The names people actually write for your brand, separated by commas. Include spelling variants, other identities, and any abbreviation your market uses. Matching is case-insensitive and whole-word. These find your brand in AI answers and classify your keywords as branded or non-branded, so they come first. Your website does not belong here.

For a fictional label called The Gilded Wardrobe, a good list is: Gilded Wardrobe, gildedwardrobe, GildedWardrobeParis, GWD

• Gilded Wardrobe, the name itself. This alone also catches "The Gilded Wardrobe", so there is no need to list the version with the article.
• gildedwardrobe, the closed-up form that appears in handles and in prose written by people who saw the web address first.
• GildedWardrobeParis, a separate identity you publish under. Worth its own entry, because the run-together spelling is not matched by either of the two above.
• GWD, the abbreviation the trade actually uses. Short forms are safe here, because matching is whole-word.

Note what is missing. The Gilded Wardrobe is redundant: the shorter entry already matches it, and listing both counts the same mention twice. Gilded and Wardrobe on their own are too generic and would pull in answers that have nothing to do with you.

Country is sent to each Assistant as a user location on the request itself, wherever the Assistant accepts one, so it searches and replies as though your customer were standing in that market. Not all of them do. Grok's API takes no location at all; Claude accepts an undisclosed list of countries and may decline yours; and a handful of markets cannot be targeted on a particular Assistant even though every other one takes them. An Assistant that is not given the market still runs and still answers, and in practice the language of the prompt carries much of the locality on its own, so the result is a weaker signal rather than a wrong one. One country per space: if you sell in several markets, run a space per market.

Industry fixes the peer group your benchmarks are measured against on Industry Insights. Once saved it cannot be changed.

If your industry is not represented in the list, choose the closest match and email support@martekio.com. A near match is worth more than a precise one at first: a benchmark needs at least two other clients in the same sector before it can be shown, so a brand new sector of one would leave you with no comparison at all. We add industries on request and move your account over afterwards, keeping all of your history.

This is what Martekio will ask the Assistants. Three ways to build the list, and you can mix them:

• Import from Search Console, the fastest honest start: your real Google queries, pre-classified against your brand identifiers.
• Upload a CSV if you already keep a keyword list. The format is documented in the guide panel.
• Type them in using the keyword editor.

Click Save when you are done.

Organization settings include Branding (your primary brand color and an optional Logo URL that replaces the logo Martekio looks up from your website, displayed in the Insights title bar, on your space card and in your Visibility Report) and the Integration kill switches. When you save a Logo URL, Martekio downloads the image once and stores it, so your logo keeps working even if the original page changes or goes away; use Save again to pick up a new version. PNG, JPEG and GIF are supported.

A dropdown of the space's working tools, displayed when you are inside a brand Space.

Visibility Report opens the Generate Visibility Report modal: a ready-made AI-assistant prompt that builds a comprehensive AI visibility report from your own data via the Assistant Connector over a 30, 60 or 90 day window, or All time, which covers every analysis from your first to your most recent (30 days by default). Where Martekio offers more than one report, the Report picker at the top chooses which one you are working on, and each keeps its own saved settings; customize colors, font, format (PDF, HTML, PPTX, or JSON), report language, actions count, time range, and an introduction, then generate the report directly with Claude. Your choices are saved to the Space automatically. The report instructions themselves are set by Martekio and are not edited or exported here. The JSON format is a content-repurposing export: drop the file into your AI content tools (ChatGPT, Claude, Gamma, and similar) to turn the report's numbers and commentary into posts, decks, and summaries.

Real-Time Prompt Check runs a single prompt live across your selected AI Assistants (see Real-Time Prompt Check). Prompt Synthesis turns keywords into prompts and lets you preview them before adding. Visibility Backlog opens your saved to-do list of recommendations, with a count of open items. Cross-Industry Citations (Beta) and Daily AI Visibility Research close the menu.

A card reads in three bands. At the top, the brand's logo, name and industry, with the status pills beside them: whether the Space is Active, Paused, in a Trial or Cancelled, and Agency Owned where the Space belongs to an agency account you were invited to rather than to your own.

Under that, four columns: Overall Share of Voice and Sentiment, each with its latest figure, a line of the last 90 days and the move since a date it names; the Analysis Schedule, with the last and next run and the number of analyses and prompts behind them; and the Brand Websites registered for the Space.

At the foot, the Space owner, the collaborators, Manage Collaboration, any tags, and the Space-level actions: duplicate, archive, and reactivate when the Space has been deactivated. A Space that has not run yet replaces all of that with a single line, Waiting for insights, and a Launch Analysis button, or Start Analysis when its schedule is off.

Opening a Space, and expanding one. These are two different things and two different targets. Clicking the card itself, or the Open Space button at its top right, opens that brand's Insights. Clicking Expand detail at the foot of the card opens the summary in place, without leaving the list.

The summary reports your most recent analysis, named once at the top as "Insights from the latest analysis:" with its date and the Space's client number in brackets, which is the reference to quote if you contact us about what you are reading. It is made of four panels: Non-Branded Share of Voice and Branded Share of Voice per AI Assistant, your Top Cited Competitors, and Sentiment by AI Assistant. The two Share of Voice panels also carry the move since the last analysis at least a month earlier, so each says which way it is going as well as where it stands.

Google AI Overview is deliberately left out of the summary panels because it renders on only about half of all queries, which would make any ranking against the other Assistants unfair; it has its own Google AI Overviews Insights page.

Everything named in the summary is a way in. Click an AI Assistant in a Share of Voice panel to open the Response Explorer filtered to that Assistant and that prompt type; click one on a Sentiment card to display all of its responses. Click a competitor to display every response that cited them, or the small external-link icon beside the name to visit that competitor's own website in a new tab. Every one of these drills opens on the same analysis the panel is reporting, not on your whole history, so the responses you read are the ones behind the number you clicked.

A key beside the panel's title names the colors actually on screen, so it never offers you one you cannot see. In the other panels the color carries nothing; the bar's length is the reading.

On an account that can hold more than one Space, the duplicate icon at the foot of a card copies that Space's configuration into a new one: its brand identifiers and websites, prompts and prompt categories, competitors and the brands you have dismissed, site labels, influential sources and the ones you removed, bot settings, tags and brand color. You are asked for a name, because two Spaces cannot share one.

The copy starts empty of history. No analyses, citations, trends or recommendations come across, because those describe runs that happened in the other Space. Its schedule is left off, so nothing runs and nothing is charged until you turn it on in Analysis Settings, and prompt regeneration is off too, so the copy keeps the prompts it was given rather than rewriting them on its first run. The card carries a Duplicated Space pill from then on.

Only a fully configured Space can be duplicated; if something required is missing, the refusal names it. A duplicate uses one of your account's Spaces exactly as creating one does.

A Space you were invited to can be duplicated too. The copy is created in your own account, which must be able to hold more than one Space and have room for it, and it uses one of your Spaces, never the inviting account's. It must also fit your account's prompt allowance, or the duplicate is refused with both numbers. The Space's tags and its Search Console connection stay behind, because they belong to the inviting account, and that account's administrators are emailed to say who made the copy. You gain no other access to the original Space.

To analyze a copy once without scheduling it, use Start Analysis on its card or in Analysis Settings. It runs a single analysis and leaves the schedule off, at most once a day, and is available to the Space's administrators.

The archive icon at the foot of a card takes that Space off your own My Martekio Spaces list and nothing more. Collection keeps running to its schedule, nothing about the Space or your billing changes, and everyone else who works on it still sees it exactly as before. It is for tidying a long list down to the brands you are actually working on.

Archiving is offered while you can display more than one Space, and the last one cannot be archived, so your list can never be emptied. To bring a Space back, open Help & Settings > Unarchive Spaces, which lists everything you have put away; unarchiving takes effect at once and the Space reappears in place. The confirmation you are given when archiving names that route, because it is the only way back.

Choose a consumer (B2C) or a business buyer (B2B) and describe them: age, gender, city and budget for a consumer; job title, company size, industry, buying role and company location for a business buyer. Only the name is required, and you can add a short line of your own about them. Martekio checks the spelling and grammar and writes the description in the language of your prompts, offering its wording beside yours.

Connect Delve.ai once per brand. In My Organization > Integration, on the brand's Space or any of its persona copies, switch Delve.ai on and enter your Delve.ai API key. Martekio checks the key with Delve.ai before saving it, stores it encrypted (AES-256-GCM), and afterwards displays only its first characters. The connection is shared by every persona copy of that brand, and only an administrator of the brand's Space can change it. Switching Delve.ai off disconnects it without deleting the key.

Choose your Delve business. The first time, pick which business in your Delve.ai account holds this brand's audience. A business whose website matches your brand's is marked, but you always choose. You can change it later.

Browse the gallery. Each card is one persona: photo, name, age, city, job and a summary. Filter by where the persona comes from (your website, your customers, social, research), search by anything in it, and open a card to read the full profile: goals, motivations, needs, pains, what makes them act, interests and values. Personas Delve.ai built from a competitor's audience are switched off by default, and a persona living outside your Space's country is marked Outside your market. Refresh fetches the latest personas from Delve.ai.

Use This Persona. Martekio writes a short first-person description of the persona and suggests a name, both of which you can edit. It leaves out every brand, product, competitor and website name, so the answers are not led towards any brand, and it leaves out the location when the persona lives outside your market, so the analysis keeps running in your Space's own country.

Confirm Persona displays the persona applied to three of the Space's real prompts. Generate Personified Prompts rewrites every prompt of the Space for that person in Validate Personified Prompts, beside the prompt it came from. Edit any of them; Rewrite writes the ones you changed again, and Write Missing fills any you cleared. Nothing is stored until Save, which asks you to confirm because it replaces the Space's prompts with the personified ones. The card then carries the persona's name, and a persona from Delve.ai is marked Persona powered by Delve.ai.

Until the Space's first analysis with the persona starts, you can change it: open the persona and choose Change Persona, which takes you back to Select a Persona Partner, so you can switch between Persona Lite and Delve.ai too. The new persona is written from the Space's original prompts, so any changes made to the prompts since the previous persona was saved, including prompts added, are not kept. Once an analysis has started, the persona is locked, so that the Space's whole history describes one audience. To analyze another audience after that, duplicate the original Space again.

Once the persona Space has run, Compare with Parent in the persona's view sets its answers beside those of the Space it was duplicated from: how often the brand is mentioned and cited, its Share of Voice, and how often the answers speak to the persona's stated need, overall, by AI assistant and by prompt, plus how far the cited sources changed.

AI answers vary a little every time, even when nothing has changed, so every figure is judged against the parent Space's normal range: the lowest and highest it scored across its analyses from 60 days before the persona's first run. Above or below that range means the persona really changes how AI assistants treat the brand; within it is normal variation. The parent run nearest in time is shown for reference. Only the prompts and assistants both Spaces ran are counted, and only analyses run after the persona was saved.

The modal opens on a Summary of four cards, Cited, Share of Voice, Mentioned and Persona reflected, each with the persona's figure, the nearest parent run and the normal range on one line. Cited and Mentioned are shares of answers: an answer counts once whether it names or links to the brand once or several times. Persona reflected is the share of answers that use the persona's own words for what it needs, and Explore on that card displays those answers in the Response Explorer. Below the Summary, Persona Analysis vs Normal Range charts each measure over time, By AI assistant and By prompt compare each with the nearest parent run (By prompt lists only the prompts that moved by more than one answer), and Cited sources shows how far the cited domains changed. Each section ends with a What this means panel, and the list on the left jumps between them.

