AI Visibility: Why Tracking One Platform Gives You the Wrong Number

Count every time the review site G2 gets quoted by an AI assistant. Only 11.3% of those quotes come from ChatGPT. The rest come from assistants most people never check.

First, a definition, because everything here rests on it. A citation is what you get when an assistant quotes your page and shows a link to it in the answer. It is the closest thing AI search has to a ranking, and it is what every AI visibility tool counts.

Now look at what those counts actually show. Of all the citations G2 receives across six major assistants, ChatGPT accounts for 11.3%, and Perplexity for more than three times that. Wikipedia gets more of its citations from Perplexity than from anywhere else, at 37.5%. YouTube gets 41.7% of its from Google's AI Overviews.

In other words, each assistant quotes a different corner of the web. That is the central problem with how AI visibility is measured today.

Most AI visibility tools track ChatGPT and report the result as your score. Two independent measurements say that is no longer a reasonable shortcut. It is a measurement error, and this piece shows why and what to do instead.

The market stopped being one product

Between June 2025 and May 2026, ChatGPT's share of global assistant web traffic fell from 76.4% to 52.7%. That is 23.7 points, a 31.0% relative decline, and the steepest fall of any platform tracked.

Read that carefully. It does not mean ChatGPT is shrinking. The category grew fast enough that it can lose 23.7 points of share and still gain users.

What it means is the shape changed. Two platforms now hold 80.0% of tracked share between them, and several platforms people actually use are not separately tracked at all. A year earlier, one platform held three quarters on its own.

So a tool reporting ChatGPT visibility is reporting on a bit over half a market, and describing it as the whole thing.

The platforms disagree about sources, sharply

That would be manageable if the platforms cited roughly the same pages. They do not.

Each one reaches for its own ecosystem first: its own index, its own partnerships, its own idea of a trustworthy source.

Look again at the G2 figure. If G2 judged its AI visibility on ChatGPT alone, it would be watching about one ninth of the times it actually gets quoted, and missing the other eight ninths completely.

That is the whole argument. The disagreement between platforms is not noise to be averaged away. It is the most informative thing in the data.

And the list of surfaces keeps moving

Even a multi-platform tool works from a fixed list, and the list ages faster than the tooling updates.

Grok now reads X posts and mentions in real time, which turns social content into a retrieval surface an answer can quote. That was not true a few months ago.

Any audit fixed to six platforms is out of date by construction. The honest thing for a vendor to publish next to a visibility score is the date the surface list was last revised.

Five rules for a defensible AI visibility report

None of these need new software.

1. Report per platform, never blended. The assistants disagree far too much to average. On one site in the probe, the six ranged from 1.2% to 54.2% of its citations depending on which assistant you asked. A single blended score hides exactly that. Show the rows.

2. State the surface list and its date. Which platforms, pulled when. If a platform is missing, say so, rather than letting absence read as zero.

3. Separate citation from traffic. These are different measurements and the gap is enormous. In Pew Research Center's study of real browsing behaviour, across 68,879 searches by 900 US adults, clicks on the sources inside an AI summary ran at 1%. Being cited is brand exposure. It is worth having. It is not a traffic channel. Why citation share is not traffic covers the three architectures behind that gap.

4. Count how often each page is reused, not just your total. Take the pages of yours that get cited at all, then ask how many citations each one earns. Wikipedia averages 8.11 citations per cited page. YouTube averages 1.87. In other words, a Wikipedia page gets quoted again and again, while a YouTube page is usually quoted once and forgotten.

That distinction matters because a total rewards publishing more pages, while the average per page rewards publishing better ones. A report showing only the total will push you towards volume.

5. Track exposure and use separately. Google's AI Mode reports a billion monthly users. In panel data, 0.34% of searches actually moved into it. Most AI adoption statistics live in that gap.

The adoption numbers underneath the sales pitch

Visibility tooling usually arrives with a slide of AI adoption statistics, so it is worth knowing what those statistics are made of.

Across the products in the dataset carrying a headline user number, 58.3% of those numbers describe people who were given an assistant, sold it inside a bundle, or counted through a product they already used. Not people who chose one. Some 41.2% of the stacked headline total sits on products nobody actively picked.

And of 15 adoption figures in the workbook, two are audited.

That does not make AI adoption fake. It means the headline user counts in circulation are mostly unaudited counts of exposure. A visibility strategy built on which platform reports the biggest user base is built on the least reliable figure available.

There is a further wrinkle worth planning for: increasingly, the visitor on the other end is not a person. When an agent browses instead of a human, dwell time and engagement signals collapse, which breaks the measurement stack before it breaks the funnel.

The referral number, so the picture is complete

Visibility work is usually justified by the promise of traffic, so here is the traffic, stated accurately in both directions.

AI assistants now send roughly 771 million referral visits a month, growing 117.4% year on year, and those visits convert about 54% better than non-AI sources for US retail.

It is also a low single-digit share of most sites' total traffic. The reason is arithmetic, not strategy: only 6.8% of assistant conversations include a web citation at all. A citation is a precondition for a referral, and most conversations never produce one.

So the honest framing is neither "this replaces organic" nor "this is a rounding error". It is a new channel with different economics: about a fiftieth of the volume, materially better conversion, growing fast enough that today's size is the wrong thing to judge it on.

How to improve AI visibility this quarter

Good AI visibility work starts with one table, not one score.

Pull citation counts for your own domain and your three closest competitors, across every platform you can reach, on a single date. Record the date. Put them in a table with one row per platform, and resist the urge to produce a single number from it.

Then look at where you are absent, not where you are present. Absence on a platform that holds real share of your buyers' attention is a specific, fixable problem, and it is usually a format problem – because format beats reputation in citation data. Presence on one platform and absence on four is the most common finding, and it is completely invisible to a tool watching only the first.

The category name for this work will keep changing. The discipline underneath it will not: name your surfaces, date your pull, separate citation from traffic, and never publish an average of things that disagree.

Questions people ask about AI visibility

These are the questions searched most often alongside AI visibility, answered from the citation and traffic data above.

What is AI visibility? It is how often, and how prominently, your brand or pages appear inside AI-generated answers. It is measured in citations rather than rankings, because generated answers do not publish positions.

How do I monitor AI search visibility? Pull citation counts per platform on a fixed date, repeat monthly, and report the platforms separately. Any single blended score hides the disagreement between platforms, which is the most useful signal available.

Why should I track AI brand visibility? Because buyers increasingly form shortlists inside an assistant rather than on a results page, and which assistant they use changes which brands appear. Track it as brand presence, not as a traffic forecast.

Are AI visibility tools worth buying? They are useful for measurement at scale, but check two things before you buy: how many assistants it covers, and whether it shows how often each page is reused rather than only your total citations. Without both, you get a number that rewards publishing more pages instead of better ones.

How do I improve brand visibility in AI search? Publish checkable facts, including prices where you can. Build canonical pages a model can return to repeatedly rather than campaign pages it quotes once. Then measure across platforms and fix the ones where you are absent.

---

Traffic share: assistant traffic share column of a 261-release workbook, June 2025 to May 2026. Citation figures: live Ahrefs index pull, five domains across six answer platforms, 7 September 2026. Source-click rate and AI Mode entry rate: Pew Research Center and SparkToro on Similarweb panel data. Full method in the report, How AI Changed the Way People Search.