How to check whether Google AI Overviews knows about your company

Search your buyers' questions in ordinary Google, note whether an overview appeared, whether you were named in it, and which sources it linked. Three separate things, and recording only the second one is why most people cannot explain their own results.

Why "did an overview appear" is the first question

Overviews do not show for every query. If no overview appears for your most valuable question, then there is nothing to be named in, and your effort belongs in the ordinary results for now. People spend months trying to appear in a summary that is not being generated.

So keep three columns: overview shown, you named, sources linked. Over several weeks the first column tells you where this even matters, and it is often the most surprising output of the whole exercise.

Why you have to do this by hand

Google says AI Overview performance is included inside your overall Search performance reporting rather than reported separately. So there is no report that tells you how often you appeared in an overview. Whatever tool you use, somebody or something has to search the question and look at the screen. That is not a gap in your setup, it is the current state of the feature, and it is worth knowing before you go looking for a dashboard that does not exist.

How to ask

Use the words a buyer would type, and do not include your company name. A search for your own brand will show your own pages and tells you nothing about whether you would be recommended.

Check from a signed out browser and, where relevant, for the location your buyers are in. Results differ by place, and a local question answered from the wrong place is not your answer. If you sell in more than one country or language, that is more than one set of readings, and combining them into one number hides the only thing you needed to know.

The six things to record for each question

• Whether an overview appeared, and roughly how long it was.
• Which companies were named, and in what order.
• Every source it linked, by full URL.
• Whether your page ranked in the ordinary results underneath, which is a separate and still valuable position.
• Where you searched from, meaning the location, language and whether you were signed out.
• The date. Everything else is meaningless without it.

A worked example

Here is a real reading, taken blind on our own site. In September 2026 we asked Perplexity six questions about our own category, with no brand named in any of them, for aiknowsus.com. We then asked it to audit its own answers and say where we had appeared. It reported that it had not cited or recommended aiknowsus.com in any of the six.

The bad way to read a result like that is "we are nowhere, we are failing". The useful way has three parts. It is a starting line, and a starting line is the only thing a later reading can honestly be compared with. It was taken blind, so it reads the market rather than our own name. And because nothing in it can be turned into a percentage, the conversation moves straight to the sources, which is where the work is.

The same capture shows what an engine will not do for you. In batch after batch, asked which competitor came out ahead most often, Perplexity declined to name one, saying its own previous answers had not produced a comparable live tested result that would support such a claim. If the engine will not draw a ranking out of one run of its own answers, neither should you.

Apply both of those to a run of twelve questions here. Take out the questions where no overview appeared at all, because those are ordinary search work and do not belong in this measurement. Among the ones that did generate an overview, being linked under some of them means the mechanics work: you are indexed, relevant and readable. Being named under one and not another means your page had something liftable on one question and not on the other, so compare those two pages and find the difference. And ranking in the ordinary results underneath is real traffic that no overview took away, so keep it in its own column.

Then re-run the same twelve a month later and expect the number of overviews to move on its own. Nothing went wrong. That column moves without you, which is exactly why you record it.

The brand accuracy check, kept separate

Search your company name too, but keep those results in a different list. You are looking for wrong prices, confusion with a similarly named company, and out of date descriptions. Fix those first, because an assistant repeating a wrong price is doing damage while being unnamed is only a missed opportunity.

The check people skip, and it is the cheapest one

Confirm the page is actually indexed. A surprising share of pages that businesses are trying to get summarised have never been indexed at all, and no amount of rewriting will help until that is fixed. Check it directly rather than assuming, and check that the page shows its content without scripts running, because a page that looks full in your browser can arrive nearly empty.

What this check cannot tell you

It cannot tell you what other people see. Overviews vary by country, language, device, account and over time, so your reading is your reading.

It cannot be turned into a rate from a handful of searches on one day. A rate needs a larger set and repeated readings, and it still needs a range around it.

It cannot tell you why an overview was not generated, because that decision is Google's and is not published.

And it cannot promise anything. No one can guarantee a place in an overview, and a page that appeared last month can be absent this month without anything changing on your side.

How often

Once a month for the full set is plenty. Overviews come and go for the same query, so a weekly check mostly records that variation rather than your progress. Keep every reading rather than overwriting one number, because six monthly readings is evidence and two weekly ones are not.

Common questions

Why do I see an overview and my colleague does not?

Because generation is decided per query and varies with location, language, device and account, and the feature itself changes. This is normal. Record where each reading was taken and stop treating a disagreement as an error.

Can I check this without doing it by hand?

Partly. Any approach still comes down to issuing the search and reading the result, since the data is not broken out in reporting. Whether you do it yourself or use a tool, keep a handful of manual readings as a sanity check, and record the location every time.

An overview named a competitor and cited a page about us. Why not name us?

Because the summary took the names from the pages it summarised and your page supplied a fact rather than a name in a candidate list. This is the clearest possible sign that you need to be in the roundups and listings that rank for that question.

Is being linked the same as being recommended?

No. Three states matter here: linked, named, and ranking underneath. Linked means the mechanics work. Named means the buyer read your name. Ranking underneath still sends visits even when neither of the others happens.

Should I count my ordinary ranking in this measurement?

Keep it in a separate column and never merge it. They are different outcomes with different value, and a team that merges them can report progress while the thing they actually wanted got worse.

How many questions should I track?

Start with ten to fifteen that you can actually run every month by hand. A set you keep is worth far more than a larger set you abandon after one reading, and the same set over six months is what makes the numbers mean anything.

What to do first

Open a sheet, put your twelve most valuable buyer questions in it, and add the six columns above. Then spend one hour searching them from a signed out browser and filling in the first column only: did an overview appear. That single column will tell you which questions this page is even about, and it will probably remove a third of your list.