The headline section, titled with your own brand name once one is set. One row per AI Assistant ranks the Assistants your Space collects (Claude, ChatGPT, Gemini, Perplexity, Google AI Mode, and Grok or Microsoft Copilot where they have been added) by visibility rate, with the leading Assistant marked in the green KPI color. Click a row to expand it and reveal the four combinations of classification and measure for that Assistant, plus the number of prompts processed, with links to the underlying responses. Google AI Overview is not ranked here: it renders for only around half of all queries, so it has its own tile at the end of the row.

Two pill groups above the rows decide what "visible" means, and the heading restates both so a collapsed section or a screenshot still says what it measured. The first group is the prompt classification, Non-Branded or Branded. The second is the measure: Citations counts the responses that cite your website as a source, and Mentions counts the responses that name your brand whether or not they link to it. Citations is the default and the harder number. The two can be far apart, and a brand the Assistants talk about constantly without ever citing is a real and common finding, so it is worth switching between them rather than reading only one. A classification your Space does not track is grayed out rather than hidden, so you can tell the measure exists.

Of all brand citations made across you and your competitors, what percentage were yours. Split into Non-Branded Share of Voice (your share across non-branded prompts) and Branded Share of Voice (your share across branded prompts).

On the Visibility Overview the visibility rows and the Share of Voice rows sit in one section under one heading, and they share a single Non-Branded / Branded choice: the pills above the visibility rows move both halves at once, so the card can never read "Branded Visibility" directly above "Non-Branded Share of Voice". The choice is remembered for your next visit.

Share of Voice has no Citations / Mentions choice of its own, deliberately: a share needs a denominator covering every brand, and competitor mentions are not something the data can supply, so a mentions-based Share of Voice would be a number we cannot honestly compute. Share of Voice is always built from citations.

The visualization beneath the cards follows them: switch the cards to Branded and the visualization displays Branded only, switch back and it displays Non-Branded only. The visualization's own Non-Branded and Branded pills still work, so you can put the second one back beside the first to compare them, and doing so never moves the cards. Non-Branded is the stronger competitive signal, because it reveals whether AI Assistants favor you when your brand is not named in the prompt.

If you have filed your prompts under Prompt Categories, the visibility and Share of Voice sections each gain a collapsible by category breakdown. The heading names the measure it is drawn on, so it reads "Citation Visibility & Share of Voice by Category" or "Mention Visibility & Share of Voice by Category" depending on the Citations / Mentions pills above.

Expand it to compare how you perform across your categories, say your Competitive prompts versus your User Need prompts, ranked as bars. It follows the same Branded / Non-Branded and Citations / Mentions choices as the section above it, so the whole page answers one question at a time. Share of Voice by Category is the exception and is always citation-based, for the reason given under Share of Voice.

An AI Assistant dropdown scopes the breakdown to a single Assistant or all of them (your choice is remembered). Click any category bar to open the Response Explorer filtered to that category's prompts and the selected Assistant.

A What this means panel above each breakdown sums it up: your strongest and weakest category, where competitors hold most of the conversation or your visibility is lowest, any category where you earn nothing, figures resting on too few responses, and in comparison mode the category that moved most.

The breakdown is computed live from your responses, so it always reflects how your prompts are filed today, across your whole history. It stays collapsed until you open it, and displays a short prompt when you have no categories, or none with prompts assigned yet.

Identifies the competitive brands most frequently cited by each AI Assistant, in the same per-Assistant grid as the headline. It displays your curated brands by default, both your Competitive Threats and your Tracked competitors, with a Threats/Tracked / All Brands toggle in the section header to switch to every brand cited whether you curated it or not.

A second toggle chooses what the bars measure: Share of Voice, the default, gives each brand's share of every citation in the analysis and adds your own brand as the first bar so the percentages have something to be read against; Citations gives the raw count instead. The Share of Voice bars deliberately do not add up to 100%, because the share is of every citation in the analysis and the brands you curated are usually a small part of that; the gap is everything not on screen, from untracked brands to publishers and sources like Reddit.

Bars are colored by what a brand is: green for your own brand, red for a Competitive Threat, slate for a Tracked competitor, gray for one you have not curated, with a key under the visualizations.

When a comparison analysis is selected, every brand keeps one row with two bars: the analysis you picked on top and the comparison beneath it in a lighter shade of the same color, drawn to one shared scale so the two are directly comparable, with a legend above the grid naming which is which. A brand new to this analysis is tagged NEW and has only the upper bar; a brand the earlier analysis cited and this one did not is grouped at the foot of the card under Not cited on that date, with only its lower bar.

The rank numbers count competitors, so 1 is your strongest competitor, and your own brand sits above them without a number because it is not competing with itself. A Competitive Movement Summary highlights brands gaining, losing, newly emerging, or no longer appearing, and a takeaway names the single most cited competitor.

Google AI Overview gets its own tile at the end of the headline row rather than a place in the ranking, because Google renders it for only around half of all queries. The tile follows the row's Branded / Non-Branded toggle, exactly like the Assistant cards beside it: it reports the percentage of rendered AI Overviews in the selected facet that cited your site as a source. The dedicated Google AI Overviews insights cover the surface in full.

ChatGPT recommends products alongside its answers to purchase intent questions, as a carousel of product cards inside the answer. This section lists every prompt in the selected analysis that drew one, every product and price on it, and who is selling them. Only ChatGPT returns a carousel, and only for buying questions, so prompts that ask for advice rather than where to buy will not draw one.

The card is in two halves. Sellers, on the left, is one row per retailer, with the share of your analyzed prompts it appeared on and a marker where it is selling products of yours. The retailer is rarely the brand that makes the product: goods reach the carousel through resellers, so the shop named on a card is usually somebody other than the manufacturer. Each seller is matched against your competitor lists as they stand today, so one on neither list is flagged for you to decide about, and can be added to your Threats or Tracked list from the row. Where ChatGPT names the seller but gives no website, Martekio resolves the retailer to a domain from the sites your own analyses have cited; where it names no seller at all, the site the product is listed on is used instead. If neither is available the row reads Seller not stated and the curate buttons are grayed, and you can add the competitor by domain in the Competition Explorer instead.

Prompts, on the right, lists each prompt that drew a carousel and the full shelf ChatGPT presented, in the order it presented it, with the seller under each product and the price in a column so you can read down it. Your own products are marked, in green where you are the seller and amber where somebody else is. Clicking a seller filters the prompts to the ones that retailer appeared on, and the shelf stays complete with the other products dimmed, so you can still tell what it is standing beside and at what price. Clicking a product name opens the page ChatGPT linked it to, in a new tab. Not every card carries a link, and a few point at the seller's home page rather than at the product, which is why the link is named after the seller.

Three controls narrow it. The seller pills above the left column choose which retailers are listed and count sellers. The Branded and Non-Branded pills above the right column choose which prompts are displayed and count prompts. The products pulldown asks how your own goods reach the shelf: All products, Sold by your own brand, Sold by another retailer, or No products of yours, the prompts that drew a shelf and carried nothing of yours at all, which is often the largest of the four and the one worth reading first. Each option but the resting one carries the number of prompts behind it, so an option reading 0 tells you the answer without clicking.

It selects prompts, not products: a prompt qualifies when any one product on its shelf matches, and the shelf is then displayed in full with the products that did not match dimmed, so you can still tell what yours is standing beside and at what price. On No products of yours nothing is dimmed, because there the whole shelf is the finding: it names who took a slot you never reached.

A retailer with nothing left under the filters on the right stays in the list rather than disappearing, but moves below a divider reading No products presented under these filters, so on a long list the retailers that still have prompts are always the ones at the top. Within each group the order never changes as you filter, so a retailer only ever moves from one side of the divider to the other. A download button on the Prompts heading exports what is on screen as a CSV, one row per product, so you can pivot it by seller, prompt or classification in a spreadsheet.

The card contributes a line to the What this means panel at the top of the page, saying how many of the prompts that drew a carousel presented a product of yours and whether you or a reseller held the slot, with the carousels that carried nothing of yours and the retailers still waiting to be classified called out beneath it. It describes the analysis rather than whatever you have filtered the card down to, so exploring the card never rewrites the summary. It deliberately makes no claim about a retailer undercutting you: a carousel card carries a title and a price but no product identity, so the only honest comparison is between two listings of the same product, which the shelf already displays side by side.

Two things to know before drawing conclusions from it. Shopping results are local: prices and availability reflect a location inside your Space's country that ChatGPT chooses, so the shelf here will not match what you personally get in ChatGPT unless you are in the same country. And the carousel is volatile: the same question asked twice hours apart can return different products at different prices, so read a single analysis as a sample rather than a fixed ranking.

ChatGPT may present sponsored ads alongside its answers. This section lists every prompt in the selected analysis that drew one, grouped by prompt, with the advertiser behind each ad and whether it was Shown to the user or only Served only, meaning returned in the response but never displayed. Only ChatGPT carries ads today, and they are sparse, so a Space can legitimately have none.

Filter pills narrow the list two ways. All prompts with ads, Non-Branded and Branded split it by the prompt's classification and are mutually exclusive, so picking one clears the others; All is the state the card opens in. Advertiser is a pulldown offering All advertisers, Own brand, Threats, Tracked and Uncurated, matched against your competitor lists as they stand today rather than as they stood when the ad was captured. It chooses which prompts appear, never which ads inside them: a prompt always lists its complete set of advertisers, with the ones that did not match dimmed. That matters because most prompts draw more than one ad, so trimming them would display a single advertiser where three had bid, and hide the most useful thing on the card: that a rival was displayed on a prompt where your own ad was served and never shown. An option with nothing behind it under your current filters is grayed rather than hidden, so you can always tell the difference between a choice that does not apply and one that is simply not there.

Your ad placement is a second pulldown, offering Shown and Not shown, the prompts where your ad reached the user and the prompts where it was returned but never displayed. Pair either with Branded or Non-Branded to split it by prompt type.

Every count on this card is a number of prompts, never of ads: a card reading 68 is 68 prompts that drew an ad between them, which on one real Space was 135 ads. The counts on the Branded pills are numbers of prompts, so a pill always delivers exactly what it says, and they respond to the other two controls: pick Threats and Branded stops reporting every branded prompt and reports how many a Competitive Threat actually bid on. A control never counts its own selection, so the three options always stay comparable with each other. The two pulldowns carry no count, to keep every control on one row.

Above the prompts sits an Advertisers panel. It opens with the share of prompts each kind of advertiser turned up on, Your brand, Threats, Tracked and Uncurated, which is the quickest read of how contested your prompt set is. The four will not add up to 100%, and should not: one prompt often carries a threat and an uncurated brand at once, so each figure answers "how often does this turn up" rather than splitting a whole. The denominator is every prompt the analysis ran, not just the ones that drew an ad, and the line states it: a Space whose 90 prompts produced ads on 68 of them reads its own brand at 33%, not 44%. It narrows with Branded and Non-Branded, which split the prompt set, but not with the advertiser or placement controls, which describe ads.

