Is my brand visible in AI search? How to find out in an hour

The queries that matter are the ones asked by people who do not yet know you exist. "What does [your brand] do?" tells you how assistants describe you, which is useful, but it is not where deals are won or lost. Most of your list should be unbranded.

A mix that works for most brands: four category queries ("best [category] for [audience]", "top [category] tools in the UK"), two comparison queries that name a competitor, two problem queries phrased the way a non-expert would phrase them, one "alternatives to [market leader]", and one branded query to test how accurately you are described. Pull the wording from sales calls, support tickets and your search console data rather than inventing it at your desk. If the queries are ones nobody asks, the audit measures nothing.

You do not need every assistant. Three is enough for a first audit, and ten queries across three assistants gives you thirty answers, which fits comfortably in this step. Choose the three your buyers most plausibly use: for most UK firms that means ChatGPT plus two of Gemini, Perplexity, Copilot and Claude. If ChatGPT is the one that matters most to your market, our ChatGPT visibility tracker page covers what makes it different in more depth.

Practicalities: open a fresh chat for every query so earlier answers do not colour later ones, and note whether the assistant searched the web before answering. Perplexity searches by default; ChatGPT and Gemini sometimes search and sometimes answer from training data, and none of the engines fully publish the rules for when each happens. The distinction matters because an answer from training data can be years stale. Record the first answer you get. Do not coach the assistant into mentioning you; a mention you had to ask for tells you nothing about what your buyers see.

Set up a spreadsheet with a row per query per assistant and five columns. Named: yes or no. Position: first brand mentioned, one of a list, or an afterthought near the end. Claims: quote verbatim what the assistant says about you, including pricing, positioning and intended audience, because paraphrasing hides errors. Sources: which domains the answer cites, if any. Search: whether the assistant visibly used live web search.

The sources column earns its keep later. The domains assistants cite for your category queries are, in effect, the publications and pages that currently decide whether you exist in AI answers. That list becomes your outreach and content target list, so record it carefully even when you are not named.

Rough yardsticks, not a formal score. Named in more than half of your unbranded answers: a solid baseline. First-named on any category query: genuinely strong. Never named on unbranded queries: the invisible pattern. Judge accuracy separately from presence: a materially wrong claim, such as the wrong pricing tier, a discontinued product or the wrong target market, is a fail for that row regardless of how prominently you appear.

One caveat. Assistants are not deterministic, and the same query can produce a different answer an hour later. Treat a single absence as weak evidence and a pattern across ten queries as meaningful. This audit is a snapshot, not a measurement instrument, which is exactly why you should keep the spreadsheet and repeat it.

Invisible. You are absent from most unbranded answers. The usual cause is not technical; it is that the sources assistants lean on do not mention you. Check your sources column: if the cited domains are review sites, industry publications and comparison pages you have never appeared on, that is the gap. Next step: earn coverage on those exact domains, and publish pages on your own site that answer your category queries directly and in plain language.

Misdescribed. You are named, but the claims are wrong. You are in common company here. In our July 2026 test of 30 brands, 27 were described with at least one materially wrong claim when the AI answered without live web search (full study). Next step: make the correct facts unmissable. A plainly written about page, consistent descriptions across directories and third-party profiles, and current pricing on pages assistants can read all shrink the room for confident guessing.

Outranked. You are named, but consistently behind the same two or three competitors. Read what the cited sources say about them. Assistants tend to echo the framing of their sources, so the fix is rarely "mention us more" and usually "the comparison pages that dominate this category describe the winners in detail and us in a single sentence". Closing that gap is slower work than the other two patterns, but it is at least precisely targeted: you know which pages to improve or displace, because your spreadsheet names them.

Keep the spreadsheet and re-run the same ten queries monthly by hand. Visibility shifts when models update or engines change how often they search the web, and you will only notice those shifts if the queries stay constant.

If you want the first pass done for you, our free AI visibility checker runs this style of audit on demand. To be straight about its current state: it tests Claude with live web grounding today, support for ChatGPT, Perplexity, Gemini and Copilot is rolling out, and it does not run on a schedule, so repeating the check is a manual step, exactly as it is with the spreadsheet.

Whatever the result, you will finish the hour knowing which of the three patterns you are in and what to fix first.