AI Visibility Tools: How to Check Whether AI Assistants Recommend Your Brand

Your buyers ask an AI assistant who to pick before they ever open Google. Want to know whether you're in that answer? You can run a free check on your own domain. This piece is about what to look at, what to skip, and how to measure it without fooling yourself. It first ran as a LinkedIn article. The version here adds the part people asked about in the comments: how to tell a real change from everyday noise.

Buyers ask the assistant first

Someone needs a vendor or a tool. More and more of them type that straight into ChatGPT, Gemini, Perplexity or Claude and take the shortlist it hands back. If you're not on it, you haven't lost a click. You never made it into the room.

Classic SEO tells you where your pages rank. It doesn't tell you whether an assistant names you when a buyer asks. That's a separate signal, and most teams have never looked at it.

What “AI visibility” really means

Two things, and people keep mixing them up. Does the assistant name your brand in the answer at all? And does it cite your own site as a source, or someone else's? You can get named all the time and still have none of your own pages cited. The model is describing you with what other people wrote about you.

That gap is bigger than most marketers expect. In 2026 Dmitrij Żatuchin analysed 167,551 citations behind AI replies about 128 brands in 12 markets and 13 languages. 85.7% pointed to pages the brands didn't own: reviews, forums, press, comparison sites (arXiv:2606.25787). Those were reputation questions, and the split moves with the kind of question, but the direction holds. Showing up in AI is largely about what the rest of the web says about you, not what you publish yourself.

Why one score isn't enough

Put the same buying query to the four big models and you'll often get four different shortlists. One names you, another leaves you out, same query, same day. A single blended “AI visibility score” hides the very gaps that cost you the recommendation. You want it split per assistant, with named and cited kept apart.

How to check it by hand

You can do this yourself in an afternoon. The method matters more than the tooling.

1. Write down the questions your buyers ask, five to ten of them, in their words, from before they know you exist. Leave your brand name out. A query with your name in it only tests whether the model has heard of you, which is a different thing (recognition isn't recommendation).
2. Run each one through ChatGPT, Gemini, Perplexity and Claude, with web search on wherever the assistant offers it.
3. Note three things per answer: were you named, was your site cited, and who got recommended instead.
4. Freeze the list and repeat it on three different days. One pass is a snapshot. Three passes start to show what holds.
5. Read the sources, not just the names. The pages the assistants quoted are the places that speak for your category. If you're not on them, that's your to-do list.

It works. It's also slow, and it drifts, because the answers move from one run to the next. That drift is why tools exist.

What an AI visibility tool should do

A good one runs your buyer prompts across the assistants on a schedule and keeps the list frozen, so a comparison over time means something. It tracks named and cited as two separate signals, per engine. It flags the rival that shows up in your place and shows which pages the answer leaned on. And it tells you how much of a change is just the engines moving on their own. If you're weighing vendors, we keep an honest comparison of the monitoring tools, ours included.

How to tell a real change from everyday noise

This was the question under the LinkedIn post, and it's the right one. AI answers aren't stable. Repeat a prompt and a share of the verdicts flips. The set of cited pages turns over day to day as well; the figures are in how much AI answers change day to day. So a before-and-after built on one scan is a guess.

We handle it the way a lab would, with a control. Since September we've run one frozen question set on our own domain, four engines, three repeats a day, and only that one domain. Nothing about that site changes between runs, so whatever moves is the engines' own wobble. The median day-to-day movement per engine is the noise band. A client's move counts as a change once it clears the band and holds across three daily reads. Inside the band we call it noise. The dashboard shows the band next to every move, so you can check the label yourself. It's one domain, so it's a yardstick rather than a market study. While an engine has too little panel data we say so and fall back to a default band. The full method is public: how every number is computed.

What to ignore

A lot of “AI optimization” advice is selling a rewrite: change your pages so the models quote you more. Be careful with it. A 2026 survey went through 45 separate studies and found no repeatable, cross-platform effect from that kind of rewriting (Martinez, arXiv:2607.14035). Same tactic, different model, different result. In one end-to-end test inside that survey, rewriting page bodies to sound more quotable made them less likely to be retrieved, not more.

Two more things to drop. A single blended score with no per-engine split, because it hides where you're losing. And the idea that a file like llms.txt moves citations. It's agent-readiness hygiene, worth having, and not a lever (what llms.txt is and who reads it).

Two dull things do seem to help. Be worth quoting, in plain language a model can lift. And show up on the pages the model already trusts, which, as the citation study shows, are mostly not yours. The rest tends to be noise that changes run to run.

Where to start

If you've never seen which sources speak for your brand inside AI, start there. Run your top buyer questions. Check whether you're named and cited, and see who gets recommended instead. Most brands haven't looked yet, so the ones who look first get a real head start.

The discussion

This piece first went out on LinkedIn: the full article, and the post where the conversation is happening. Read them and weigh in there.

Related reading: does AI recommend your business, or just recognize it · how to check if AI cites your brand · how much AI answers change day to day · you can't buy AI citations. See also: daily AI visibility monitoring · the free check · how every number is computed.