Below it is one row per brand with its status, how many ads it ran, across how many prompts, and what share of the prompts in scope that is, on the same denominator as the summary above so the two read together. That is where curation happens: a brand on neither of your lists carries Threat and Track buttons. Each is grayed once that list is full, with a hover saying so, since one cap (25 by default) governs both lists, and picking one adds it to that list immediately and re-reads the section, so the brand, its ads and the pill counts all move together. A brand bidding against your prompts that your Space has never classified is invisible to every competitor visualization in the product until you classify it, which is why this list is the first thing in the card.

Clicking an advertiser opens the Response Explorer filtered to the responses that carried an ad from that brand, within the analysis you were reading. It arrives as a named, removable filter like any other, so you can clear it or add to it.

Both drill-throughs sit on the prompt row. Clicking the prompt opens it in the Response Explorer, filtered to that prompt and to the analysis you were reading, so you land on the same responses rather than that page's own default date window. The megaphone beside it opens the same Sponsored Ads in This Response detail you get from the Response Explorer, with each ad's title, text, image, landing site and position, plus any watched competitors cited in the same response. The number next to them is how many advertisers appeared on that prompt.

For every prompt in an analysis, Martekio also captures Google's first page of organic results. This section takes each page an Assistant cited and checks whether it was on that first page, then reports, per Assistant, what share of its citations also rank on Google. It is always a citation measure, which the heading says, and it keeps its own Non-Branded / Branded pills rather than following the visibility section's: comparing Assistants on non-branded prompts while reading Share of Voice branded is a legitimate way to work, and the two axes diverge hard here. Each row is clickable through to the Response Explorer, and Google AI Overview and Google AI Mode appear as controls at the top: they draw on Google's own index, so they are the reader's calibration point rather than competitors in the ranking.

Read it as a signal of how much each Assistant leans on Google, not as a report card on your own SEO. It counts every citation an Assistant made, competitors and third-party sources included, not only pages on your site. A low figure means that Assistant is drawing on sources Google does not rank on page one for that prompt; a high figure means the two largely agree.

The section is marked Experimental. Assistants with very few citations in the selected analysis are flagged as a low sample, and the metric always describes the selected analysis rather than an average over time, because it drifts between runs.

Four sections (Visibility Rate, Share of Voice, Sentiment and Owned Domain Citations) visualize each AI Assistant over time, and they work the same way in all four. Each section opens as a compact table (latest value, change, a small trend line on the same scale, and the low and high over the range); the toggle at the top right switches to Visualization, one Assistant's visualization at a time, or Combined, every Assistant as a line on one visualization with a legend you can click to hide or display an Assistant. The choice applies to every section and is remembered, and Reset filters returns it to Table.

In the Visualization view a row of Assistant logos sits above the visualization: click a logo to display that Assistant, or use the arrows either side to step through them. The choice is shared, so picking ChatGPT in one section turns every section to ChatGPT and the page reads as one story about one Assistant. Every visualization in a section shares one scale, so a taller line really is a higher value whichever Assistant is displayed.

Each visualization leads with a headline: the latest value, and beneath it the change since the start of the date range, green when it rose and red when it fell; the Assistant ahead on the latest value has its headline in green. Hover over a date on any visualization and every visualization on the page marks that date with its own value there, so one hover reads the same analysis across Visibility, Share of Voice, Sentiment and Owned Domain at once.

Every line breaks wherever collection stopped for a stretch rather than joining up across it, and rests on a visible dot at each reading, so a period in which nothing ran reads as a gap instead of as a trend.

The opening section. A grid comparing branded and non-branded citation and mention rates for each Assistant: the share of that Assistant's answers in which your brand appears. These are the same four metrics the How visible is your brand? section of Visibility Overview ranks, so the point-in-time view and this trend agree. Rates rather than raw counts, because a count climbs whenever more prompts are analyzed even as your share of answers falls.

A two-row filter bar drives every card together: row one picks the branding axis, either Non-Branded or Branded, and row two the type, either Citations or Mentions, so the section always displays one measure at a time. Click any point to read the answers behind it in the Response Explorer, scoped to that analysis and that Assistant.

A grid of per-Assistant Share of Voice trend visualizations, one card per AI Assistant (set by your Display Assistants preference). A shared pill bar flips every card between Non-Branded and Branded in lockstep. Non-Branded is the default because it best reflects unprompted recommendations.

Tracks the tone of mentions over time from -1 to +1, as per-Assistant visualizations, with the most recent score as each card's headline. Reveals whether strategy changes or external events are shifting how Assistants characterize your brand.

Visualizes the trend of direct citations to your domain over time per Assistant. Pills flip every card between Non-Branded and Branded citation share. A rising non-branded share signals that Assistants trust your content as an authority for your industry, not just for your own brand name. Click any point to read the answers behind it in the Response Explorer, scoped to that analysis, that Assistant and the classification displayed.

Which individual prompts are gaining or losing AI visibility. For each prompt Martekio measures the citation rate, the share of AI answers that cite your brand, across the date range you have selected, and compares it with the same length of time immediately before it: pick a 14 day range and it compares those 14 days with the 14 before them.

Very short and very long ranges are clamped to between 7 and 90 days, because below a week there are too few answers per prompt to mean anything, and beyond a quarter the comparison window reaches back further than most Spaces have data.

A prompt is Improving when it is cited in a larger share of answers now and Declining when its share falls, and the change is stated in percentage points (cited in 60% of answers, up from 40%, is +20 points). Click a prompt to open it in the Response Explorer, or click an Assistant icon to visualize that Assistant's trend for it.

Measures how consistently your brand stays cited across the trailing window of analyses. Every tracked prompt scores from 0% (never cited) to 100% (cited every time) and lands in one of three bands: Consistent (at or above 80%, green), Inconsistent (20 to 80%, amber), or Underperformer (below 20%, red). A prompt has to appear in at least five analyses before it is scored at all, so a prompt you added last week is left out rather than judged on two data points.

Consistency Trend draws two lines at each analysis, pooled across all Assistants: the share of your scored prompts that are Consistent (green) and the share that are Underperformers (red), so the green line rising and the red falling is your brand becoming more reliably cited. Hover a point for the counts behind it, including the Inconsistent prompts in between. The line breaks where no analysis ran. Click any point on the trend to open the prompt table as it stood at that analysis, filtered to that line's band (Consistent or Underperformer) across all Assistants; click any prompt in it to open it in the Response Explorer.

Below it, Consistency by AI Assistant breaks the three bands down into one bar per Assistant. Click any band segment to open the prompt table filtered to that Assistant and that band, and the band is named in a chip beside the prompt count in the modal title; click the chip to clear it and display every band. Prompt detail opens the same table unfiltered. Download Prompts exports whatever the table currently displays, with every filter and the current sort applied, as a CSV.

Controls: pills flip between Non-Branded and Branded and between Citations and Mentions, and the page's date range applies: it selects which points the trend draws and moves the per-Assistant bars to the end of the range. Scores still reach back before the range start so every point keeps its full eight-analysis window. Every prompt is scored over the trailing 8 analyses, roughly 4 weeks at the standard twice-weekly cadence. Non-Branded consistency is the stronger signal that your brand is competitive for the prompts tested.

The Curate Competitors button at the top right of the Competition Explorer opens the modal where all competitor curation happens: brands are detected; competitors are chosen. Curation is all-time, independent of the date range on the page. To help you judge each brand at a glance, rows show the title of its most-cited page as a subtitle ("Draftboard vs Homie, Comparison & Alternative" reads as a rival; a bare domain name reads as a directory). Clicking a brand name opens its citing responses in the Response Explorer (all time), so you can read what the Assistants actually say before deciding; the external-link icon beside the name opens the brand's website in a new tab, and the page-title subtitle links to that specific page.

The tabs, left to right: All Brands, a divider, then the two chosen sets, Threats and Tracked (each shown against the same cap, e.g. n/25), another divider, and finally the source piles Shopping Sellers (only on a Space that collects ChatGPT Shopping), New Brands and Removed. Threats sits before Tracked because it is the set that drives these insights. Every tab offers instant search (which searches every brand regardless of the tab you are on, including removed ones), sortable columns (Competitor, Citations all time, Recent last 30 days, First seen), its own short help note, and a CSV download that matches exactly what you are looking at (the button shows the count, and respects the tab, the search box, and any tier or band you have selected).

All Brands: every brand ever cited alongside yours. Each row carries one Track / Threat control: press Track to add it to your monitored set, or Threat to flag it as a Competitive Threat straight from here (no need to track it first). A brand holds one status at a time, so choosing one clears the other. A Remove action removes noise (a video platform or a directory is a detection, not a rival), and is available on every tab except Removed itself.

The citation distribution bar: the fastest way to work through thousands of detected brands. It sorts every brand by how many citations it holds and splits them into four candidate groups: Primary (the brands behind the first half of all citations), Secondary, Tertiary, and the Tail. Each group is broken into citation bands, and each shows how many of its brands you still have to look at, turning to a tick when you have handled them all. Click a group or a band to filter the list below to just those brands. Start at Primary: a handful of brands there usually accounts for more of your category's visibility than the entire tail. You do not have to process everything, leaving most brands uncurated is perfectly normal.

Shopping Sellers: the retailers in your latest ChatGPT Shopping carousel that you do not follow yet. They compete for the same customers at the moment of purchase, often without ever being cited in an AI answer, which is why they can be missing from your other tabs' rankings. A Shopping column shows the share of ChatGPT prompts whose carousel listed each one, and the tab opens sorted by it. Follow a seller with Track or Threat as on any other tab. A seller you follow stays on screen with its button lit, so you can see the change took and undo it; the tab count drops straight away, and the next time you open the tab it lists only the sellers still unfollowed. Only sellers with a website your Space has seen cited are listed, because your lists match competitors by domain.

New Brands: brands cited for the first time ever in the last 30 days with 10 or more citations, your potential new competitive threats. Triage each one with Track, Threat, or Remove.

Removed: everything you removed as noise. Untick a row to restore it at any time; nothing is ever deleted.

Tracked: your monitored radar, the set the Tracked Brands view and the API reflect. Track the brands worth watching but secondary to your real competition: publications, marketplaces, retailers carrying your products, and emerging names. Competitive Threats are for the brands you go head to head with. Switch Track off to remove a brand, or press Threat to promote it to Competitive Threats (which moves it off Tracked, since a brand holds one status at a time). Removing a tracked brand asks you to confirm first, because that clears its Tracked status; restoring it later brings it back uncurated.

Threats: your real rivals (25 by default), driving the Competitive Threats visualizations and Citation Gaps. The Remove action takes a brand off your threats; it does not become tracked automatically.

The basket: on the Threats and Tracked tabs, a basket icon after a brand's name means it sells in your latest ChatGPT Shopping carousel; hover it for how many prompts and on which date. It describes the brand, not how you added it, so it also marks threats you flagged before, and it disappears if the brand stops appearing in the carousel.

Two buttons at the bottom of Curate Competitors for setting up quickly. Auto Select ranks the brands AI Assistants cite alongside you by all-time citation count, the same figure the Citations column displays, then has an AI classifier judge each of the top brands: a genuine rival, or simply a retailer that stocks you, a marketplace, a price comparison site or a publisher. Your own brand on its other domains, and anything you have already removed, are excluded before the classifier is consulted at all.

