How to check your own AI visibility in 30 minutes

You can do the first pass yourself

None of this replaces a proper, repeated audit — it can't, for reasons that show up clearly by the end of this playbook — but it will tell you, honestly and in your own words, whether AI visibility is a real problem for your brand before you spend anything finding out more precisely.

Before spending anything, you can run a rough version of what an AI visibility check actually does, by hand, in about half an hour. You need four browser tabs — ChatGPT, Gemini, Perplexity and one more search-capable assistant — and somewhere to write down what you see. Nothing here requires a technical background; it requires patience and a willingness to read your own brand's answers critically.

Step 1: write eight category-first questions

The mistake almost everyone makes here is testing their own brand name first. That question is the easiest one to win — an assistant that already half-remembers you will usually confirm it, and it will tell you almost nothing about the buyers who don't yet know you exist. The questions that matter are the ones a buyer types when they're discovering the category, not confirming a name they already have. Write them the way a stranger to your brand would type them, not the way your own marketing describes the category.

1. 01Two discovery questions: “best [category]” and “recommended [category] for [use case]”.
2. 02Two problem or use-case questions: “best [category] for [specific need]” and “how should I solve [the problem your category solves]”.
3. 03One trust question: “most trusted [category]” or “most reliable [category]”.
4. 04One comparison question: “[your brand] vs [your closest competitor]”.
5. 05One brand question: “what does [your brand] do” — this is the only one that names you.
6. 06One question specific to your category (a location, a price band, a use case only you and a few rivals actually compete on).

That mix isn't arbitrary. Discovery and problem-based questions are where most of the buying journey actually happens, before anyone has a shortlist of names in mind. The single brand question is there for comparison, not because it's the important one — it's the easy control question that tells you whether the assistant knows you exist at all before you go looking at the harder ones.

Step 2: ask each one in four assistants

Run each of the eight questions once in each of the four assistants — eight questions across four engines gives you thirty-two short conversations. That isn't an arbitrary number: it's the same shape a free automated Snapshot runs, just without the repetition or the classification done for you.

Keep each answer as it came, including the parts that seem irrelevant. An assistant's hedging — “it depends on what you're looking for” — or its refusal to commit to a single name is itself useful information about how confidently it can speak about your category at all.

Ask the same question twice in the same sitting and you may get two different shortlists — that isn't a bug, it's how these systems answer.

Step 3: log three things for every answer

1. 01Who got named, and in what order they were named.
2. 02What status you got: absent, mentioned in passing, considered as one option, actively recommended, or presented as the buyer's first choice.
3. 03Which sources the assistant seems to be leaning on — it will often say, or you can directly ask it what it's basing the answer on.

That third point is worth the extra minute it takes. Asking an assistant directly what it's basing an answer on will sometimes surface the actual source — a review platform, a comparison site, a specific publication — and that's the single most useful piece of information in this whole exercise, because it tells you exactly where to go and try to change something.

It's also worth noting, for each answer, roughly what kind of source seems to be behind it: your own site, a competitor's, a review platform, a publication, a blog, a community thread, or something social or video-based. You won't always be able to tell precisely, but even a rough category — “this reads like it came from a review site” — starts to show you which kinds of source your category leans on, and which of them you're actually present in.

Step 4: check the facts, not just the mentions

While you're reading through the transcripts, read every sentence that mentions your brand specifically and flag anything wrong — the wrong city, an old price, a service you dropped, a country you now ship to that the assistant doesn't seem to know about. This is the manual version of a Brand Truth check, and it's often more revealing than the ranking itself, because a single wrong fact repeated confidently is more damaging than a low ranking that's at least honest.

This is also the point to check whether the assistant is quietly getting something about you wrong, not just leaving you out — the wrong city, an old price, a service you no longer offer. A wrong fact stated confidently is arguably worse than being absent, because a buyer reading it has no way to know it's stale.

Step 5: pick three fixes, not thirty

By now you'll have a pattern, not just a list. Pick three things from it: one fact to correct at the source that's actually publishing it, one gap in the kind of source your competitors are in and you aren't, and one page that could plainly and directly answer one of the questions you lost. Trying to act on everything at once is how this kind of exercise dies after week one — three concrete, source-anchored fixes are far more likely to actually happen than a spreadsheet of thirty vague ones.

Where doing it by hand breaks down

The method holds up. What breaks down is doing it properly, repeatedly. A single run of one question is one sample of a system that can answer differently five minutes later — you can't tell a real pattern from a one-off from a single pass. Assistants also drift as they read new material, so this month's answer isn't a fixed fact about your brand; it needs asking again next month, and the month after that, to know whether anything actually changed.

And doing this correctly every month, for enough questions, across enough assistants, while logging it consistently and comparing it to last month's log, turns into a lot of unglamorous admin fast — the kind of task that gets done diligently for two months and then quietly stops. That's not a failure of discipline so much as an honest description of what repeated, structured measurement actually costs to run by hand.

That's exactly what a free Snapshot automates: the same eight-question, four-engine shape, run without you copying transcripts into a spreadsheet, with each answer already classified and a first look at your AI Shortlist Score and which of the five numbers behind it is weakest.

Doing the manual pass first isn't wasted effort even so. It teaches you to read an AI answer the way this whole exercise requires — critically, watching for what's missing and what's quietly wrong, not just skimming for your own name. That habit is worth keeping even after the monthly measurement itself is automated. If the manual pass above turned up something worth understanding properly, run the free AI Visibility Snapshot — it takes minutes, and it does this same thing wider, and repeatably.