Be the name the AI recommends
Your clients ask ChatGPT or Perplexity for "a good osteopath nearby." Here's what these AIs read to recommend you — or skip you.
More and more, your next client doesn't type "osteopath Lyon" into Google. They ask ChatGPT, Perplexity, or Google's built-in AI: "a good osteopath near Croix-Rousse?" And they get a short answer — two or three names, one line each. If you're not on that list, you don't exist for that client. This isn't the Google SEO of before, and it works differently.
What the AI reads to decide who to cite
An answer engine doesn't rank pages, it synthesizes. To reply, it gathers what it finds about you all over: your site, your socials, your reviews, directories, mentions elsewhere. Then it favors what's clear and consistent — who you are, who you help, where you work — and what shows real expertise. A vague, generic, or absent profile gives it nothing to cite. It moves on to the next one: the one it understands better.
Why an independent starts with an advantage
It's counterintuitive, but these AIs favor the specific, not smooth high volume. A big chain produces interchangeable content; you have a precise angle, a clientele of your own, real know-how — exactly what an answer engine needs to give a useful answer instead of a vague one. Your real advantage is being identifiable. The AI still has to be able to identify you — and that's where the rest is decided.
If you're not on that list, you don't exist for that client.
The levers that actually matter
From most worthwhile to gimmick, in order.
Don't block the AI crawlers. The dumbest trap: a robots.txt — or an SEO plugin — that blocks GPTBot (ChatGPT), PerplexityBot, ClaudeBot, or Google-Extended. If they can't read your site, you're absent from the answers, full stop. Get that line checked before anything else.
Tend to Bing, not just Google. ChatGPT leans heavily on Bing. Create and complete your Bing Places listing on top of your Google Business profile: it's the channel almost no one thinks to fill in.
Give it structured data to read. On your site, schema.org markup (LocalBusiness type, in JSON-LD) spells out your name, address, hours, services, and area. That's what machines read first — a one-time job for the person who handles your site.
Get cited somewhere other than your own site. AIs trust outside validation: client reviews, trade directories, and especially the Reddit threads or forums where your audience asks its questions — models draw heavily on those. One useful answer under a real question weighs more than ten "optimized" pages on your own site.
llms.txt? A bonus, not a priority. It's a small file that summarizes your site in markdown for AI agents. Adding it costs almost nothing — but as of today, no major provider (OpenAI, Google, Anthropic) has committed to using it, and nearly all crawlers ignore it. Add it if you feel like it, without expecting miracles.
What doesn't work
Stuffing a page with keywords: answer engines read meaning, not density. Copying generic "optimized" content: that's exactly what the AI sets aside, because it can't attribute it to anyone. And vanishing for three months: a dormant account sends the opposite of trust.
The part you can automate
Half this list is settled once: the robots.txt, the Bing listing, the structured data on the site. The other half plays out every week — regular, recognizable content rooted in your trade, on your networks and where your audience looks. That continuous part is what took a time nobody had — least of all someone running a business alone. An AI that reads your site and your real work can produce that flow — posts and visuals calibrated to your brand, across your five networks — so you become, little by little, the name the other AIs end up recommending. You give it your real material, not a blank prompt, and the rest follows.