It then displays what it picked, in two tabs, Threats and Tracked. Each brand carries its citation count, the title of its most-cited page and a link to open its website, so an unfamiliar domain can be judged without leaving the modal. Anything the check could not place is marked Validate and kept, so those are the rows worth looking at first. A third Skipped tab lists everything the check rejected and why. If you disagree with one, Threat and Track beside it move that brand straight into the list you want, so the check is a suggestion rather than the last word. Those buttons gray out when the list is already full. Reject anything that is not a real competitor with the Reject button on its row.

Save selections appears only once you have opened both the Threats and Tracked tabs, so the review cannot be skipped. Nothing is written until you save, so Cancel leaves your existing lists exactly as they were. Anything the check is unsure about is kept rather than discarded, which is why the review step matters. Brands you have already removed are never selected, and neither is your own brand on its other domains.

Reset clears both lists and leaves your removed brands alone. Both ask you to confirm before they run. Reset writes immediately and cannot be undone; Auto Select only replaces your existing lists at the moment you press Save selections, so Cancel at any earlier point costs you nothing.

Your curated brands in one section, with four buttons in the section header to switch between Competitive Threats, Tracked Brands, Other, every brand Assistants cited that sits in neither list, and Combined, all of them on one visualization with each bar colored by what the brand is (red for a threat, slate for tracked, gray for uncurated, green for you) and a key above the visualization. The visualization, the detail table and the Citation Gaps section below all follow that choice, so everything on screen describes the same set; a brand holds one status at a time, so it is represented in exactly one of the three views.

Other and Combined draw the ranking visualization alone: there is no single curated list behind them, so they carry no detail table and no citation gaps. Other is the place to spot a brand worth promoting to Threats or Tracked with Curate Competitors; Combined is where you check how your curated brands rank against the whole field.

Taking Competitive Threats as the example: a ranking visualization on top and a detail table below. The visualization ranks the Competitive Threats cited in the selected window across every Assistant, in Non-Branded (who Assistants suggest organically for your category, the strongest competitive signal) and Branded (who else gets cited when users ask about you by name) leaderboards, with its own CSV export scoped to your threats.

The detail table below lists every flagged threat, including any not cited in the selected window, so the two counts differ by design. It shows Last Seen (the most recent analysis date the threat was detected on), Frequency (the total times it was detected across all Assistants), Citations across every Assistant in the window, and a Momentum pill. Every column heading sorts. Click a threat's name to open its citing responses in the Response Explorer, click the external-link icon beside the name to visit its website in a new tab, or click the magnifier to open the Competitor Deep-Dive modal described below. A placeholder invites you to flag competitive threats via Curate Competitors when none are set.

The Rank by control in the section header decides what the ranking visualization measures. Citations counts links, so a brand cited through three of its pages in one answer scores three. Responses counts the AI answers that cited the brand, once each. Share of responses expresses those answers as a percentage of every answer in the period, which holds steady when you widen the date range.

ChatGPT Shopping ranks brands by the share of ChatGPT prompts whose product carousel listed them. It is the one measure that is not about citations, so it reaches the retailers that sell to your customers without ever being cited. It covers ChatGPT only, the one assistant with a shopping carousel, and the visualization says so. Your own bar there is your own website's share of the carousels, products you sell directly; a retailer reselling your products is a competitor bar. The bars do not open the Response Explorer on this measure.

On every measure, a basket after a brand's name means it sells in your latest ChatGPT Shopping carousel. Hover it for the share and the date.

Momentum is a sortable column in the Competitive Threats detail table, not a section of its own. Each threat carries a pill comparing its citations in the selected window against the previous window of the same length: a signed percentage, or New when the brand was not cited at all in the prior window. Hover the pill for the two raw counts behind it. A dash means there is no comparable prior window, which is what an All Time range produces.

Momentum colors read from a threat perspective: new entrants are blue, rivals gaining ground are red or amber, stable is gray, and declining or disappearing rivals are green, because a fading rival is good news for you.

A battleground view: for every prompt you track, Citation Gaps checks whether any of your Competitive Threats was cited while your brand was not cited in any response to that prompt. These are the battlegrounds your rivals own outright, ranked by how many citations they collected there; each row shows the prompt with the competitors cited. Click a prompt to open the Response Explorer on the AI answers behind it. Closing a gap usually means creating or improving content that answers that exact prompt.

The section opens with a ranked bar visualization of the competitors cited on the most gap prompts, counted over every gap in the selected window rather than only the rows listed. Click a bar to narrow the list to that competitor's gap prompts; a filter bar above the list names the competitor, with a Clear button to return to every gap. Each row leads with how many of the answers to that prompt in the window cite a competitor, for example 93 of 131 answers cite competitors, rather than a bare citation count. The competitors cited appear as logo chips in their status color, with any beyond the first four folded into +N more. Only Branded prompts carry a badge; a row with none is Non-Branded.

If you have filed your prompts under Prompt Categories, a Prompt Category dropdown narrows the list, so you can ask where your rivals are beating you specifically on, say, your Competitive prompts. It offers All Categories and an Uncategorized option, and appears only once some of your prompts are filed. Each row also carries its category as a badge. Note that a category can legitimately return no gaps: prompts naming your own brand are usually ones you already win, so an empty result there is good news rather than a missing answer.

Answers one question: are more brands entering your field than before? A brand enters your field once an AI Assistant has cited it in a second response, counting only prompts each Assistant had already answered in at least three earlier analyses before the brand first appeared. One-off mentions are left out, and so are brands that are only new because you added a prompt or an Assistant. Counting starts from a Space's fifth analysis; until then the section says it is waiting for more insights.

A single summary line gives the number of brands that entered in the selected period and the pace per analysis, so a Space that runs more often is not read as more crowded, and says whether that pace has risen, fallen, or held against the period of equal length before. Below it, a table ranks the ten brands cited by the most Assistants, then in the most analyses, with Display all opening the full list; its Assistants, Analyses, and Responses columns count the selected period, the same responses a brand's name opens in the Response Explorer.

Both lists carry inline Track, Threat, and Remove actions (hidden for invited collaborators, who can still read them); clicking a brand name opens its citing responses in the Response Explorer, and the external-link icon opens its website. The prompt-type pill row (All, Branded, Non-Branded) switches the whole section to the brands that entered through that type of prompt, and is remembered per client. Beside it, Exclude media & blogs etc. hides the domains that are sources rather than brands: see Sources, and the Exclude media & blogs etc. filter below. This section is always visible, even before you have tracked any competitors.

Not every domain an AI Assistant cites is a brand competing with you. Many are sources the Assistant is quoting: news sites and trade press, magazines, blogs and review sites, regulators and government bodies, trade associations, conferences, directories and price-comparison listings, encyclopedias, social and video platforms, forums and app stores. Martekio works out which is which for every cited domain, and the Exclude media & blogs etc. pill on Newly Cited Brands hides the sources so the list reads as the brands entering your market. Leave it off and each source is labeled with a Source chip instead, because a source being cited for your topics is often worth reading even though it is not a rival.

Your own curation always wins. A brand you have Tracked or flagged as a Competitive Threat is never treated as a source, whatever Martekio would otherwise have concluded, because you have told us it belongs in your field. Where there is not enough information to decide, the domain is shown as it is today rather than hidden: nothing disappears because Martekio was unsure.

Opened from the magnifier at the end of a Competitive Threats table row, this modal is a per-competitor workbench organized in two tabs. Overview displays KPI cards (total citations, distinct URLs, their citation share versus you, and the count of prompts where you never appear, shown in red when above zero), the Competitive Overlap benchmark comparing their citations against yours on the same prompts with a per-prompt table (each row jumps to the Response Explorer for that exact prompt), and the Most Cited URLs ranking with an average-citations-per-URL concentration signal (each URL links to the live page, marked with an external-link icon).

Citations lists every individual citation with status filters and sorting. An All / Non-Branded / Branded filter stays pinned while you scroll and re-slices everything in the modal, the amber "prompts where you never appear" chip jumps straight to the Citation Gaps section, and a View Responses button opens the Response Explorer scoped to this competitor, the current filter, and the selected window.

The per-competitor drill-down opens with a Competitive Overlap section comparing that competitor's citations against your own across the same analyzed prompts. A benchmark bar displays the two totals side by side, and a table lists the prompts where the competitor is cited most, each with a three-state Your Presence indicator: cited in the same responses as them, cited on the prompt but never alongside them, or not cited at all. A chip counts the prompts where the competitor is cited and you never appear. Filter by Prompt pills (All, Branded, Non-Branded) scope the whole drill-down by prompt type.

The rank pools each brand's Non-Branded answers across its Assistants, leaving out Google AI Overview as every cross-Assistant comparison does, and is displayed once at least two other brands were measured. Strongest and weakest compare your Non-Branded citation rate with the industry average on each Assistant. The gap over the date range comes from Citation Rate Over Time: Improving when, on average across the Assistants, the gap has moved at least one point in your favor between the first and last month, Slipping when it has moved against you, Steady otherwise.

A row of comparison cards (one per metric) benchmarking your brand against your sector over the selected date range on eight metrics: branded citations, non-branded citations, branded mentions, non-branded mentions, average citations per response, share of voice (your citations as a share of every citation in your answers), sentiment (how positively answers that mention you describe you, from -1 to +1) and citation position (where your site first appears among the sources cited, where 1 is first and lower is better). Each card displays your value, a pair of bars (you in blue, your industry in grey, the same key as Performance by AI Assistant), and the gap to your industry's average in points, or in citations per response on the last card. A gap under one point (0.1 citations per response) reads as level rather than as a win or a loss, and a metric not yet measured for both you and your industry reads No data rather than a tie at zero.

A metric only counts an Assistant when both you and the sector have a figure, so the averages always share the same denominator. The sector average is the mean of the other brands Martekio tracks in your industry: your own results are never part of it, each brand counts once no matter how many keywords it tracks, and Martekio's own demonstration spaces are excluded. Sectors with fewer than two other tracked brands display no comparison at all rather than an unreliable one.

Paired horizontal bars breaking down your branded and non-branded citation rates against the sector average for each Assistant individually, with the gap in points on each row. Where your bar is longer than the average you are outperforming peers; where it is shorter, that gap is your growth opportunity. The comparison is brand-versus-peers within the same Assistant, so the position stays honest. The section header states how many other brands the average covers.

The citation rate is the share of your Non-Branded AI answers that cite your own site. Each month pools every analysis you ran that month, so it is the share of that month's answers. The sector line averages the same brands the bars above compare you with, each brand counting equally, and is left out for any month fewer than two other brands were measured. A month with no analysis is a gap in the line, never a zero. A dashed grey line marks the sector's average across every AI assistant except Google AI Overview, month by month, a reference for judging each assistant against assistants in general. With one assistant selected, a shaded band runs from the weakest to the strongest brand in the sector that month, shown once at least three other brands were measured. A table above the chart gives, per assistant, your latest rate and the sector's, the gap in points, how that gap has moved since the start of the range, and your rank among the brands measured, and a What this means panel under it sums up where you lead and trail.

