The test that hurts
Open ChatGPT and ask the question your prospects are already asking: "Recommend a [your trade] in [your city]". Look at the answer. In most cases, three or four businesses are cited, with a summary of their strengths. If you are not among them, those prospects go to the ones who are, without ever seeing your name.
This behavior is no longer marginal. A growing share of provider searches now starts in a conversational AI rather than on Google, especially among people under 40. And unlike a Google results page where ten links share the attention, an AI answer cites two to four names. Spots are scarce, and they are being decided right now.
Where do AI engines get their answers?
A model like ChatGPT combines two sources. First its training memory: everything it read from the public web while it was being built, with a lag of several months. Then live search: for local or recent questions, it queries the web at the moment of the question and synthesizes the pages it finds.
In both cases, the raw material is the same: your website, your Google listing, professional directories, customer reviews, articles that talk about you. A model cannot recommend a business the web says nothing clear about. It only "knows" you if the web describes you.
Why are you missing?
Four causes come up in nearly every case we audit:
- Your website does not say who you are. A homepage showing "Welcome" and photos, without a sentence like "[Name], [trade] in [city] since [year]", is unreadable for a model.
- Your AI crawlers are blocked. GPTBot, ClaudeBot or PerplexityBot banned in the robots.txt, often by a plugin setting nobody ever approved. The result: AI engines that search live cannot read you.
- No structured data. Without Organization or LocalBusiness JSON-LD, the model has to guess your business and your service area instead of reading them.
- The outside web does not mention you. No directory, no reviews, no article: nothing confirming your existence beyond your own website. Models are wary of businesses described by a single source.
The 5 levers to exist
- Spell out your identity in plain words. On the homepage and the about page: who, what, where, since when, for whom. Facts, numbers, city names. That is the material the models cite.
- Add structured data. An Organization or LocalBusiness JSON-LD block with name, address, phone, service area and services. Ten minutes of work, readable by every engine.
- Publish an llms.txt. The file that summarizes your website for AI engines. Our free generator creates it for your domain.
- Open your doors to AI crawlers. Check the robots.txt: GPTBot, ClaudeBot, PerplexityBot and Google-Extended must be allowed through.
- Multiply consistent mentions. An up-to-date Google listing, directories in your industry, regular customer reviews. Every external source repeating the same information strengthens the model's trust.
How long does it take?
AI engines that search live, like Perplexity or ChatGPT in search mode, reflect your changes within a few days to a few weeks: that is the fast lever. The models' training memory, on the other hand, updates with each new version, on a rhythm of several months. Concretely: what you fix this week feeds live answers as early as this month, and the models' deep knowledge within six to twelve months. One more reason not to wait: your lead is being built while your competitors sleep.