Most conversations about AI visibility happen entirely in the abstract. Someone read that ChatGPT is "eating search". Someone else thinks their site must be in there somewhere. Nobody has actually checked.
You can check. It takes an afternoon, it costs nothing, and it tends to settle the argument quickly. The hard part is not the tooling. It is being willing to look at a few answers you might not like.
Here is how to do it properly.
Cited is not the same as ranked
The first thing to let go of is the idea that your ranking tells you. It does not, at least not on its own.
When an AI tool answers a question, it is not reading off the top blue link. It is assembling an answer from a handful of sources it considers clean, specific and defensible, then sometimes naming them and often not. You can rank first for a term and still be absent from the answer a person actually reads. You can rank fourth and be the one quoted.
So "are we visible in AI search" is really three separate questions. Does the tool mention your business at all? Does it link or cite a specific page? And is the thing it says about you correct? Those come apart more often than you would expect. A tool will happily describe your business warmly while citing a competitor's page for the actual claim.
You measure all three the same way: you ask, and you write down what comes back.
Step one: ask the questions your customers ask
Make a short list of the real questions a customer would type, in their words, not yours. Not "commercial printing Leeds" but "where can I get 500 booklets printed near Leeds this week". Aim for ten to twenty. Cover the obvious ones, a couple of comparison ones ("X versus Y for a small business"), and one or two where you would genuinely hope to be the answer.
Then run each one through the four tools that matter for most UK businesses right now: ChatGPT, Google's AI Mode, Perplexity, and Claude with web search switched on. For each prompt, record four things in a simple sheet:
- Were you mentioned at all?
- Were you cited or linked, and if so, which exact page?
- How were you described, and was it accurate?
- Who else showed up, and were they cited more prominently than you?
That last column is the one people skip and the one that teaches the most. AI answers are comparative by nature. Knowing which three competitors keep appearing next to you, and which page of theirs gets quoted, tells you more about what to fix than any score out of a hundred.
It is tedious. Twenty prompts across four tools is eighty answers to read. It is also the most clarifying thing a small business can do about AI search, because it replaces a vague worry with a specific list.
Step two: check whether the bots even reach you
The prompts tell you what the tools say. Your server tells you whether they can see you in the first place.
Every major AI provider now sends documented crawlers, and they split the job up. There are training crawlers (OpenAI's GPTBot, Anthropic's ClaudeBot), search crawlers that index pages specifically for AI answers (OpenAI's OAI-SearchBot, Anthropic's Claude-SearchBot, PerplexityBot), and on-demand fetchers that grab a page the moment a user asks about it (ChatGPT-User, Perplexity-User). Google-Extended controls whether your content feeds Gemini, separately from normal Search.
Search your access logs for those names. If the search crawlers are visiting and the pages they fetch return cleanly, you are reachable. If a firewall, an aggressive bot rule, or a robots.txt line is quietly turning OAI-SearchBot or PerplexityBot away, you can be the best answer on the internet and still never appear. We have found exactly this on sites that were blocking AI search bots by accident, usually a leftover rule from someone trying to stop scrapers a year earlier.
While you are in the analytics, add one more check. Since June 2025, ChatGPT tags the links in its answers with a marker (utm_source=chatgpt.com) so referred visits show up as referral traffic from chatgpt.com, with Perplexity and others appearing the same way. Filter for those sources and you can see the clicks AI answers are actually sending you, as opposed to the mentions that never get clicked.
What these checks cannot tell you
None of these measurements is clean, and pretending otherwise just sets you up to over-react.
The prompts are a snapshot, not a metric. AI answers are non-deterministic: ask the same question twice and you can get different sources, and answers shift with phrasing, location, personalisation and the model's mood that week. Treat one run as a reading, not a verdict, and repeat it monthly rather than agonising over a single result.
The referral data undercounts badly. A great deal of AI traffic still lands as "direct" in analytics, and any answer that satisfies the person without a click leaves no trace at all. Being useful inside an answer nobody clicks is real visibility that your dashboard cannot see. For most small businesses the click numbers are still small, so the prompts matter more than the referral chart for now.
And the tools genuinely disagree with each other. Being cited well by ChatGPT tells you very little about whether Perplexity or Google AI Mode will quote you, because they draw on different sources and weight them differently. That is why you check all four separately instead of assuming one stands in for the rest.
Don't buy a tool until you have done this once
There are paid platforms that automate all of this, run your prompt set on a schedule, and chart your citation rate over time. Some of them are good. Most of them cost more per month than the half-day of checking that would tell you whether you need one.
So the buying heuristic is simple: run the manual version once, by hand, before you pay for anything. If twenty prompts across four tools, plus a log check, once a month, is genuinely too slow for the size of your business, then a tool earns its place. If you have never done it once, you are buying a dashboard to measure a problem you have not confirmed you have.
For a ten-person business, the manual version is usually enough on its own. The point is not a beautiful chart. It is knowing, this month, whether you are in the conversation or not.
We run this same triangulation across ChatGPT, Google AI Mode, Perplexity and Claude on the sites we look after, because a ranking now describes a smaller and smaller part of whether the website is doing its job. The answer the tools give is the part most owners never see, and it is usually the part worth fixing first.
If you want a hand running the first pass on your own site, or turning the list it produces into actual changes, have a 20-minute chat. Bring the questions your customers ask. We will go and look at what the tools say back.