A grid with one tile per displayed Assistant, arranged in even rows. Each tile displays the top 10 domains for the selected date range, ranked by the share of a brand’s answers that cite them, averaged so every brand in your sector counts equally whatever the size of its prompt set. Google’s own search links and ChatGPT’s product-search links are left out, and a site’s subdomains count as one. Rising and Falling split the date range at the midpoint of each Assistant’s own analyses, so an Assistant that started collecting later is still compared over its own history, and compare each domain’s share in the two halves, counting only brands measured in both, so a brand that starts partway through cannot make its own sources look like they are surging. A domain needs at least three citing answers to rank, and the views are offered once each half holds at least two analysis dates. Bars follow the same colours as every chart that ranks brands, so a red bar is one of your Competitive Threats. Domains cited by every Assistant get a star.

A spreadsheet-style table displaying every AI response that mentions your brand, with around twenty filterable and sortable columns covering the Assistant, the origin keyword, the prompt sent, the full reply, sentiment and its explanation, share of voice, and the branded, competitive, and citation-URL counts. Filtering is instant as you type.

Smart loading: by default the grid fetches only your displayed Assistants for the trailing 30 days, widening automatically to 60, then 90 days, then full history if that window is empty. A Load All Data action pulls full history on demand.

Shared with Prompt Check: the same grid, cell renderers, and drill-downs power the Prompt Check results view, where a single typed prompt is run live across your selected Assistants.

Cited page filter: arriving from a visualization that counts citations per page, such as Owned Domain Citations on Google AI Overviews, filters the grid to exactly the responses whose citations include that page. It appears as a removable Cited page step in the filter funnel, so you can see the constraint and clear it. There is no way to set it from the filter bar; it exists so a bar reading 8 opens the 8 responses behind it rather than one example.

Prompt Category filter: if you have filed your prompts under categories in Manage Keywords & Prompts, the filter bar gains a Prompt Category dropdown so you can read only the responses to, say, your Competitive prompts. It offers an Uncategorized option too, which is a quick way to find prompts still waiting to be filed. The dropdown appears only once you have categories with prompts assigned, and lists only those present in the responses currently loaded. Categories are resolved live, so a prompt you re-file is reflected the next time the page loads, across your whole history rather than only new runs.

The first column of every row carries a Display Citations icon. It opens a modal listing the sources the Assistant referenced, distinguishing branded citations to your own domain, competitive citations, and other influential sources, so you can see exactly which pages the Assistant drew on.

A collapsible Share of Citations by Brand panel at the top breaks that response's citations down by the brand each one points to, with your own share marked You and tracked rivals marked Competitor. Because it counts every citation in the response, your share there matches the Share of Voice column in the grid exactly; click any brand to filter the list below to its citations.

Where a source provides one, each citation shows the page's title above its URL, and a Google Maps place (a business or location an Assistant cited) is labeled with a Place badge and its name.

Some Assistants run their own background searches before writing an answer, and where they report them, Martekio records them. The Fan Outs column gives each response a count of those hidden searches, and the icon beside it opens the full list of queries. Four Assistants expose them today: ChatGPT, Claude, Grok and Microsoft Copilot.

For every other Assistant the cell shows a dash rather than a zero, because nothing was reported, which is a different thing from a search that found nothing; a genuine zero from an Assistant that does report them shows as 0. A Fan Outs filter above the grid narrows the view to the Assistants that report them, so you can read only the responses where this evidence exists.

The ChatGPT Shopping column answers Yes or No for whether that response carried a product carousel. Only ChatGPT returns one, and only for purchase intent questions, so most rows read No. Click the icon on a Yes to open the shelf exactly as ChatGPT presented it: each product with its seller underneath, the price in a column so you can read down it, and your own products marked in green where you are the seller and amber where somebody else is. Product names that ChatGPT linked open in a new tab. A ChatGPT Shopping filter above the grid narrows the view to the responses that carried one.

ChatGPT may present sponsored ads alongside some of its answers, and Martekio records them. The ChatGPT Sponsored Ads column answers Yes or No for each response; only ChatGPT carries ads today, and they are sparse, so most rows read No. Hover a Yes to display how many ads the response carried and how many of them were shown to the user.

Click the icon on a Yes to open a modal listing each one: the advertiser, the ad's title and text, its image when one is present, its landing site (which opens in a new tab), whether it was shown to the user or only served in the response, its position, and a tag when the advertiser is your own brand or a watched competitor. The modal also lists any watched competitors also cited in that response; clicking an advertiser or a competitor refilters the grid to every response that cites that site.

Served and shown are kept separate on purpose: an ad can be returned in the response without being displayed. A Has Sponsored Ads filter above the grid narrows the view to just the responses that carried an ad.

A funnel above the grid displays how your filters narrow the responses, one step per filter in the order you applied them. The first step is All responses in the current time window or selected analysis, and it is always shown, even with no filters on. Each later step gives its count and its share of the step before it, so you can see which filter is doing the narrowing; the last step also gives, in brackets, its share of all responses.

A step that removes nothing is dimmed, and the bars lighten as the funnel narrows. Filters set from a column heading appear as steps too. The × on a step removes that filter, and Clear all in the funnel's corner removes them all while leaving the Time filter as it is. The last step's count is the number of responses in the grid.

A heat matrix of your influential sources: one row per source ranked by total citations, one column per displayed Assistant. Each cell shows the citation count with the number of unique URLs in brackets, shaded in the Assistant's color: the deeper the shade, the higher that source ranks for that Assistant, with each column's leader in bold. A dash means that Assistant never cited the source in the period.

A Top Assistant column closes each row with the logo of the Assistant where that source takes the largest share of the Assistant's citations, share rather than raw count, so a high-volume Assistant does not win every row by default.

The takeaway names the single most influential source, clicking any cell opens the citation URLs for that source and Assistant, and a Top 10 sources overall strip below the matrix summarizes the leaders with each one's share of all citations to your tracked sources in the period.

A two-column grid of trend visualizations for the four user-selected hero sources (Reddit, Amazon, LinkedIn and YouTube by default). Each card opens on a Combined line with per-Assistant toggle pills. Drag a card by its handle to reorder them; the order is remembered. The Manage Influential Sources button sits in this section's header. These visualizations surface shifts in which sources Assistants lean on.

A modal (opened from the Manage Influential Sources button) to add, remove, activate, deactivate, or change which sources are promoted to hero status.

Auto Hero Selection is a switch that stars the four most-cited sources for you, refreshed each time you open the modal, and locks the stars while it is on; turn it off to choose them yourself, starting from whatever it last picked.

Exactly four are heroes, no more and no less, and the list holds up to 20 sources in total, leaving room for your own alongside the 9 that ship by default. Save stays disabled until four are starred, so changing a hero means unstarring one and starring another. A starred source is locked while it is a hero: its toggle and delete button are disabled, and unstarring releases both.

Inactive sources disappear from Influential Sources; active non-hero sources still appear in Top Sources by Assistant and Other Sources list.

A grid of all tracked sources beyond your hero sources, ranked by total citations. Each card shows the favicon and name, citations, unique URLs, and a per-Assistant breakdown. A takeaway highlights which secondary source matters most and flags any tracked platforms not cited at all.

Influential Reddit Contributors: ranks the Reddit users who contribute to the most distinct threads cited for your brand, up to 25 of them. Each thread counts once however many times it was cited, so someone active across many discussions outranks someone who happens to be in one heavily cited thread. Each user card shows the number of cited threads they contributed to, how many of those they started (as OP), their share of all cited threads, the subreddits they participate in, and a View Posts link.

Automated and moderation accounts are detected and kept out of the default list, since they are not people you can reach out to, and where a Space has any, People and Bots pills appear: the two are separate lists rather than one with the bots mixed in, because they answer different questions, who to approach and what is automated. The bots are still listed, because an automated account cited repeatedly really does shape the answers; the pill's hover names the accounts it holds. A minimum of 30 distinct cited Reddit threads is required before the list appears.

Thread and user extraction: when a Reddit URL is cited, Martekio fetches the thread and extracts the original poster, commenters, full text, subreddit, and date, powering the contributor analysis and thread viewer.

Inaccessible Reddit Citations: detects Reddit URLs that can no longer be reached (banned, private, quarantined, or deleted) and lists each one with its status, the Assistants that cited it, citation count, and last cited date, revealing where Assistants recommend content users cannot reach.

Similar Subreddits: discover and track subreddits related to those already in your citations, expanding your Reddit monitoring.

The headline section, five KPI cards scoped to the AI Overview surface. AI Overview Render Rate is the proportion of your tracked prompts where Google rendered an AI Overview, so it describes Google's behavior rather than your brand's.

Its card breaks the rate down twice: a donut splits your tracked prompts into the Overviews rendered for non-branded and for branded prompts and the queries that got none, and a Prompt category list beside it gives the render rate for each of your Prompt Categories, highest first, so you can see which parts of your catalog Google answers with AI and which it still answers with links. A category is listed once it has at least three tracked prompts, because with fewer a single result swings it between 0% and 100%. Each prompt takes its category by the same rules as everywhere else: today's category if it is still on your list, otherwise the one it had when the analysis ran. Click a category to open the Response Explorer on that category's AI Overview responses for the analysis.

The other four measure you within those rendered Overviews and are laid out as two pairs, one measure to a row: Citation rate · Non-Branded beside Citation rate · Branded, then Mention rate · Non-Branded beside Mention rate · Branded. Citation rate is how often an Overview cites your site as a source; Mention rate is how often it names your brand whether or not it links to you. Citations lead because they are the harder number and the one Share of Voice is built from. Non-branded leads each pair, being the more diagnostic signal for organic discovery.

Every card carries its change against the comparison analysis, and an Explore link that opens the Response Explorer on the rows it counted: the same analysis and Assistant, filtered to rendered Overviews and, on the four brand cards, to the classification plus the mention or citation the card measures.

A conversion matrix that maps your organic Google ranking against whether you are cited in the AI Overview, so you can see how traditional ranking translates into AI Overview presence and where the gaps are. Click any keyword in a quadrant to open its actual response in the Response Explorer, especially useful in Ranking overlooked, where you rank organically but the Overview cites someone else. The other three quadrants are Fully visible, AI-only visibility, and Invisible.

Mirrors Visibility Overview for the AI Overview surface: which competitors are most cited in AI Overviews, and how often Google links directly to your own domain. Rows navigate to the evidence: click a competitor (or an owned page's example query) to open the matching responses in the Response Explorer for the selected analysis.

With a comparison analysis selected, both lists show how each count has moved since it, competitors carry the same NEW and rank-movement tags as Visibility Overview, a newly cited page is tagged NEW, and anything the compared analysis cited that this one did not is named under No longer cited. The organic-ranking grid beside them shows the change in each of its four cells and in its headline rate, colored by whether the move is good news: more keywords Ranking overlooked is a rise shown in red.

Three trend sections that track Share of Voice, citation and mention volume, and sentiment over time, all scoped specifically to the Google AI Overview surface. Pills switch the first two between Total, Branded, and Non-Branded, and the Citations & Mentions visualization between Citations and Mentions.

A summary block pairing a headline Page Views number with an area visualization of the daily trend. Asset requests and bot traffic are excluded, so only real human page visits are counted. A meta line reports the number of days in the window and the average page views per day.

Tracks real human visits that arrived from a ChatGPT referral: did the Assistant actually send a customer? A bold KPI shows total human visits from ChatGPT with a paired trend visualization. This section is scoped to ChatGPT because it is currently the only AI product that reliably exposes both UTM tagging and a stable referrer; the scope will broaden as other Assistants ship comparable signals.

