A wave of free “AI visibility scans” has arrived, and some of them never query a single AI platform. They read your domain name, ask you your industry and location, and then a language model writes a plausible-sounding essay about what ChatGPT or Gemini “would likely” do — scored out of 100, gated behind your email, ending in a sales call. We build measurement tools for a living, so we know exactly what a real check requires — and what corners get cut when it's theatre. Here are the six tells. Run any report you've been given past them, including ours.
Tell 1: no actual answers, anywhere
A real citation check asks the platforms questions and shows you what came back — the query, the answer, who was named, per engine. If the report contains verdicts but not a single quoted answer, nothing was measured. There is no reason to hide the receipts except that there aren't any.
Tell 2: hedge-words doing the work of data
Read the verdict sentences closely. “Would likely draw on”, “unlikely to surface”, “almost certainly not citing”, “may have some footprint”. That is the grammar of a model guessing, not a system reporting. Measurement says “we asked twelve questions; you were cited in two” — no “likely” required.
Tell 3: one model “scanning” five platforms
“Powered by [one AI model], checked across ChatGPT, Gemini, Copilot…” is a logical impossibility dressed as a feature. One company's model cannot inspect another company's answer engine — it can only speculate about it. Checking ChatGPT means calling ChatGPT; checking Perplexity means calling Perplexity. Anything else is one model doing impressions of the others.
Tell 4: the form wrote the report
Notice what they ask before the “scan”: your industry, your location, your email. Then notice the report saying things like “this [your-town]-based [your-industry] business”. When the personalisation traces back to the form rather than your website, the report was generated from your three answers, not from anything discovered. A real scan reads your site and can tell you things you didn't type in.
Tell 5: impossible speed for the claimed work
Genuinely querying multiple AI platforms with multiple questions takes minutes and costs real money in API calls — which is why honest citation checks cost a few pounds, and why ours streams the answers live while you wait. A “full five-platform scan” in 30 seconds is the time it takes a model to write an essay, not run one.
Tell 6: the score exists to book the call
Watch the shape of the funnel: email required before results, a frightening score, no fixed prices anywhere, and a “book a free discovery call — priced on the call” ending. A tool built to measure gives you the result and a price list; a tool built to harvest leads gives you a fright and a calendar link. Neither is illegal — but you should know which one you're holding.
To be fair: what inference can legitimately do
Checking entity signals — whether you have structured data, a Google Business Profile, a knowledge-graph presence — is real, useful work, and some of the advice in these reports (claim your profiles, add schema, fix your citations) is perfectly sound. The dishonesty isn't the advice; it's dressing plausibility up as measurement and scoring it out of 100. A guess with a number on it is still a guess.
The one question that settles it
Ask the provider: “show me the actual answers.” Which questions were asked, to which engines, and what came back, word for word. A real check has them instantly — ours shows you all 60 answers arriving live, names who was recommended instead of you, and locks the questions so a rescan is a fair before/after. Our technical check is free and shows its working, no email required. If a provider can't show the answers, you already have yours.