How to audit your own AI visibility in an afternoon.

The biggest DIY mistake is testing one flattering query, seeing your name, and relaxing. Build a small map instead. Three tiers, five queries each: category queries ("best [what you do] in [your city]"), service queries ("who does [specific service] for [specific kind of client]"), and decision queries ("[your name] vs [competitor]," "[your name] reviews," "is [your name] good").

Use the words buyers use, not the words your website uses. Nobody asks ChatGPT about "integrated growth enablement solutions."

Test ChatGPT, Perplexity, Gemini, Claude, and Copilot. Three rules keep the data honest: fresh sessions every time, so personalization and chat history don't flatter you; identical wording on every engine, so the comparison means something; and dated screenshots of every answer, because memory is not a record.

For every query, log four things: whether you were named, your position in the answer, what the engine actually said about you, and which sources it cited. The last two are where the findings live. A mention that describes the wrong specialties is a problem wearing a win's clothing. And the citation list under a Perplexity answer is a free map of exactly which sources you need to show up in — usually a review platform, a trade publication, or a round-up you're not in yet.

Score yourself roughly: named on a service or decision query, 2 points. Named on a category query, 1. Hallucination about your business, minus 2. Out of a possible ~25, most businesses we've seen score under 8 on their first pass. Under 4 is common and fixable.

Three patterns cover almost everyone. Invisible everywhere: your third-party evidence layer is thin — engines have nothing to quote about you. The fix starts with infrastructure and citations, not your website. Visible on browsing engines, missing from base models: your record is newer than the last training snapshot; keep building and wait out the cycle. Named but described wrongly: you have a hallucination with a stale source behind it — find which cited page is feeding the error, because that's the page to fix or displace.

Having watched many first audits, we can save you some surprise. The most common discoveries, in rough order: total invisibility on queries you assumed you owned; one competitor named on nearly every engine, usually not the one you watch; at least one factual error about what you do or where you are; and a citation list dominated by two or three sources you've never deliberately worked on.

Whatever you find, write it down in a form you can re-run. The single highest-value habit from this whole exercise is a spreadsheet with your fifteen queries down the side, five engines across the top, and a dated tab per month. Three months of that and you have something most businesses never see: their AI visibility as a trend rather than a rumor.

Honesty about the limits: fifteen queries is a snapshot, not coverage — we run 80+ because service-level visibility hides in the long tail. One afternoon is a data point; the value compounds when the same map is re-run weekly and you can see answers move. And tracing a hallucination to its source takes tooling and stubbornness that's hard to improvise on a Saturday.

But run the afternoon version anyway. Worst case, you confirm you're fine. The usual case: you watch an AI recommend three competitors to a buyer who was, functionally, asking about you — and the rest of this blog suddenly gets very practical.