Three visualizations side by side: Traffic Proportion (AI bot share of total visits), AI Bot Distribution (a pie of the split by AI bot, built from whichever bots actually visited), and Weekly Bot Activity (the same split over time). Together they answer how exposed your site is to AI crawlers, which matter most, and whether interest is growing.

One row per date and bot, showing the date, bot name, visits, top page visited, and an HTTP status code breakdown colored by class (2xx green, 3xx blue, 4xx amber, 5xx red). Rows are grouped under collapsible month headers, all collapsed by default.

Search and sort: a search field filters by bot, page, or status code and auto-expands matching months. Every column header is sortable, defaulting to date descending.

Drill-down and export: the Top Page Visited cell opens a Top Pages Visited modal (up to 50 URLs with its own CSV download) when a bot hit more than one page that day. A Download Bot Visits button exports the full table to CSV.

Once Martekio has connected your server logs, Analysis Settings > AI Bot Tracking > Manage AI Bot Tracking opens in two tabs. AI Bots chooses which bots are counted. A few listed bots, such as bingbot and PetalBot, carry a Not an AI crawler tag: they are classic search crawlers, and while switched on their visits count as AI bot traffic.

Server Log reports what the last import read from your log: how many lines, how many were understood and how many were skipped, and why. Martekio reads the standard combined log format, the default on Apache and Nginx, so lines from another kind of log, or with extra fields at the end, are skipped. Examples of skipped lines are listed with their line numbers, and the download button at the bottom left saves up to 1,000 of them, spread across the file, for your developer. If most of a log cannot be read, the import is refused and your previous data is kept, rather than replaced with nothing.

The same tab lists Paths Not Counted as Human Traffic. A server log cannot tell a page a person opened from a request that page's own scripts make in the background, and a busy background request can outnumber real page views. Paths listed here are left out of human page views and ChatGPT referrals, while AI bot visits to them still count. Martekio marks a busy path Recommended when it looks like a background request: nobody arrives at it directly, it is always requested by your own pages, and its responses are a fraction of a page's size. Nothing is excluded until you choose it and save, and changes apply from the next import.

Your server logs are imported after each of your scheduled analyses, so Agentic Traffic displays the date of the last import, and the days since then are added at the next one. Each import reads only log files it has not read before, and a request that appears in two of your files is counted once.

If you would rather filter your logs before sending them, Preparing Your Logs at the bottom of the modal lists what you can remove, what must stay, and how to mask IP addresses, then explains how to upload your logs with the private key Martekio gives you, using rclone, the AWS command line tool or any other S3-compatible tool. It prints to paper or PDF for whoever prepares the files.

After each analysis, Martekio generates actionable recommendations for improving your Share of Voice. Each is typed (content, competitor, sentiment, coverage, and paid competition when a competitor runs ChatGPT sponsored ads beside your prompts), prioritized high, medium, or low, and grounded in the specific citation, reference, and competitor data from your latest analysis. Expandable supporting data reveals the metrics behind each suggestion.

Visibility Opportunities has two tabs over the same findings. Recommendations shows one analysis at a time, with an Analysis picker for stepping back through earlier runs. Recommendations are written fresh on every analysis, so a problem that persists is described again in new words rather than carried forward, which is why the same advice can appear several analyses running. Impact Actions is the other view: it groups those recommendations by the work they describe and ranks them.

Analysis after analysis, individual recommendations pile up and start to repeat. Impact Actions groups them by the work they actually describe, so an issue raised in six analyses appears once here instead of six times in six different wordings, and ranks them by how many separate analyses reached the same conclusion. Because each analysis is written without knowledge of the last, that agreement is evidence rather than repetition, which makes this your priority list: the changes most worth making, in the order your own data supports.

Display Supporting Data on any action reveals which analyses raised it, as a dot per analysis, along with the full detail from the most recent one and the complete history. Actions that stop being raised fall away on their own, and work you complete and that stops appearing is marked resolved. Ranking by recurrence begins once a space has three analyses; below that the list is ordered by priority.

Add Recommendations to Backlog. Each recommendation card has a clipboard icon that adds the item to your Backlog with one click. Backlog items persist across sessions. Open the list with the Visibility Backlog button at the top of this section, which carries a count of your open items, or from Visibility Backlog in the Tools menu in the top navigation bar; both open the same list. Click again to remove an item; an In Backlog pill shows the current state.

Mark backlog items implemented. The Visibility Backlog modal has two tabs, Backlog and Implemented, each showing a count. On any open item, click Mark as Complete to close it out: the item moves to the Implemented tab and records the date and the name of the person who recorded it, shown as "Completed [date] by [name]". A success toast confirms the move. The count on the Visibility Backlog entry in the Tools menu counts open items only, so it drops as you complete work.

To reopen an implemented item, switch to the Implemented tab and click Reopen; it returns to the Backlog and a toast confirms it. An implemented item cannot be removed directly, because removing it would destroy its record; reopen it first. Completion is per space and shared with everyone who can see that space, so a team or agency works the same list together.

During a free trial, the Impact Engine runs as a preview: every analysis still generates the full set of recommendations, but only the highest-ranked ones are shown. The rest appear as blurred cards behind a panel that explains what a subscription unlocks. Impact Actions works the same way, showing its top-ranked actions with the remainder behind a panel. Past analyses, dismissing, the Backlog controls and CSV downloads unlock with a full subscription. Everything generated during the trial is kept, so the complete history is waiting the moment you subscribe.

Open Analysis Settings and choose Define your Factset, which reads Edit your Factset once you have stated any. Add up to 20 short statements that are true and checkable, each with an optional link to where you publish it: when you were founded, how many countries you ship to, what your flagship product is called, which certifications you hold. Keep them specific. A short list of precise facts is checked far more sharply than a long list of vague ones, which is why 20 is the limit.

If you would rather not start from a blank page, Suggest from AI responses reads your recent answers and extracts the claims the assistants are already making about you. Each suggestion names the assistants that made it, so you can tell a claim one of them makes from one they all make. Edit them, correct anything wrong, and add the ones you want checked. Nothing is saved until you add it and press Save.

Suggesting takes a minute or two, because Martekio is reading your stored answers and working out which claims in them can be checked. A panel displays which step it is on and how long it has been going.

Each suggestion can also display what the assistant actually wrote: the sentence it came from, word for word, taken straight from the stored answer. The suggestion itself is our summary of the claim, so if you want to judge it properly, read the quote.

Where the assistants disagree with each other, the suggestion is displayed differently: instead of one line to add, you get each version they gave, the assistants behind it and the words each of them used. Add the version that is correct, or ignore them all and write your own. These are worth reading before anything else on the list: a claim your assistants cannot agree on is one your audience is already getting two answers to.

Your facts are written in your space's content language, the one set in My Organization, because they are compared against answers in that language.

Accuracy is the share of answers checked against your current list of facts that contradict none of them. An answer checked before you added a fact was never checked against it, so it does not count until the next analysis checks it. Open errors counts every finding still open; one seen only in answers checked against an earlier version of your facts is still open, and the figure says how many of those there are, which is why accuracy can read 100% beside an open error.

Hover a cell for the detail: how many open errors that assistant has on the fact, or how many of its answers were checked without a contradiction.

Findings appear in the Factset section of the Impact Engine, with a count of the open ones beside the heading. Each card puts what you state directly above what the assistants say, names the assistants that said it, and tells you how many separate analyses found the same thing, which is the measure of how settled the problem is. Display the evidence and you get the assistants' own words, every quote taken verbatim from the stored answer, with an Explore link that opens that exact answer, filtered to the assistant and analysis that gave it.

Below the quotes, Sources these answers cited lists the pages the assistants linked in those answers, most-cited first, with your own pages marked Your site. A page listed there was cited, which does not mean it states the claim: open it and check before contacting anyone.

Each finding is labelled with its origin, because what you can do about it depends on it. Cited sources means at least one of the answers linked to pages, so there may be a page behind the claim: check the cited pages, correct any of your own first, then ask the publishers of the others to correct theirs. Assistants that search the web pick up a corrected page in their next answers.

From model training means none of the answers cited a page, so the claim most likely comes from what the assistant learned in training, and there is no page to contact. The fix is to make the correct fact easy to find: state it plainly on your own site, for example on your About or history page with matching structured data, and correct it on the reference sources models learn from, such as Wikipedia and Wikidata. Assistants that search the web can pick this up quickly; answers from training change only with new model versions, so a finding like this can stay open for a while.

Sometimes the assistants' version is defensible and your fact is the one that needs to be more precise. A company with several founding dates is the classic case. Editing the fact in Analysis Settings is then the better fix.

Each card carries a Next step written for its origin. When you decide to act, use the clipboard button on the left of the card to add the finding to your Backlog, the same control as on recommendations; the Backlog item is titled for the work, checking the cited pages or publishing the correct fact.

When the work is done, mark it complete in the Backlog and the card reads Done, awaiting re-check. Martekio keeps checking, and marks the finding resolved once two analyses in a row no longer find the claim, so you get the confirmation rather than having to take your own word for it. If the assistants repeat the claim after you completed the work, the card reads Done, still being said, which is normal for a claim from training. A resolved finding reopens if the claim comes back.

The × button under the Backlog button dismisses a finding, and asks you why (on a dismissed finding, the same spot reopens it): if the assistant was right and your stated fact is out of date, say so, because that is the answer that makes the check sharper for everyone.

An empty Factset section is good news, not a missing feature: it means the assistants did not contradict anything you stated. If you have not stated any facts yet, it says that instead.

Every implemented item is a row in the table under the visualization, in the order you shipped them, with the date the advice arrived, the date it went live, the days between, and a Result icon. The prompts it is measured on are listed under its name; click one to open its responses in the Response Explorer. Click the Result icon to open the item's Attribution Result: three measures, Citations, Mentions and Share of Voice, each as a before-and-after pair with a chip saying whether it came out ahead of, behind, or the same as the Comparison Prompts, and a tile stating the conclusion in words. An item you added to the Backlog yourself, rather than from a recommendation, is labelled Client Added. The difference from the control is the finding, not the raw movement: if everything you track rose together, the work did not cause it.

AI answers do not change the day you publish. Pages have to be recrawled and the assistants asked again, so the three weeks after your go-live date are excluded from the measurement entirely: they are neither the before nor the after. A card shows Evaluation started until then. Once two analyses have run past that point, a result appears, roughly three and a half weeks after you shipped. It is marked as an early read while more analyses are still to come, because the figure firms up as they arrive. Where the analyses either side are too few or too close together to distinguish a real change from noise, the card says Not enough data rather than printing a number you should not act on.

The Implementation timeline tracks your non-branded visibility across the period your work spans, with every go-live date marked and a switch between Share of Voice, Citations and Mentions. On its own it is context rather than proof: it has no control behind it, so it includes everything that moved for reasons unrelated to you. A break in the line is a period with no analyses.

Hover over an item in the table and the visualization switches to that item's own two lines: its Impact Prompts in green and its Comparison Prompts in gray, the same figures its result is computed from. The periods its measurement used are shaded and labeled with their dates, Before, Excluded from analysis and After, and a red line marks the go-live date. Click an item's name to pin it, so the visualization holds it while you move around the page; click again, press Escape, or use the cross on the visualization to let it go.

Pin up to three items and a Compare button appears in the table header. It opens their results side by side: the result, the Impact Prompts before and after, the net change against the Comparison Prompts, the Comparison Prompts before and after, and a short What this means summary. The visualization and the table header stay in view as you scroll, so the visualization is always beside the rows you are hovering.

