AI Visibility Tracker
Track how often AI engines (Gemini, ChatGPT, Perplexity) mention your brand over time, with weekly email reports. Runs free on Google Gemini.
AI Visibility Tracker for the trend line Search Console can't give you
Someone asked you to add an 'AI visibility' line to the monthly report, and you're comparing a paid tracker against the Search Console data you already have. Here's the honest version: Search Console's own generative-AI report gives impressions only, merges AI Overviews with AI Mode into one number, and has no query dimension at all. A tracker that runs the same prompts on a schedule tells you something Search Console structurally cannot — but only if it runs enough of them, and only until the model version underneath it changes.
Built and reviewed by Sayed Hasan, founder of SEOs Hut · Updated 28 September 2026 · Free, no signup
Is Google visibility the same as AI visibility?
No. Google Search Console's dedicated Generative AI report shows impressions only — no clicks, no CTR, no position — and it merges AI Overviews with AI Mode into a single figure with no query dimension. Clicks from those surfaces exist only inside the undifferentiated main 'Web' total. An AI visibility tracker measures something different: whether your brand is named in an actual AI answer, sampled over repeated prompts.
Search Console does not show you AI Overview or AI Mode performance
It shows you impressions — nothing else. The dedicated Generative AI performance report has exactly four dimensions: Pages, Countries, Dates and Devices. No query dimension, no click count, no CTR, no position. It also merges AI Overviews and AI Mode into one figure, so you can't even tell the two apart. If two links from the same site appear in one generative feature, Google counts that as a single impression, not two. Whatever your dashboard calls 'AI visibility' from Search Console data, it is not measuring what most people think it's measuring.
What each measurement surface could actually report, by date
This is the real problem with any 'AI visibility over time' chart: the instruments and the models underneath it both changed while you were collecting the data.
AI Overviews start appearing in the Performance report
Google's changelog clarifies that AI Overview impressions are logged in the main Performance report — inside the ordinary Web totals, not broken out.
AI Mode data joins the same totals
Google's changelog: AI Mode data now counts towards Performance report totals — merged with AI Overviews, with still no way to separate the two.
Bing ships the first citation-specific report
Bing Webmaster Tools' AI Performance report launches in public preview — Total Citations and a sample of grounding queries, explicitly not placement or ranking data.
Google announces a dedicated Generative AI performance report
A report built specifically for gen-AI surfaces, separate from the main Performance report — impressions only, still no query dimension.
The dedicated report finishes rolling out worldwide
Available globally, with the same four dimensions it launched with: Pages, Countries, Dates, Devices.
Still no click, CTR or query data for gen-AI surfaces, from Google's own tools
Anything calling itself an 'AI CTR' or an 'AI Overview click count' from Search Console is describing a number Google does not publish.
What each surface reports, and what it leaves out
| Surface | What's reported | What's missing | Counted separately from ordinary search? |
|---|---|---|---|
| Search Console: Generative AI report | Impressions, by Page / Country / Date / Device | No clicks, no CTR, no position, no query dimension | Merges AI Overviews and AI Mode into one number |
| Search Console: main Performance report | AI Overview treated as one ranked position; a click on a link inside it counts as a click | No breakdown of which specific link inside the AI Overview got the click | Folded into the same totals as ordinary Web results |
| Bing Webmaster Tools: AI Performance report | Total Citations, cited pages, a sample of grounding queries | Explicitly not placement, ranking or authority data, by Bing's own wording | Its own separate dashboard |
| A repeated-prompt tracker | Named-in-answer rate across a fixed prompt panel, run on a schedule | No standard formula — vendors rarely publish their counting rules | Independent of Google or Bing entirely |
Scroll the table sideways to see every column.
None of these four report the same thing, which is exactly why two tools — or two tabs of the same Search Console property — can disagree without either being wrong.
From a spot-check to a programme you can defend
Manual spot-check
A handful of prompts, run by hand, whenever someone asks. Free, and honestly labelled as directional only — not a trend, just a look.
- n = 5-10
- No fixed schedule
- Fine for a one-off question
Scheduled panel
The same fixed prompt list, run on a fixed cadence, so at least the comparison is apples to apples over time. The questions buyers actually ask change faster than any keyword list you built the panel from, so plan to refresh it periodically.
- Same prompts, every time
- Weekly or monthly cadence
- Comparable period to period
Fixed-n statistical panel
A sample size chosen to hit a stated confidence-interval width before you start, not discovered after the fact. Published research on citation-share stability found the CI width crosses a 0.05 target at roughly n≈30 for Gemini and n≈40-50 for other topics.
