The AI visibility metrics worth tracking are the ones tied to a fixed question set, split by engine, and reported as change over time. Blended scores and raw mention counts move for reasons that have nothing to do with your work.
Most tools count mentions while your analytics show no matching traffic. That gap is the whole subject of this post: which signals survive scrutiny, which do not, and how to tell the difference from your own data.
What does an AI visibility metric actually measure?
An AI visibility metric starts simple: does an AI system mention you? But “mention” covers a lot of ground.
One mention might be your company name appearing as a cited source in a Google AI answer, where the reader can see you and click through. Another might be ChatGPT paraphrasing your content without naming you. A third might reference your brand with no link for the user to follow.
All three read as “mentions” to a tracking tool. They deliver very different value. A platform that reports them as equal is hiding the signal in the noise.
Real metrics separate those cases. Position in the answer matters. Frequency matters. Whether the reader can see your name and click matters.
Why the engines cannot share one score
Here is the finding that broke our own first attempt at a blended number. Between 31 May 2026 and 25 August 2026 we ran 10,872 probes across five engines, and they did not behave alike at all.
| Engine | Probes | What it did |
|---|---|---|
| Google AI Overviews | 7,494 | Produced an AI Overview on 3,400, or 45.4% |
| ChatGPT | 913 | Answered every probe |
| Perplexity | 911 | Answered every probe |
| Claude | 792, since 20 June 2026 | Answered every probe |
| Gemini | 762 | Usable citation data on only 94; last probe 3 August 2026 |
As at August 2026, Google is the only one of the five that often declines to generate an AI answer at all. The others always answer something.
Average those together and you get a number where “Google did not answer” and “ChatGPT answered and named a competitor” collapse into the same missing citation. They are not the same event and they need different fixes. Report the engines separately or the metric is a fiction.
Gemini in that table is the second lesson. When an engine stops returning usable data, the honest move is to mark it and stop counting it, not to carry a flat line and let the reader assume it means zero mentions.
Why your tool counts mentions but your analytics do not show the traffic
This is the frustration most teams face. The visibility tracker reports a spike. Analytics shows little or no matching traffic bump.
Three things cause it. AI answers reduce clicks, because the reader gets a summary inline and often does not need a source. Paraphrased mentions hide your brand, so the reader does not know to click through. And many AI answers include no clickable link at all, so the mention happens with no traffic pathway.
A tool that counts mentions is reporting activity. It is not reporting business value, and the two diverge more than most dashboards admit.
Which metrics actually predict results?
Three signals separate movement from noise.
Traffic sourced from AI engines. Measurable in Google Analytics under Acquisition. Is referral traffic from AI systems growing month on month? This is the only one of the three that connects directly to money.
Page-level citations, not brand mentions. If a specific page is cited as a source for a competitive query, that page is the one to watch. Brand-level counting tells you nothing about which work paid off, which is the same flaw that makes sitewide traffic useless for measuring SEO ROI.
Change against a fixed question set. A jump from three mentions to ten is momentum. A jump from 150 to 155 is noise. Both are meaningless if the questions changed between runs.
Split branded from non-branded before you read anything
This is the single most common way an AI visibility report flatters its owner. Probing your own brand name asks whether the engine has heard of you. Probing buying questions asks whether it recommends you. The gap between them can be total.
Our own numbers over the 60 days to 25 August 2026:
| Bucket | Probes | Times cited |
|---|---|---|
| Branded, questions containing our name | 68 | 14 |
| Non-branded, questions a buyer would type | 876 | 0 |
The engines know who we are when asked by name, and never reach for us otherwise. A report that mixed those two buckets would have shown us a respectable-looking citation rate and told us nothing true. The full write-up, including what we changed, is in 876 probes, zero citations.
When does AI visibility work deserve a place in your roadmap?
Before adding it, ask three questions.
Are your customers using AI to search? If your audience asks ChatGPT or reads AI Overviews to find solutions in your space, visibility there matters. If they search traditionally or browse industry directories, it is secondary.
Are you already ranking on Google? If you are not on the first page, AI Overviews are usually a distraction. Our own July 2026 Search Console month makes the point: 311 impressions and 7 clicks for the whole site, with our largest single query at average position 96.2. Nothing was going to cite a page nobody could find.
Is your content strong enough to cite? If your pages carry no verifiable facts, mentions will be sparse no matter how well you track them. We found 10 checkable data points across our own nine articles in August 2026, with seven articles carrying none, and every article scoring between 22 and 47 against our own 55-point E-E-A-T threshold. That was the real problem, not the tracking.
For local services, B2B niches and early-stage businesses, AI visibility usually ranks below rank tracking and technical SEO. For expertise-driven fields and content-heavy brands, AI mentions are worth tracking because the audience is already asking generative engines for expert answers.
If you are not sure where you stand, the free audit covers your readiness for AI answers and your general search health, and the agent readiness check scores whether an engine can read your homepage at all.
Frequently asked questions
Does appearing in ChatGPT or Google AI answers actually drive traffic?
Sometimes. If the engine includes a clickable link and you are cited as a source, traffic can follow. But many AI answers paraphrase content without citing sources, and inline answers reduce click-through because users get their answer without leaving the tool. Raw visibility does not guarantee traffic.
How often should I track my AI visibility to distinguish real changes from noise?
Weekly tracking lets you see a month-over-month trend without overreacting to daily wobbles. One new mention in one week can look dramatic. The more important rule is to keep the question set fixed between runs, because a changed question set makes the next number unreadable.
Is AI visibility tracking worth doing for a local or niche business?
It depends on whether your customers use AI to search. A local plumber whose customers search 'near me' or use directories will usually see better return from rank tracking and Google Business Profile work. An expertise business whose customers type 'how do I' and 'what is' questions into an AI engine is more likely to benefit.
Should I prioritise improving my Google rankings or my AI visibility first?
Google rankings first, in most cases. AI systems often cite pages that already rank well. Our own site is the cautionary version: in the full month of July 2026 we took 311 impressions and 7 clicks, with our largest query at average position 96.2, and separately, across the 60 days to 25 August 2026, our 876 non-branded AI probes returned 0 citations.
Why does a Google-only AI visibility number mislead?
Because Google frequently declines to generate an AI answer at all. Across 7,494 Google AI Overviews probes we ran between 31 May and 25 August 2026, an AI Overview appeared on 3,400, or 45.4%. A summary score cannot tell you whether you lost the citation or the answer never existed.