What an AI Mention Is — and Why You're Looking in the Wrong Place

Your buyer is asking a chatbot the question they used to type into Google. They read one written answer instead of choosing from ten links, and that answer names a handful of companies. If yours is not one of them, the decision is often made before a single page loads.

So founders go and check. They open Google Analytics, then Search Console, then their rank tracker. All three come back empty, and the obvious conclusion is that AI has never heard of them.

That conclusion is usually wrong. It is also common enough to have its own genre of thread:

That question is asked by people who already run analytics, already have Search Console, and already know their Google rankings. They are not missing a tool. They are missing a definition.

This piece answers three things: what an AI mention actually is, why every tool you already own reports zero, and what you can genuinely check today without paying anyone.

What Actually Counts as an AI Mention?

An AI mention is when a generated answer names your brand. That is the whole test. A clickable source is a bonus, not a requirement.

That one sentence undoes most of what search taught us. In classic search, being visible and being clicked were the same event — you appeared as a link, and the click was the proof. Inside a generated answer those two come apart completely. The engine can describe your product, name you as the option worth considering, and never show a URL. The buyer reads your name, forms a view, and moves on.

There are three states, and holding them apart is the whole job:

• Named and cited — your brand appears in the answer and as a clickable source. The only state your analytics has any chance of seeing.
• Named, not cited — the answer says your name with no source attached. Real influence, zero trace.
• Absent — the answer covers your category and you are simply not in it.

The middle state is the one that decides deals, and the one nobody instruments. It produces no click, no session, and no row in any report you currently open.

Why All Three of Your Tools Report Zero

Each tool is built around a click or a query. A mention with no source attached produces neither, so as far as those tools are concerned it did not happen.

Take them one at a time.

Google Analytics records a visit. If a chatbot named your brand and attached no link, nobody clicked, so there was no visit for GA4 to record. Referral traffic from AI hosts is real and worth reading — but it can only ever count the mentions that carried a link and got clicked. It is a floor, never a total.

Search Console reports Google's own surfaces. A Perplexity answer, a Claude answer or a ChatGPT answer never passes through Google's systems, so Search Console has nothing to say about them.

A rank tracker reports your position in a list of links. A generated answer is not a list. There is no first, no third, no page two — so there is no position to report.

None of these tools is broken or badly configured. They are measuring accurately. They were built to watch a click, a query and a rank — and a mention is none of those things.

Which is why so many teams end up doing it by hand:

The answer, largely, is everyone.

Where AI Mentions Actually Happen

Mentions happen across at least five surfaces, and they behave differently enough that "am I visible in AI" is not one question.

Two things follow from that table.

Strong on one is not strong on all of them. The engines retrieve and reason differently, and practitioners compare them constantly:

That thread recurs precisely because the answers disagree with each other.

They are not equally observable. Some attach visible sources you can read straight off the screen. Others name a brand in prose and stop. An honest measurement plan starts by saying which surfaces you can actually see into, and which you are guessing at.

Does Ranking #1 on Google Mean AI Mentions You?

No. It helps, and it is not the same thing.

We put a 115-question interrogation to eight AI systems — ChatGPT, Claude, Gemini, Grok, DeepSeek, Kimi, Perplexity and Copilot — asking each to narrate its own path from a user's question to a cited URL. Eight out of eight said search position does not determine citation. Eight out of eight also said a retrieved page is not the same as a cited one: being fetched proves nothing on its own.

That result is worth reading in both directions.

It is the strongest agreement in the entire set, and it cuts against the industry's own marketing. A model reciting SEO content would have told us authority and rank drive citations. All eight said the opposite.

It is also eight self-reports, not eight measurements. A model has no privileged view of the retrieval system it sits inside. Treat it as a strong hypothesis, not a proven law — and be wary of anyone who quotes a finding like this without saying so.

Practitioners arrived at the same place by a different route:

What You Can Genuinely Check Today, for Free

You are not stuck waiting for a budget. Three checks cost nothing and can be run this afternoon.

Ask your own buying question, several times

Take a real question a buyer would ask before choosing a vendor in your category. Put it to two or three engines. Then ask it again. The answers move between runs, which is itself the finding — a single run tells you almost nothing, and a rate across ten runs tells you a lot.

Read who is named

When your brand is absent, the answer still names somebody. That list is your real competitive set inside the answer, and it is frequently not the list you would have written yourself.

Check your referral data for the AI hosts

Traffic from these tools does appear in GA4 when a source was attached and clicked. The data sits a few clicks deep, which is why most small teams never look — but it is there. Read it as the floor of your visibility, never the total.

Run all three and you will have something no dashboard has given you: a dated, first-hand reading of whether you are in the answer.

What Nobody Can Honestly Tell You Yet

There is no trustworthy public benchmark for a "good" AI mention rate, and you should be careful with anyone who offers you one.

We know this because we audited our own corpus first. Across eight AI systems narrating their own retrieval, we found fourteen precise-sounding statistics asserted with no source behind any of them. One system fabricated its citations outright — an invented URL, invented passage text quoted from it, an invented citation format — inside the document that read as the most rigorous in the set. Confidence of presentation told us nothing about whether anything real sat underneath.

So the honest position is short:

• You can measure your own rate. Ask the same question ten times, count how often you appear, write down the date. That number is real because you generated it.
• You cannot benchmark it against the market. No credible public baseline exists yet.
• A vendor's benchmark is a vendor's sample, usually collected in a way that makes their product look necessary.

Measure yourself against yourself, with dates, and you are already ahead of a market still arguing about whether the thing can be measured at all.

Conclusion

An AI mention is your name inside somebody else's answer. Not a link, not a rank, not a visit — which is precisely why the tools you already pay for report nothing. The fix is not another dashboard. It is asking the real buying question, several times, on more than one engine, and writing down what you see with a date beside it.