126 million prompts later: the mention-vs-citation gap in AI search

Semrush just analysed 126 million AI-search prompts - one of the largest looks yet at how brands really show up in AI answers. Buried in the data is a distinction most teams collapse into one: being mentioned and being cited are not the same thing, and the gap between them is bigger than almost anyone assumes.

When a marketer says "are we showing up in AI," they usually mean one thing but measure another. The Semrush 2026 AI Visibility Index makes the confusion impossible to ignore, because at 126 million prompts it has the scale to show that the two things people conflate genuinely diverge.

What the study looked at

Published on 26 June 2026, the index analysed 126 million US AI-search prompts from January to April 2026 across ChatGPT, Gemini, Google AI Mode and Google AI Overviews, benchmarking 22 industries. It scaled up an earlier, much smaller sample into something big enough to see patterns that anecdote misses. Two of those patterns are worth building a strategy around.

Pattern one: mentioned is not cited

The headline finding for practitioners: the overlap between the brands an engine mentions and the domains it cites can be as low as around 30% on Gemini. Read that slowly. A brand can be named in the answer without being credited as a source, and a source can be cited without being the brand the answer names. They are two different currencies.

This matters because most measurement tracks one and calls it visibility. If you only count mentions, you miss whether the answer actually points anyone to you. If you only count citations, you miss whether you are the brand being recommended by name. Optimising for one while ignoring the other leaves half your picture dark.

"A brand can be named in the answer without being credited, or credited without being the one named. They are two different currencies."

Pattern two: almost no one is universally visible

The study identified a "universal" set of just 36 brands that held top-100 visibility across all four platforms every month of the analysis - names like YouTube, Google, Reddit, Amazon, Apple and Walmart. Thirty-six. Across four platforms. Every month. The implication is that durable, cross-platform visibility is rare and hard, and that the engines disagree with each other enough that winning on one is no guarantee of winning on another. There is no single "AI visibility" number; there are four, and they move independently.

The measurement gap is the real story

The most actionable numbers in the report are about measurement itself. Semrush found that 45% of marketing leaders cannot accurately measure their brand's visibility in AI answers, and only 9% have tools that track across all the major platforms. Put those beside the two patterns above and the problem is obvious: the thing is genuinely multi-dimensional - mention versus citation, across four disagreeing engines, changing monthly - and most teams are trying to read it with a flashlight.

There is a payoff signal too. Organisations with a fully unified SEO and AI-visibility workflow reported increased traffic or leads 81% of the time, versus 36% for teams managing the two separately. The teams treating AI visibility as a measured, integrated discipline are pulling ahead of the teams treating it as an occasional curiosity.

What to do with this

1. Track mention and citation separately. Ask, for each priority question: are we named, and are we cited? They are different, and closing one gap is a different job than closing the other. A brand mentioned-but-not-cited needs better sourcing; a brand cited-but-not-named needs stronger entity presence.
2. Measure per engine, not in aggregate. With only 36 brands universally visible, assume the engines disagree about you too. Know where you win and where you do not, platform by platform.
3. Make it continuous. Monthly change is baked into the data. A one-time audit is a snapshot of a moving target. Occasional manual checks are why 45% of leaders cannot measure this - the cadence is the problem.
4. Unify the workflow. The 81%-versus-36% gap is the clearest ROI signal in the report. Treat SEO and AI visibility as one integrated practice, not two teams that meet occasionally.

Why the two currencies diverge

It helps to understand why mention and citation come apart, because the reason points to the fix. A mention comes from what the model knows and decides to say - it names the brands it associates with the category, drawn from everything it has absorbed. A citation comes from what the model retrieves and links in the moment - the specific pages it pulls to support the answer. Those are two different mechanisms. You can be strongly associated with a category (mentioned) without having a single clean, retrievable page the model wants to cite. And a thorough page can get cited as a source without your brand being the one the model names as the recommendation.

The fix follows from the mechanism. To earn more mentions, you strengthen the entity - be described, consistently and by many sources, as a leading option in your category, so the model's associations include you. To earn more citations, you give the model something worth retrieving - clear, specific, current pages that answer the question cleanly enough to quote. They are related jobs, but doing one does not automatically do the other, which is exactly why measuring them separately is the starting point.

A simple way to run this yourself

You do not need 126 million prompts to apply the lesson. You need your own top questions, run honestly:

1. Write your 15-25 real buyer questions - the ones a prospect would actually type, in their words, not yours.
2. Run them on each engine that matters to your buyers - typically ChatGPT, Gemini, Perplexity and Google's AI surfaces - and log two columns per question: were you named, and were you cited.
3. Read the gaps. Named-but-not-cited points at a sourcing problem (give the model a page worth citing). Cited-but-not-named points at an entity problem (get described as a category leader more widely). Missing from both is where competitors are winning.
4. Repeat on a schedule, because the data shows it moves month to month. A snapshot lies; a trend tells the truth.

Done by hand this is a real chore across four engines and two dozen questions every month, which is precisely why 45% of leaders cannot keep it up - and precisely the case for tracking it continuously rather than heroically.

The takeaway

The value of a 126-million-prompt study is that it turns hunches into structure. The structure here is simple and a little uncomfortable: AI visibility is not one number, it is mention-and-citation across four engines that disagree and change monthly - and most teams cannot see it clearly. The ones who can, and who treat it as one measured discipline, are already turning that clarity into traffic and leads. The first step is the honest one: stop asking "are we in AI" and start measuring, separately and continuously, whether you are mentioned, cited, or missing.

Related: Brand mentions beat backlinks for AI citations - the 2026 data →