How to check whether AI answers mention your brand

There is no position to track and no ranking report. Here is how to find out whether you appear, and what changes it.

A growing share of questions never reach a results page. Someone asks a model, gets an answer, acts on it. If you are not in that answer you were not considered, and unlike a ranking there is no position ten to console you. You are cited or you are not.

Check before you theorise

Most discussion of this subject is speculation. The useful first step takes ten minutes: ask the questions your customers would ask, in their words, and write down what comes back.

Two things to get right. Ask in a fresh conversation with no prior context, because a model that has been discussing your company will mention it. And ask several times — answers vary between runs, so a single check is a snapshot rather than a baseline. A mention on one run and silence on the next is normal, and averaging across runs is the only honest way to read it.

The AI answer checker automates the tedious part: the same question across four engines, run several times, with mentions and cited sources reported separately. That separation matters more than it sounds.

Want to check yours?

Run the AI answer checker →

Mentions and citations are different things

You can be named in an answer that links to a competitor. You can also be cited as a source without being recommended. These are different problems with different fixes, and a tool that collapses them into one number hides the thing you need to know.

Being mentioned but not cited usually means you are known but your pages are not the clearest statement of the thing. Being cited but not mentioned means your content is useful and your positioning is not landing. Neither is fixed by the same work.

What actually gets cited

Across the citations I have looked at, the pattern is consistent enough to plan around, and duller than most people hope.

  • Pages that answer the question directly, near the top, in a complete sentence. Models quote sentences. A page whose answer is spread across three paragraphs offers nothing convenient to lift.

  • Structure a machine can parse. Real headings, real lists, question-shaped subheads. Not schema tricks — just ordinary clear structure.

  • Specific, checkable claims. Numbers, dates, named methods. "Industry-leading approach" gives a model nothing.

  • Sources with existing standing. Which is the ordinary authority problem, unchanged.

Notably absent: keyword density, word count, and most of what on-page checklists still measure.

The fix is usually editing, not writing

When a competitor gets cited and you do not, the cause is rarely that they have content you lack. More often you both cover the topic and their version contains a liftable sentence while yours does not.

Find the page you would want cited. Read it as a machine would: is there one sentence that answers the question completely, on its own, out of context? If the answer depends on the paragraph above it for its subject, it cannot be quoted. Rewrite it so it stands alone and move it up the page.

That is often the whole intervention. It is unglamorous, takes twenty minutes, and does more than a rewrite.

What not to do

Ignore anyone offering to get you into AI answers through a technical trick. There is no markup that makes a model cite you, no file to upload, no submission process. The mechanism is that your page is a good source and the model found it.

Be sceptical of dashboards that re-report what a manual check tells you. Run the check yourself first and see whether the output would change a decision. Often it would not, and the honest answer is that you need a clearer page rather than a subscription.

Think carefully before blocking AI crawlers. Blocking removes you from what a model can cite. That may be right if your content is your product, but it is a business decision with a real cost on both sides — worth making deliberately rather than by copying someone's robots file.

How often, and what to watch

Monthly suits most sites. Weekly for a few weeks after changing something you want to see land, then back to monthly. Answers drift as models update, so what matters is direction over months rather than any single run.

Track three things: whether you are mentioned, whether you are cited, and what gets cited instead. The third is the most useful and the one most people skip. That list of competing pages is your work queue, and it is usually shorter than expected.

The AI visibility guide covers the strategic picture. This is the part you can do this afternoon.

A worked version of the check

Concretely: take the three questions that most reliably precede someone buying from you. Not brand questions — the ones asked before your name is known. "How much should X cost", "is it worth paying someone for X", "X versus doing it myself".

Ask each in a fresh conversation across the engines you care about. Record whether you appear, which sources are named, and what kind of page each source is. After three runs per question a pattern usually emerges: either the same handful of sources keeps appearing, in which case you know exactly what you are competing with, or the sources vary wildly, which tells you the topic has no settled authority and is unusually winnable.

That second case is the opportunity most people miss while worrying about the first.

What to do with a topic nobody owns

When citations are unstable, the question is under-served rather than locked up. A page that answers it plainly, with specifics, from someone who has done the work, tends to start appearing within weeks rather than the months an equivalent ranking effort would take.

The title and meta writer is useful here for a reason that has nothing to do with titles: writing five framings of the same page forces you to notice which one actually states the answer. That framing is usually the sentence worth putting first.

Get new Serpish things, when they’re ready.

Subscribe