How to check whether Perplexity knows about your company

There are two separate questions, and you need both: does Perplexity name you when somebody asks about your category, and does it describe you correctly when somebody asks about you by name. The first is your visibility. The second is your accuracy. They are measured differently and must never be added together.

Why this engine is the easiest one to check

Perplexity shows numbered citations beside almost every answer. That means a check here gives you three things at once: whether you were named, which pages decided the answer, and therefore what to do next. On engines that answer from memory you get only the first of those, and you are left guessing at the cause.

It also reacts quickly, because it retrieves at the time of the question rather than recalling. So this is the place to test whether a page you published last week is doing anything at all.

Check one: are you named at all

Write down ten questions your buyers actually ask, in their words. Not your product name, not your category as you describe it internally. The question a person types when they have not decided yet.

Ask each one in a fresh conversation. Never put your company name in the question. A question that names you gets an answer about you, which tells you nothing about whether you would have come up.

For each question, write down whether you were named, who was named, in what order, and which sources were cited. The sources matter as much as the names.

Check two: is what it says about you right

Now ask about you directly, in a separate conversation, and keep these answers in a separate list. Ask what your company does, what it costs, who it is for, and whether there are any concerns about it.

Look for three failures. Wrong pricing, which is the most common and usually comes from an old page or a marketplace listing you forgot about. Confusion with another company of a similar name. And out of date facts, such as a product you retired or a location you left.

Read the concerns section carefully rather than defensively. It is the closest thing you will get to hearing what a buyer is told about you when you are not in the room, and it is often accurate.

The seven fields to record per run

One row per question per run, with the following seven fields. This is the whole method and it fits in a spreadsheet.

• Date. Non negotiable, because your only real output is change over time.
• The exact question, copied.
• Mode used, if you switched between a quick answer and a deeper research mode. They read different amounts and are not comparable.
• Named, yes or no, and in what position.
• Who else was named, in order.
• Every citation URL. Full URLs, not domains.
• Whether any of your own pages were cited, which is a separate result from being named.

A worked example

Here is our own reading, taken on this engine. In September 2026 we asked Perplexity six questions about our own category, with no brand named in any of them, for aiknowsus.com, which is our own site. Afterwards we 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. There was no position for us to hold.

That reading is uncomfortable and it is far more useful than a flattering one. Nothing in it can be turned into a percentage and argued about, so the conversation moves straight to the sources. It was taken blind, so it is a reading of the market rather than of our own name. And it is a starting line, which is the only honest thing to compare a later reading against.

The same capture shows what happens when you push an engine for a ranking it cannot support. 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. That is the clearest statement we have that a single run does not establish a ranking, and it came from the engine itself.

So read your own ten questions the way that run has to be read. Being named in two of ten is two out of ten, not twenty per cent. If your own pages are cited on questions where you are not named, the mechanics are fine and the writing is the problem, so read those pages as a stranger and ask which sentence an engine could have repeated. Usually there is not one. And if a single directory keeps turning up across your questions, getting that listing complete and current outweighs anything you could write this month.

Then re-run the same ten a week later and expect the count to move on its own. Nothing broke. You have learned that the variation between runs is about the same size as the result you are measuring, which is exactly why you keep readings rather than a number.

Reading the result honestly

Ten questions is a small sample. If you were named twice, the honest statement is not that you are at twenty per cent. It is that you were named twice out of ten and you need many more questions before a rate means anything. Ask the same set again a week later and expect the number to move on its own.

Never combine the blind set and the named set into one figure. The named set will nearly always mention you, so including it produces a number that can only look good and that nobody outside your company would believe.

What this check cannot tell you

It cannot tell you why a page was not cited, only that it was not. A page can be fetched, read and left out of the citation list, so absence is weaker evidence than presence.

It cannot be compared across modes. A deep research reading and a quick answer reading are different instruments, and a change between them is a change of instrument rather than of position.

It cannot tell you what buyers actually see, because your questions are your guess at their questions. That guess improves when you take the wording from real sales calls and support tickets instead of from your own marketing.

And it cannot promise anything. No one can guarantee a place in an AI answer, and a check that looks good one week can look worse the next for reasons that have nothing to do with you.

Common questions

How often should I run this?

Your top three questions weekly, the full set monthly. This engine is fast enough that weekly readings carry real information, which is not true everywhere. Keep all the readings rather than overwriting a single number.

Why did two runs of the same question disagree?

Because each run searches again and can retrieve a different set of pages. That variation is the engine working normally. Take three readings before you believe any single result, and report a range rather than a point.

My competitor is cited for a fact about my own company. What do I do?

Open their page. If the fact is wrong, write to them with the correct value and a link to your own page stating it, and keep a dated copy. If the fact is right, the problem is that they stated it and you did not. Publish it on a page built for that question, in text, with a date.

Is being cited the same as being recommended?

No, and this check is where the difference becomes obvious. You will see questions where your page is citation two and the recommendation names two other companies. Track them separately: citations tell you the mechanics work, mentions tell you the content persuaded.

Should I check in incognito or signed in?

Signed out or a separate profile for the blind set, so that you are closer to what a stranger sees. If you must use your own account, at least start a fresh conversation for every question so that earlier answers do not shape later ones.

How many questions before a rate means something?

More than most people have the patience for. Ten questions tells you which of your pages are reachable and which sources matter, which is useful on its own. A rate you could report to a board needs a much larger set and repeated readings, and it still needs a range around it.

What to do first

Do check two before check one. Ask Perplexity what your company does and what it costs, today, in one conversation. If it is telling buyers a wrong price, that is active damage and it is usually fixable in an afternoon, while being unnamed is only a missed opportunity. Fix the wrong fact, then start the blind set.