It's almost never random. When ChatGPT, Claude or Perplexity answers a buyer's question and your business isn't in it, there's a specific, checkable reason — usually one of seven. We built a scanner that checks 40+ signals AI models use to read and trust websites; these seven account for most of the damage we see. Each is fixable, and none requires magic.
1. Your site blocks the AI crawlers
GPTBot (ChatGPT), ClaudeBot, PerplexityBot and Google-Extended have to read your site to cite it. Security plugins and some hosting firewalls block them by default — the digital equivalent of locking the door and wondering why nobody visits. Check your robots.txt, or run the free check, which tests every major AI crawler's access in one pass.
2. The models can't tell what you actually are
Humans infer from photos and layout; models need it stated. If your site never says — in machine-readable structured data — what you do, where you operate and what credentials you hold, the model can't confidently recommend you. Confidence is the currency: an AI would rather name a directory it understands than a business it half-understands.
3. Your pages describe — they never answer
Buyers ask questions. Models cite pages that answer them, in the first paragraph, with specifics: what, where, from what price, how fast. Most small-business sites open every page with a slogan. “Proudly serving South Wales since 1998” answers nothing; “We fix leaking boilers across Cardiff, usually same-day, from £85” is citation material.
4. You never name the places you serve
Ask a model “plumber in Penarth” and it recommends pages that say Penarth. If your service area lives only in your head — or in a footer map image — every local query goes to a competitor or a directory whose pages spell it out in text.
5. Nobody else on the web corroborates you
Models weigh third-party evidence: directories, review platforms, local press, industry bodies. A business no one else mentions is a business the model hesitates to recommend. This is why directories dominate AI answers for local trades — they carry thousands of corroborating signals. You don't need hundreds of mentions; you need your Checkatrade, Google Business Profile and trade-body listings live, consistent, and linked from your site's structured data.
6. Your site is technically unreadable to a machine
Some sites are app shells: a human with a browser sees a lovely page, a crawler sees an empty div waiting for JavaScript. Others bury content in images or PDFs. If the machine can't parse the words, everything else on this list is moot. This is invisible from the outside — it's exactly what the technical scan exists to catch.
7. You're only visible for your own name
The subtle one. Ask about your business by name and the model knows you fine — so you assume you're visible. But buyers who already know your name were never the prize. Test the questions people ask before they know you exist. That's why eleven of the twelve questions in our citation audit never mention your brand.
Which one is costing you?
Reasons 1, 2, 4 and 6 show up in the free 60-second check. Reasons 3, 5 and 7 show up when you test real answers — the £19 citation audit runs 60 of them and names who's winning instead of you. Fix in that order: technical first (provable same day), content second, then rescan. Citations follow the work by two to four months — anyone promising faster is guessing.