Your brand keeps showing up in AI answers and no one is clicking

A small business owner checks ChatGPT, sees the company name in a recommendation, and feels good about it. Then the traffic report shows nothing. Nothing broke. The answer named the brand without linking to it, so there was no click to record.

That is the difference between a mention and a citation in AI search, and it decides what you can measure, what you can act on, and where you should spend the next month of content work.

Two kinds of visibility, one of them invisible

A mention is when an AI engine names your business, your product, or your idea inside the answer text with no link attached. The reader sees you. They cannot follow you anywhere, and your analytics never learns it happened.

A citation is when the engine attributes part of its answer to a specific page of yours: a numbered footnote, a source card, a link under the summary. The reader can click. You can see the session arrive.

Both are worth having, for different reasons. Mentions mean the model associates your business with a topic, which is the groundwork. Citations are the part that produces visitors, form fills, and something you can put in a report.

How the answer engines differ

Google’s AI Overviews sit at the top of the results page and show linked sources as cards. If your brand appears in the summary text but no card points to your domain, that is a mention.

Perplexity is built around sources and lists them next to nearly every claim, so the gap between being described and being cited is unusually easy to see. ChatGPT’s search mode attaches numbered references to specific statements. Microsoft Copilot pulls from Bing, which means its source pool can differ from Google’s for the same question.

One thing that has changed the calculation: the overlap between who gets cited in AI answers and who ranks in the traditional top ten appears to be shrinking. Ranking well still helps, but it no longer guarantees you a place in the answer. Treat AI visibility as its own thing to work on rather than a byproduct of SEO.

A tracking routine a two-person team can actually keep up

No analytics tool reads AI answers for you. The work is manual, and the trick is keeping it small enough that you still do it in week six.

Write down 20 to 50 questions that matter to your business: your name plus your category, the unbranded questions a buyer would ask before they know you exist, and the comparison questions where you would sit next to competitors. Freeze the list. Changing it every month destroys the trend line, which is the only useful output.

Run the set weekly across the engines you care about. For each question, log three things in a spreadsheet: did your name appear, was there a link to your domain, and which engine produced the answer. That gives you a mention rate and a citation rate.

Watch for the two rates pulling apart. Mentions climbing while citations stay flat usually means the engines know who you are but cannot find a page of yours clean enough to point at.

Log a few competitors in the same sitting. Same session, same engine state, so the comparison holds. A competitor cited on a cluster where you have nothing is your clearest work order: pull the page they cited and look at how it is built.

Catching the clicks you do earn

Clicks from cited links should land in Google Analytics as referral sessions, but a meaningful share get filed as unassigned or direct and quietly blend into your baseline. Build a custom channel group covering the AI domains you see in your referral report, so those sessions stay in one place instead of scattering.

If you use a CRM, the useful next step is a contact property recording which engine brought someone in the first time, set from the referring domain, so you can eventually ask whether AI-sourced contacts close differently. Treat whatever number you get as a floor, not a total. Some of it will still slip through.

One reason this matters more than the volume suggests: people arriving from an AI answer have already read a synthesized explanation and chosen to go deeper. That is a further-along visitor than a cold organic click, and it deserves a landing experience built for someone who has questions rather than one built for someone who has never heard of you.

Turning a mention into a citation

Answer first. If the direct answer to a section’s question sits in the third paragraph behind caveats, an engine looking for an extractable statement will take someone else’s. Lead each section with the claim and put the nuance underneath it.

Be consistent about who you are. Homepage, about page, product pages, third-party listings and profiles should describe your category and your customer in the same words. Inconsistent descriptions make it harder for a model to build a stable picture of your business.

Add structured data where it genuinely matches the page, and validate it. Markup that contradicts the visible content hurts more than it helps. If you are not comfortable editing your site’s code, this is a small, well-defined job to hand a developer.

Show who wrote it. Named authors with real credentials, a visible update date, outside sources, and examples that came from your own work all feed the same trust judgment. An unattributed post that has not been touched in years is fighting a losing battle no matter how sound the advice inside it is.

Then revisit. Rerun the numbers, and if a page that used to be cited stops being cited, someone published something more current. Plan a review every few months for the pages that earn you anything.

What this data cannot tell you

Two people asking the same question on the same day can see different sources. Location, personalization, and the engine’s own variation all move the result. Your weekly check is one sample, not the truth.

So do not react to a single week. Look for movement sustained across a month or more before you rewrite anything. And keep the claims you make internally proportionate to the instrumentation you have. Saying that citations on one query cluster rose after you restructured three pages is defensible. Attaching a pipeline dollar figure to AI search is, for most small teams, further than the data reaches.

Background reading: AEO mentions vs. citations: Key differences explained