Automatic product data extraction from any e-commerce platform. Shopify stores use the .json API for fast, complete data (titles, descriptions, prices, variants, images, availability, barcodes). Other platforms are parsed for schema.org Product structured data, falling back to meta tags when none is found. No manual data entry required.

Each saved feed is a card showing its title, an Active and Published toggle, the URL-pattern and keyword-filter row, and a metrics row (product count, in-stock ratio, locale-variant merge count, stale badge). Feed History logs every generation with status, product count, duration, and who triggered it. After each run the feed is validated against the Martekio schema, with a modal summarizing a pass or fail verdict, error and warning counts, and a clickable quality breakdown (products with or without prices, images, and categories).

Download feeds in JSON (recommended), CSV, or TSV. Feeds are enriched with real Martekio citation data per product (citation count, citing AI Assistants, last cited date, and visibility trend) only where that exact URL has direct citations. An in-app guide explains the self-hosted deployment model: download your feeds, host them at a public HTTPS URL, and register that URL with the AI platform. OpenAI ACP settings (feed ID, account ID, merchant, country) are stored on the client profile and included automatically in feed headers.

An AI answer is a local answer: ask the same question from Mexico and from Spain and you often get different brands and different sources. The Market to check picker lets you run your prompt as a user in any of 101 markets would, covering the whole of Europe and the whole of Latin America, and it opens on your own Space's country so the common case is one click. Choosing another market also sets the content language for the check, because that is what a local user would be answered in. The market you last used is remembered for next time.

An Assistant that is not switched on for your Space is listed but cannot be selected, and says Not enabled. It is listed rather than hidden so you can see what Martekio measures and what you are not yet collecting; paid add-ons such as Grok and Copilot read this way until they are added to your plan. If a previous selection included one, it is dropped when you open the window, so the check never waits on an Assistant that would return nothing.

Some AI Assistants cannot be pointed at a country at all, because their APIs take no location: Grok is one. Others accept a market in general but not every one of them, so a handful of markets are unavailable on a particular Assistant. And Claude accepts an undisclosed subset of countries and may decline the one you chose, in which case it answers without it.

The modal says which of the Assistants you selected are affected before you run, each row is marked, and the results name the market and flag any Assistant that was not given it, so a difference between two Assistants is never mistaken for a difference between markets. An Assistant that cannot be given the market still runs and still answers: the language of your prompt carries much of the locality on its own.

For every Assistant you see its full reply plus the visibility insights computed from it: Share of Voice, your brand mentions and citations, competitor citations, and sentiment. The market the check ran in is named at the top. A summary highlights which Assistant gave your brand the strongest presence, and the same drill-downs you know from Response Explorer are available on each response.

Beneath the per-Assistant results, a Top Cited Domains panel ranks the ten domains cited most across all the Assistants you tested, pooled together. Each domain shows which Assistants cited it and how often, alongside a page and citation count, and expands to the exact cited URLs. Martekio crawls those pages for their titles, so a link reads as the page's real title with its URL beneath (falling back to the URL when no title is available). It is the fastest way to see which sources the Assistants leaned on for your prompt.

Google AI Overview can be selected like any other Assistant, but it is the one worth knowing about before you pick it. It does not answer directly: Martekio submits the prompt to a queue and waits, and a single check waits about a minute before giving up. A scheduled analysis has no such hurry, submits every prompt together and waits far longer, so the Assistant that reports nothing here will still be measured properly on your next run. If it returns an error on a check, read that as the queue being slow rather than as an absence of AI Overview for your prompt.

For questions about your own metrics, Tekka now selects from a toolbox of more than twenty read-only Martekio data tools and answers strictly from what they return, so every figure is your actual data, never an estimate. The toolbox spans the whole platform: current Share of Voice and its trend over time, sentiment by AI Assistant, your watched competitors and citation gaps, improving and declining prompts, your tracked prompt list, the subreddits and Reddit threads AI cites, your influential sources, AI crawler traffic and AI referrals to your site, ChatGPT sponsored ads on your topics, your industry benchmark, Google AI Overviews performance, the technical health of your cited pages, your analysis history, and who has access to your space.

Each data answer ends with a small Source line naming the Martekio data it was built from, for example Source: Share of Voice trends or Source: Impact Engine recommendations. You always know which Insights page would display the same numbers in full.

Ask "what should I do to improve my visibility?" and Tekka answers with your brand's own prioritized recommendations from the Impact Engine, generated from your latest analysis, rather than generic best-practice tips. She can also answer what to prioritize, from your Impact Actions, and what you are working on, from your Visibility Backlog: what has been picked up, what is finished, and whether a recommendation was ever acted on. For general methodology ("how does one improve AI share of voice?") she draws on Martekio's documentation instead, and for current tactics she can synthesize the latest Daily AI Visibility Research and weekly top tips.

Tekka also explains the platform itself: what a metric means, how an Insights page works, what Branded versus Non-Branded is, straight from the same documentation this guide comes from. And she is a full API reference: ask about any endpoint of the Martekio Public API and she answers with parameters, example responses, and working code samples in whatever programming language you ask for.

Tekka and the AI Assistant Connector share one set of Martekio data tools, so a question asked in Tekka and the same question asked in Claude or ChatGPT through the connector are answered from the identical data. Use whichever is closer to hand; the numbers always agree.

If your plan includes API access, you can also ask Tekka a question over HTTP and get her written answer back in the response, so her analysis can sit inside your own reporting tool, a reporting job, or a team chat bot. It is the same assistant, the same data, and the same answers you get here. Each space can ask a set number of API questions per day (25 by default), which resets at midnight UTC; questions you ask here in the app never count against that allowance. Answers typically take between 4 and 21 seconds, because Tekka reads your data before writing the reply. Full request and response details are in the API reference, under Tekka. Open the API reference.

The modal opens on two things: the connector's server URL, and your personal API token, masked until you reveal it. Both have a copy button beside them. For the supported path below you only need the server URL; the token is there because it is the same token the public API uses.

The supported path for both assistants is a sign-in, with no token to copy. Paste the Martekio server URL into the assistant's connector settings (in claude.ai, Settings then Connectors then Add custom connector; in ChatGPT, Settings then Connectors then Create). You are then sent to a Martekio sign-in and consent screen where you log in and choose which space to grant. The consent screen states exactly where your data will be sent, so you can confirm you recognize the destination before allowing access.

Note: custom connectors are not available on the free ChatGPT plan; you need a paid plan (and may have to enable Developer Mode) before the option to add one appears. Claude has no such requirement: custom connectors work on every Claude plan, including Free, which allows one custom connector. A desktop app signs in with the same account, so a connector added on the web is already there.

If your account has access to more than one space, a web or ChatGPT connection can report on any of them. You pick a default space at the consent step, then switch simply by naming the brand in chat, for example "now show me Arden Noir". A connection only ever reaches spaces you have access to, and that access is checked on every request.

An administrator can cut off programmatic access for everyone at once from My Organization, under Integration. There are two switches, one for the public API and one for the Assistant Connector, and they are meant for emergencies: a lost laptop, a departing contractor, an integration behaving badly. Turning one off stops every token and every connected assistant for that organization immediately. Nothing is deleted, so turning it back on restores access without anyone reconnecting. Insights pages keep working throughout.

For an agency, the parent organization's switches also cover every Space it manages, and a managed Space cannot turn its own access back on while the agency has it off. Each change is recorded against the account with who made it.

You need Owner permission on the property in Search Console. There, open Settings, then Users and permissions, then Add user, and add the Martekio service account email shown in the modal (a one-click copy button is provided) with Restricted permission. Restricted is read-only: Martekio can read your query data but cannot change anything on your property. Back in Martekio, enter your domain (suggested automatically in the form sc-domain:yourdomain.com) and click Connect.

A How to Set-up GSC button carries the full step-by-step instructions, and an Email GSC Setup button forwards those instructions to whoever owns the Search Console property (for example your SEO agency or web team) so they can add the service account for you.

The modal lists your top 100 search queries over the last 90 days, each with its clicks, impressions, click-through rate, and average position. Search Console data lags about two days behind, which is normal for Google's reporting. Refresh Keywords re-fetches the latest data from Google on demand, and Download GSC Queries exports the table as CSV.

Radar partners get a single page rather than the Insights pages: the Control Tower, which reports your prompt allowance, what you have used, when the allowance resets, and a log of your recent API calls, and lets you manage who on your team can sign in to it. The Insights pages are not offered because a Radar account is headless and has no data of its own to draw them from. Sign out from the Logout button at the foot of the left sidebar, the same place and the same button as everywhere else in Martekio.

The page holds your overall visibility (how many answers cite and mention your brand), your Share of Voice with its movement since the previous analysis, the Branded and Non-Branded split, Share of Voice by AI assistant, a trend over your recent analyses (tap a point to open that analysis), Share of Voice by Prompt Category (every assistant that ran is shown here, whatever you have chosen under Display Assistants for the desktop), your top competitors per assistant as a share of every citation that assistant made, with your own brand as the first bar and an All / Threats / Tracked switch, and brand sentiment by assistant.

One Non-Branded / Branded switch sits above those sections and scopes all four of them, Non-Branded by default: the assistant breakdown, the trend, the categories and the competitors, each of which names the facet it is showing in its own heading. Brand sentiment is the one section it does not scope, which is why that card is headed "all prompts": a sentiment score is read from a whole response, not from the kind of prompt that prompted it.

Cards and switches for things you do not track are left out entirely, see below. The one exception is Share of Voice by category, which stays and tells you how to define categories, since they are something to switch on rather than something you have decided against. Tap a competitor to open its website. Every figure is the one the desktop Insights pages show for the same analysis.

By default the page shows one analysis on its own, with no comparison. Tap Compare to an earlier analysis under your Share of Voice to pick one, and every figure on the page then carries how much it has moved since then, including each AI assistant, each prompt category, each sentiment score and each competitor. An assistant the compared analysis never measured shows no movement at all, rather than reporting every brand as new. Choose No comparison to go back.

Once one is chosen the label reads "13 Aug. compared to 10 Aug.", the analysis you are reading first. Only earlier analyses can be compared to, since the movements are always read as "since then", and choosing a different analysis to view clears the comparison.

Share of Voice movements are in points, the plain difference between two percentages, so 20.8% to 20.9% is +0.1 points and not 0.1% more. Sentiment moves on its own -1 to +1 scale, so it is never labelled points.

The page leaves out anything you have not set up, rather than showing it empty. If you track only Non-Branded prompts, the Branded card and the Branded / Non-Branded switch are not there at all, and the same the other way round. If you have not marked any competitors as Threats, the Threats pill is gone, and likewise for Tracked; with neither, the competitor section itself is gone. A pill that is still showing but finds nothing means something different and worth knowing: you curated that list, and this assistant cited none of it.

The identifiers (such as your domain) Martekio uses to recognize your brand in responses and citations, and to classify keywords as Branded or Non-Branded. Completed first; required before the sections below. Beneath it, Define your Factset (optional), which reads Edit your Factset once you have facts, is where you state what is true about your brand so Martekio can check what the AI Assistants say against it (see Factset).

