What do AI visibility tracking tools actually measure?
AI visibility tracking tools measure one of four units, and no two products count them the same way. Bing began reporting its unit free in preview and Google followed in 2026; read the unit before you read the chart.
- Citation: a link to your page shown as a source in an AI answer (Bing's unit).
- Impression: a link to your site shown in an AI Overview or AI Mode response, clicked or not (Google's unit).
- Mention: your brand named in an answer to a prompt the tool sent, linked or not (most paid trackers).
- Share: your citations or mentions divided by everyone's, for the same query or prompt set.
For marketing leads in Singapore, Malaysia, Australia, the US and Canada, this post settles one decision: whether to stop at free reports or buy a tracker, and which.
Our position: start with the three free first-party reports. For Google and Copilot they answer "are our links shown and cited?", with Google reporting impressions and Bing a sample; neither counts unlinked brand mentions. A paid tracker earns its place only when you need engines Google and Microsoft do not report on, or share against named competitors on prompts you chose. Buy on engine coverage and prompt ownership; the dashboard is the least important part of the purchase.
This is the measurement layer of our AI visibility optimisation work.
What do Bing Webmaster Tools and Search Console report for free?
Bing Webmaster Tools reports AI citations by page and by grounding query, with a citation share per query; Search Console reports AI impressions by page, country, date and device.
Bing went first. Its AI Performance report entered public preview on 10 February 2026, covering "Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations". It counts total citations, average cited pages and grounding queries, defined as "the key phrases the AI used when retrieving content".
"It is also about whether your content is cited and referenced when AI systems generate answers."
Krishna Madhavan, Meenaz Merchant, Fabrice Canel and Saral Nigam, Product Managers, Microsoft AI, Bing Webmaster blog, 10 February 2026
On 16 June 2026 Bing began rolling out, in preview, intents, topics, citation share and a compare view. Citation share is "the percentage of citations attributed to your site out of all citations shown across all sites for that same grounding query."
In our own account export on 27 October 2026, the grounding-query file carried five columns: Grounding Query, Intent, Topic, Citations and Citation Share. A Topics Portfolio beta followed on 13 October 2026, listing Top Opportunity Queries where other sites appear and yours does not, with volumes shown as ranges. For anyone working the Microsoft Bing ecosystem, Bing's report names the phrases an AI engine used to find you, and Clarity's Citations dashboard shows grounding queries too.
Google's side is thinner. The generative AI performance report reached all sites on 31 August 2026 and documents one metric: "Impressions are how many times links to your site were shown to a user in a generative AI feature on Google Search." It covers AI Overviews and AI Mode, excludes Search Labs, runs on Pacific Time and keeps the 1,000-row limit. The documented dimensions are page, country, date and device, so there is no query dimension.
Bing says its data "represents a sample of overall citation activity", and the documentation describes citations for Bing and impressions for Google, not clicks. Report both as presence signals. Never present a citation or an impression as a visit.
Microsoft Clarity adds a third free view. Citations became generally available on 13 May 2026 with grounding queries, cited pages, share of authority and AI referral traffic, drawing on Bing Webmaster Tools and Clarity signals; our Clarity citations explainer covers setup.
AI visibility tracking tools compared: engines, units, exports
Five paid trackers name between five and 16 engines on their own documentation, against two for each free report. Every fact below comes from the vendor's own pages, read 27 October 2026.
| Tool | Type | Engines named | Unit measured | Your own prompts | Competitor view | Export or API | Last verified |
|---|---|---|---|---|---|---|---|
| Bing Webmaster Tools AI Performance | First-party, free (preview) | Copilot, AI summaries in Bing, select partners | Citations, cited pages, citation share | No (Bing's grounding queries) | Share per query, no names | Export (seen in our account, 27 Oct 2026) | 27 Oct 2026 |
| Search Console generative AI report | First-party, free | AI Overviews, AI Mode | Impressions | No | No | Export; 1,000-row limit | 27 Oct 2026 |
| Microsoft Clarity Citations | First-party, free | "Supported AI experiences", not listed | Grounding queries, cited pages, share of authority, AI referrals | No | Share of authority (basis not stated) | Not stated on the page read | 27 Oct 2026 |
| Profound | Third-party | 8: ChatGPT, Perplexity, Claude, Gemini, AI Overviews, Copilot, Grok, DeepSeek; per contract | Visibility score (mentions), share of voice, citations, sentiment | Yes, run daily | Yes | REST API, Tableau connector | 27 Oct 2026 |
| Peec AI | Third-party | 6 standard (ChatGPT, AI Overviews, AI Mode, Perplexity, Gemini, Copilot) plus 10 plan-tier add-on upgrades incl. Claude, Grok | Visibility, position, sentiment, sources | Yes, daily, per country | Yes | CSV, Looker Studio, API | 27 Oct 2026 |
| Otterly.AI | Third-party | 7: AI Overviews, ChatGPT, Perplexity, AI Mode, Gemini, Copilot, Claude (via API) | Brand mentions, URL citations | Yes | Yes | CSV, PDF, API | 27 Oct 2026 |
| Semrush AI Visibility Toolkit | Third-party | 5 across reports per the page read: AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini; Prompt Tracking names AI Overviews, AI Mode and ChatGPT | AI Visibility Score (mentions), share of voice, sentiment | Yes (Prompt Tracking) | Yes, auto-identified | CSV (1,000 rows per export), PDF | 27 Oct 2026 |
| Ahrefs Brand Radar | Third-party, AI add-on | 8: AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini, Copilot, Grok (temporarily unable to gather new data), Claude (custom prompts only) | Mentions, citations, impressions, share of voice | Yes, plus prompts from People Also Ask data | Yes | API, Looker Studio | 27 Oct 2026 |
Peec reaches 16 because it counts Claude Sonnet and Claude Haiku separately, and Ahrefs lists Grok although it is temporarily unable to gather new data after policy changes. Profound says it draws insights "from the consumer experience, not API outputs"; Ahrefs runs prompts "through the free, publicly available web interfaces"; Otterly tracks Claude "via API, using Haiku model with web search capabilities"; the Otterly page does not state how its other engines are collected.