- n chosen before the run, not after
- Width target set in advance
- Model version logged with every run
Cross-engine committed panel
Accepts that some engines simply won't reach your target precision. The same research found no fixed-n protocol gets SearchGPT to a 0.05 CI width within a realistic budget — so for that engine, you're committing to a directional read forever, not chasing a number that doesn't exist yet.
- Google/Gemini figures treated as statistical
- SearchGPT figures treated as directional only
- Budget fixed in advance, not precision
Your trend line has version changes hiding inside it
Two separate problems break comparability, and neither is fixed by running more prompts.
The first is that you often can't pin which model answered your prompt. OpenAI exposes a system_fingerprint field and says outright that you may see different outputs 'due to changes we've made on our systems' — an acknowledgment that the model behind an identical API call can shift without notice. The Responses API has no seed parameter, and Anthropic's Messages API has neither a seed nor a fingerprint field at all. So a tracker that shows a flat line for three months and then a jump has no documented way to tell you whether that jump is a real visibility change or a new model underneath the same prompt.
The second is that citation distributions are not stationary even within one sampling run — the same published research that set the n≈30-50 sample-size targets also found that the underlying distribution shifts across a query sequence, which is exactly why an early-stopping rule ('it looks stable, let's call it done') is unreliable. A single measurement was never a trend; a chart built from single measurements strung together isn't one either. If you want to know why the number moved, rather than just that it did, read the actual answers behind the change before reporting the direction to a client.
The audience your trend line is trying to measure
Four ways the trend line lies to you
Each of these looks like a real signal and is actually an artefact of the setup.
Comparing your score against a competitor's from a different vendor
What happens: Two tools rarely use the same formula, the same prompt set, or the same counting rule for a brand named more than once — a disagreement between them proves nothing about either brand's actual visibility.
Do this instead: Track your own number against your own history, from one tool, on one schedule. Treat a second vendor's number as a different metric with a similar name.
Reporting Search Console 'AI clicks' or 'AI CTR'
What happens: The dedicated Generative AI report is impressions only. Any click, CTR or position figure attributed to AI Overviews or AI Mode specifically is not a number Google publishes — it's inferred, and usually wrong.
Do this instead: Report the impression trend as impressions, and be explicit that clicks from these surfaces are only visible inside the undifferentiated main Web total.
Not logging the model version behind each measurement
What happens: A visibility jump that lines up with a known model update looks identical, on the chart, to a jump caused by a real change in your content or competitors.
Do this instead: Note the date of any known model or product update next to your trend line, the same way you'd note an algorithm update on a rankings chart.
Stopping the sample the moment the number looks flat
What happens: Published research found the citation distribution itself shifts within a query sequence, which makes 'it stopped moving, so we're done' an unreliable stopping rule rather than evidence of stability.
Do this instead: Commit to a fixed n before you start, based on the target confidence-interval width, not on how the chart looks halfway through.
What actually belongs in the monthly report
Report the impression trend from Search Console as impressions, full stop. Report your prompt-panel number with its sample size and schedule attached, and flag the months where a known model update might explain a jump. If the trend moves in a way you can't explain, go back and read the actual answers behind it rather than reporting a number nobody has traced to a cause. A drop in AI fetcher activity in your logs usually shows up before a drop in citations does — check that before you assume the content itself stopped working, and find out who took the ground you lost once the trend confirms you actually lost any.
AI Visibility Tracker FAQ
How to track AI visibility?
What are the best AI visibility tracking tools?
Why should I track AI brand visibility?
Can Semrush or Ahrefs replace a dedicated ChatGPT rank tracker?
What is a 'good' AI visibility score?
Why do two AI visibility tools show different numbers for the same brand?
Does a flat AI visibility trend mean nothing changed?
Primary sources used on this page
- Google: the Generative AI performance report — dimensions, aggregation rule, impressions only — support.google.com
- Google: AI Overview position, click counting, and the AI Mode follow-up reset rule — support.google.com
- Google Search Central: instrumentation changelog, AI Overviews and AI Mode dates — developers.google.com
- Google: the dedicated gen-AI report's announcement and global rollout dates — developers.google.com
- Microsoft: the Bing Webmaster Tools AI Performance report and its documented limits — blogs.bing.com
- Sample-size and non-stationarity findings for citation-share tracking — arxiv.org
- OpenAI: system_fingerprint and why identical calls can return different outputs — developers.openai.com
Read the trend, then read behind it
A trend line only tells you that something moved. These five help you find out why.
Read the guide: How to Check Your Website Keyword Rankings (Free) and Is Local SEO Dead? No, but It Has Changed. Want it handled for you? See our AI SEO (GEO) services.