The domains that belong to your brand, used to attribute citations back to you. Associate site labels (optional), which reads Edit site labels once any are assigned, lets you give each site a name of your own, up to 30 characters, such as "Garnier UK". A label you assign appears under the site's address in the Response Explorer's Brand Website filter.

Add every domain you own at setup, including your country sites. A domain added later has its stored citations credited to you the moment you save, and one you remove stops being credited; your Share of Voice history is then recalculated for the new list in the background, usually within a few minutes, and a note is added to your latest analysis saying what changed. While an analysis is running, My Profile, My Organization and Analysis Settings are unavailable from the left rail, so a list cannot change halfway through a run.

Unclaimed Brand Websites lists the cited domains that carry your brand name but are not in the field, typically your own country sites (the .de or .co.uk of your .com): tick the ones that are yours and Add to Brand Websites puts them in the field, ready for Save, or mark a namesake owned by another company as Ignore and it is treated as a competitor instead. It is built from citations, so it fills up after your first analysis.

Each row says why it is listed: Counted for nobody means the domain matches one of your websites apart from its ending, so its citations are set aside and count on neither side; Counted as a competitor means it only carries your name, so it is already counted as a rival, which is where dealer and partner sites appear alongside fan sites that have nothing to do with you.

The market your analyses run in and the industry used for benchmarking and prompt generation. Language is the language of that market, used with the country when the AI Assistants are asked your prompts, and Use local language Visibility Recommendations has your Visibility Recommendations written in that language rather than in English, the default. The timezone sits here too (see Timezone below).

The Manage Keywords & Prompts button opens the keyword and prompt editor, documented below. Every way of getting keywords and prompts in lives there, in this order on its toolbar: Import from CSV, Prompt Synthesis, Import from Search Console, Fast Add, Prompt Research, and one at a time. Prompt Research is new: Martekio researches your brand and suggests prompts for you. Fast Add takes several prompts at once, one per line, classifies each as Branded or Non-Branded, and lets you sort the review list by Prompt or by Classification before you add them.

Anyone on the Space can comment: its administrators in Manage Keywords & Prompts, and invited collaborators in their read-only view of the same list, which leaves out the editing tools. Each comment shows who wrote it and when, and a comment can be up to 2,000 characters.

Comments you have not read are counted in three places: on Analysis Settings in the left menu for administrators (on Keywords & Prompts for collaborators), on the Manage Keywords & Prompts button inside Analysis Settings, and on each row. Opening a row's comments marks them as read.

A comment belongs to the text of its keyword or custom prompt, so it stays with the row when you edit the list; change the keyword itself and the comments stay with the old wording.

Martekio researches your brand and suggests prompts for you to track, so you are not starting from a blank list. Choose the Country, which starts on your Space's market, and, where a country is written in more than one language, the Language to write in. Set how many Prompts to generate: up to 50 per run and at least one for each user intent you select; the suggested number is a little more than your remaining prompt slots, so you have a choice. Pick the Classification split between Non-Branded and Branded prompts, and the User intents to cover: Research, Problem / Need, Recommendation, Competitor Comparison, Buyer Intent, Alternatives, and Use Case / Persona. The prompts are spread evenly across the intents you select, so the list never leans on one kind of question.

Martekio then reads your website, researches your competitors and the people who buy from you, and writes each prompt as one short question in the language you chose. A run usually takes one to two minutes. A progress window reports each step as it finishes, and the research continues if you close it, so you can return to Prompt Research later.

The results open with What the research found: a short summary of your brand, the competitors it identified and the buyer personas it wrote for, so you can judge at a glance whether your business was understood before reading the prompts. Each suggested prompt is listed with its user intent, the persona it was written for, a one-line reason, and its Branded or Non-Branded classification against your Brand Identifiers. Filter the list by intent, tick the prompts to keep, and click Add. Prompts are pre-selected up to your remaining slots, one intent at a time, and a prompt already on your list is marked Previously added. Added prompts join the list as Custom Prompts, sent exactly as written, and are only tracked once you click Save. Your latest run is kept for the Space, and Run Again starts a new one with the same settings.

Each Space can start Prompt Research five times a day, and the count resets at midnight UTC. A run counts as soon as it starts, so one that fails or is interrupted still uses one of the five; once they are used, starting another tells you to try again tomorrow. Collaborators with restricted access can open the latest results but cannot start a run.

The user intent of each prompt, the kind of question it asks: Research, Problem / Need, Recommendation, Competitor Comparison, Buyer Intent, Alternatives, or Use Case / Persona. It is a fourth, independent axis beside the Branded classification, the Keyword / Custom type and your Prompt Category, with one difference: categories are your own filing, while intents are Martekio's standard list, the same in every Space, so your results can be compared.

Set an intent from the dropdown in the Intent column on any row. Prompts added by Prompt Research arrive with theirs already set. To fill the rest in bulk, Classify Intents has Martekio suggest an intent for each prompt: it classifies only the prompts with no intent, so one you chose is never changed, unless you choose to reclassify every prompt. Review the suggestions, change any you disagree with, untick any to leave as they are, then save. Like the categories window, it saves straight to your list, and any unsaved changes in Manage Keywords & Prompts are saved before it opens. It is also in Analysis Settings, beside Manage Categories.

An intent is a description, not a change to what is measured, so changing one never creates a new keyword set version.

Your own labels for grouping the prompts you track, for example Competitive, User Need or Industry. This is a third, independent axis: it is separate from the Branded classification and from the Keyword / Custom type, so a prompt carries all three. Pick a category from the dropdown in the Prompt Category column on any row. The last option in that dropdown, Manage Categories, is where you create, rename, reorder and delete the categories themselves. If you have unsaved changes to your keywords and prompts, they are saved before it opens. Every space defines its own and nothing is created for you, so an empty list is the normal starting point.

To have one suggested, Generate Categories, at the foot of the manage window, reads the prompts you send to the AI Assistants, branded and non-branded, and proposes up to 10 short categories (three words at most), each with how many prompts it holds. Review them before anything is saved: rename any category, click its prompt count to display its prompts and move any to another category, then Save. Saving replaces your existing categories and files your prompts under the new ones, so you are asked to confirm first when you already have categories; prompts that fit no suggestion become uncategorized. Only your current prompt list is changed, never the record of past analyses.

To build a list by hand instead, Import Categories, at the foot of the manage window, takes a pasted list, one per line, so a column copied from a spreadsheet works as it is; imported names are added to whatever you already have, never replacing it, and anything already present, longer than 64 characters, or past the 50 category limit is skipped and counted for you before you commit.

The same window opens from Manage Categories in Analysis Settings, beside Manage Keywords & Prompts, so you can reach it without opening the prompt list first. Renaming a category keeps every prompt filed under it, while deleting one leaves those prompts uncategorized, so the manage window shows how many prompts each category holds before you change anything. That count is a link: click it to list the prompts filed under the category, and click any prompt there to open it in the Response Explorer. Categories are optional: leave a prompt uncategorized and nothing else changes.

Once you have categories, a Coverage line beside the search box shows how many prompts sit in each one, updating as you file them and before you save. Until any prompt is filed it reads No prompts categorized yet, with a Manage Categories button to set your categories up. From the first filed prompt, every category is listed, including those at zero, because an empty category is usually the point: most spaces discover they track a great many prompts naming their own brand and very few genuinely competitive ones, which is often the reason competitive visibility looks thin.

Switch the line from Category to Intent to count your prompts by user intent instead, every intent listed including those at zero: Buyer Intent 0 means your list never asks a buying question. The button beside the intent counts reads Classify Intents until any prompt has one, then names how many prompts still have none, and Reclassify once every prompt has an intent.

When a teammate invites you to a Martekio Space, you receive an invitation email. If you are new to Martekio it walks you through validating your account and signing in; if you already have an account it simply confirms you have been granted access to the new Space. Invitations are covered in more detail under Collaboration in the Navigation Bar.

An invited collaborator sees every Insights page and can read, filter, export and generate reports, but does not change what the Space tracks. Controls that would rewrite the Space for everyone are hidden rather than shown and refused: Curate Competitors, Manage Influential Sources, Search Console setup, and dismissing a recommendation. Ask the Space owner or an admin if you need one of those changed.

The Organization member switch on the invitation form applies to a new account only. Member or guest is a property of the person, not of a Space, so somebody who already has a Martekio login keeps the setting they have and appears under the same tab everywhere. If you invite an existing account, the confirmation tells you which they are.

To change it afterwards, open their details from Manage Collaboration. Bear in mind the change follows them into every Space they collaborate on.

On an account that holds more than one Space, Manage Collaboration is available from My Martekio Spaces as well as from inside a Space, and there it acts on the account rather than on one brand. The invitation form names which it is, so you always know what you are granting before you send it.

An invitation to the account carries an Account administrator switch, off by default, as is Can invite others. With it on, the person reaches and manages every Space in the account, including Spaces added later, with no further invitations. With it off they reach the account only and will land on a page with no Spaces on it, so to give someone the brands without making them an administrator, invite them from inside each Space instead. An administrator can always invite others, so that switch is included and cannot be turned off separately.

Inviting from inside a Space is unchanged and grants that Space alone: the person sees that brand and none of its siblings.

An administrator of a Space can appoint another one. Open Manage Collaboration, click the collaborator's name, and turn on Administrator under Access. Administrators can invite, configure and remove collaborators; a plain collaborator reads the Insights pages but does not manage the Space.

The same switch turns the role off again, with one exception: a Space's only administrator cannot be demoted, because nobody would be left to manage it. Appoint the replacement first. That also applies to yourself, so if you want to hand a Space over and leave it, make somebody else an administrator, ask them to turn your role off, and Leave then appears in My Profile.

Open My Profile from the left-hand rail. The Spaces you can access card lists every Space you have been invited to, with your role on each, and a Leave button beside it. Leaving takes effect immediately and nobody is notified, so an administrator of that Space would have to invite you back.

Two kinds of Space cannot be left, and the list says which. A Space you administer, because giving up that responsibility should be deliberate: ask another administrator to remove your administrator role first, or contact Martekio. And your only Space, because you would no longer be able to sign in. If the Space you leave was the one you normally land on, Martekio moves you to another Space you can access.

A Space you reach through your role on a parent organization is not listed here, because there is no individual invitation to give up. Ask an administrator of that organization.

Single-number summary cards for a headline metric. Martekio uses three styles: Hero cards (one row per AI Assistant, with the leader marked in green), Comparison cards (your value next to an external baseline, with a colored delta pill), and Summary cards (a headline number paired with its own trend visualization). The green color marks the winning value; there are no trophy badges.

Line and area visualizations for trends over time, grouped and stacked bars for comparisons, heatmaps for time-of-day patterns, and small-multiples grids that repeat one visualization per Assistant. Hover any series for a tooltip, and use a visualization's three-line menu to download it as an image. Visualizations respond to the section's filter pills and the analysis date picker.

Sortable, filterable tables (the largest, such as Response Explorer, are full data grids). Click a column header to sort, type to filter, and click a row or its drill-down icon to open a detail modal. Most tables offer a CSV download of the current filtered view.

Small colored cues: green for positive or leading values, amber for warnings or partial states, red for problems or negative sentiment. Examples include the Citation Consistency bands (green Consistent, amber Inconsistent, red Underperformer), Momentum pills on competitive threats, HTTP status colors in the bot table, influential-source amber badges, count badges on section headers, and stale or validation indicators on feed cards.