On its impressions estimate, Ahrefs states: "They show potential visibility, not actual audience reach." On access, Google's AI optimisation guide, updated 10 July 2026, says: "No third-party tool has access to our internal ranking or AI systems." Every paid tracker samples answers to prompts it sends.
Which tool answers which question?
Each measurement question has one cheapest honest source, and for Copilot and Google that source costs nothing. Match the question first, then the tool.
| Question you are asked | Cheapest honest source | Unit | What it cannot tell you |
|---|---|---|---|
| Are we cited in Copilot, and for which phrases? | Bing Webmaster Tools AI Performance | Citations and citation share per grounding query | Clicks (not documented); who else is cited; full counts (sampled) |
| Do AI Overviews and AI Mode show our links? | Search Console generative AI report | Impressions by page and country | Queries; clicks and position are not documented |
| Do AI assistants send us visits? | Clarity Citations or your analytics referral report | AI referral sessions | Answers that named you without a click |
| Does ChatGPT, Perplexity or Claude name us for our buying questions? | Your own prompt-set probe; a paid tracker at volume | Mention rate per prompt | What real users were shown; answers vary run to run |
| What is our share against three named competitors? | A paid tracker | Share of voice on the tracked prompts | Anything outside that prompt set |
Each layer is added only when the one below cannot answer the question in front of you.
Your gap may be ChatGPT coverage, or it may be competitor share, and the fix differs. Send us your Bing grounding-query export and Search Console AI report and we return a prompt-set map with a buy-or-wait call per engine, starting with a short call.
Which AI visibility tier do you need? A tool chooser
Three tiers cover the choice: free reports alone, plus your own prompts, or a paid tracker; the chooser picks one from your engines, prompts and rival-share need.
Formula: gap engines = the ticked engines among ChatGPT, Perplexity, Gemini and Claude. No gap engines and no competitor need gives Tier 1. Otherwise, 25 prompts or fewer gives Tier 2; more than 25 gives Tier 3. The shortlist names tools whose own documentation, read 27 October 2026, names every ticked engine (Semrush on its Prompt Tracking engines, Profound on AI Overviews for Google). Illustrative heuristic, not a vendor method.
Tier 3: paid tracker
2 gap engines and 40 prompts is more than a monthly hand check can hold. Shortlist: Profound (Google: AI Overviews only), Peec AI (Claude as a plan-tier add-on), Otterly.AI, Ahrefs Brand Radar (Claude on custom prompts only). Confirm your plan covers each engine.
The same logic as a static table
| Your situation | Tier | What you run |
|---|---|---|
| Only Google and Copilot matter, no competitor share needed | 1 | Bing AI Performance, Search Console generative AI report, Clarity Citations |
| Any gap engine or competitor need, 25 prompts or fewer | 2 | Tier 1 plus your own prompt set, checked monthly by hand |
| Any gap engine or competitor need, more than 25 prompts | 3 | Tier 1 plus a paid tracker whose documentation names every engine you need |
Why 25: at four engines that is 100 answers to read and log each month, roughly what one analyst can check by hand in a day, on our estimate.
How do you build a prompt set you own?
A prompt set you own is 20 to 40 buyer questions from sales calls and Bing grounding queries, kept in your own sheet and loaded into any tool you trial.
A vendor's prompt library measures the vendor's idea of your market. Ahrefs builds its database prompts from People Also Ask data, not your buyers' wording. If you switch tools with your own sheet in hand, your trend line survives the switch.
- Pull the Bing grounding-query export and keep the rows with a commercial or research intent.
- Add the questions sales hears on calls, in the buyer's words, for each of your five markets where wording differs.
- Tag each prompt with the engine that matters most for it: ChatGPT for research questions, Copilot for Microsoft-heavy buyers, AI Overviews for anything people still type into Google.
- Run the set monthly and log three things per answer: named or not, linked or not, which competitor appeared.
Made-up round numbers, not client data. A Singapore insurer writes 30 prompts and checks them on ChatGPT and Copilot each month. In month one it is named in 6 of 30 ChatGPT answers (20%), and Bing shows citations on 9 grounding queries. Tier 2 is enough until the board asks for share against three rivals across Australia and Canada, which takes the set past 25 prompts per market: Tier 3.
Expect noise. Answers change between runs and between modes; our analysis of ChatGPT reasoning mode citations shows a brand can be cited in one mode and missing in the other. The scoring maths sits in our share-of-authority measurement stack, and the content fixes in the GEO playbook.
Do this today
Export two free reports this week, before anyone signs a tracker contract: the Bing grounding-query export and the Search Console generative AI report by country.
Then write your 25-prompt set from those exports and a week of sales calls. Ask every vendor, in writing, which engines your plan covers and whether you can load your own prompts. Then ask how answers are collected, consumer interface or API, and shortlist only tools whose own documentation answers both.
Leapbuzz runs AI visibility optimisation from Singapore for brands across Singapore, Malaysia, Australia, the US and Canada, and the first deliverable is that prompt set, because it outlives any tool you choose.