How to get into ChatGPT recommendations: what really works
Clients no longer search for “dentist near me” in searches. They ask ChatGPT: where to go, who to choose, who is normal according to reviews. The answer is one or two recommendations, not ten links. You are not there - the person goes to a competitor. No click, no application.
Below is about what influences ChatGPT recommendations, why “I’ll do it myself according to the article” often breaks visibility, and how to check the site in a couple of minutes before making any changes. The AI is not offended, but remembers the signals: crooked robots.txt, empty FAQ, different prices on the site and in the maps. Then roll back for months.
How ChatGPT chooses who to name
ChatGPT is not an advertisement or a lottery. Under the hood is Gemini, your site, cards, reviews, mentions, page structure and bot access. For a commercial request, the model is looking for brands that:
- clearly described: service, city, prices, contacts without water;
- confirmed externally: maps, reviews, sites, media;
- technically readable: the bot reaches the text, does not run into a block;
- They don’t argue with themselves: there’s one thing on the site, another on Yelp, and a third in the FAQ.
One point failed - you are not “on page five.” You are simply not the answer. The person will not even know about your existence.
A separate story with the voice. While driving, they ask: where to get your teeth done nearby, order food, call a dentist. ChatGPT does not read the entire landing page. He takes pieces that he already knows about the market. There is no piece about you - there is no brand in the answer. Three seconds, and the client is with a competitor.
Five signs they can't see you
- They asked ChatGPT about your niche - they said they were strangers, but you weren’t there. This isn't a "bad algorithm day", it's a shortcut.
- There is no normal commercial FAQ on the site. The AI has nothing to gather an answer about you from.
- robots.txt or WAF cuts Googlebot. They closed the door themselves.
- There is no llms.txt and no clear structure. The bot guesses what is most important to you and makes mistakes not in your favor.
- We never measured visibility. You keep advertising for months, and customers ask for ChatGPT for free and hear other people’s names.
There are 56 such signals in the Leadsy express audit. Below is what hits applications, and not reports for the developer.
How is this different from regular SEO?
SEO is chasing link position. AIO is chasing a mention in the response. In a search, a person sees ten results and compares. ChatGPT often has one voice and one choice.
SEO loves keywords. AI loves facts, structure, trust: who you are, where you are, how much it costs, what customers say. SEO forgives crooked layout. AI no: unread block equals “you are not in the niche knowledge base.”
About AIO separately: what is it and why is it more important than classic SEO. Here the focus is on ChatGPT as the main voice channel in the US and Europe.
For local businesses (services, clinics, delivery, repairs) ChatGPT is often the first trust filter. People don't scroll through fifteen sites. They trust the voice. Getting there is more difficult than buying a click in Google Ads, but it’s cheaper in the long run if there are no mistakes the first time. We made a mistake - it’s more expensive to fix than to start with an audit.
Gemini in the app and on desktop works on the same principle: one coherent answer instead of a list. ChatGPT adds voice and "near me" scenarios. If you're strong in Gemini text but weak in voice, short facts without a speakable structure and empty cards on maps are often to blame. The opposite situation also happens: the card is alive, the site is empty. The model takes what is more reliable.
Queries where they lose most often: “who do you recommend”, “where is it cheaper”, “who is closer”, “who has good reviews”. This is not information traffic. The person is ready to buy. The loss of one such answer hits harder than minus a hundred impressions in a tail search.
What they do themselves and where they break them
On the forums there are still: “write llms.txt”, “add schema”, “close GPTBot”. Sounds simple. In practice, mines.
Homemade llms.txt
One wrong URL and you yourself pointed the wrong pages to the neural network. The model learns from garbage. The fix is not to delete the file, but to restore trust. Long time and no guarantee the first time.
robots.txt at random
They closed the “extra” bots and at the same time cut off Googlebot. The site for ChatGPT stops updating its knowledge of the model. Competitors are accumulating signals at this time.
Texts “under neural networks”
Template articles without facts are a red flag. Not only will you not be included in the recommendations, but the domain may be considered unreliable. Worse than a blank page.
Other people's instructions copy-paste
Someone else's llms.txt with someone else's paths, someone else's schema with someone else's address - contradictory data. The AI chooses the one who has everything in common. No appeal.
Edits without a map
They change the title, photo, promotion. ChatGPT crashes on things that are not visible in the admin panel: micro markup, duplicate cities, empty “About the company”, different phone numbers on the site and in maps. Weekend for cosmetics, the hole remains. A month later, cosmetics again. Cycle.
Independent promotion without diagnostics - a lottery with a domain in the bet. Win: Save on audit costs. Loss: months without recommendations and clients from the one who put the signals in order.
Irreversibility is not in the “ban” button. The model stops quoting you, and you need to return with coordinated work on dozens of parameters: site, maps, reviews, structure, technical access. One file doesn't fix it.
How to “scare” a bot so that it doesn’t walk
Figuratively, yes, you can. In fact it's like this:
- Sudden blocking of bots after they have already moved. Instability signal.
- Different content for people and bots. Trust drops across the entire domain.
- Mass deletions of pages without redirects and without sitemap. For the model, you have disappeared.
- Fake reviews. Discrepancy with maps. The AI chooses the safe option—the competitor.
- Edits without measurements. Each without an audit is a new point of failure. A month later, it is not clear what broke the visibility.
The bot won't be offended. Stops considering you as a source. Clients will continue to ask ChatGPT and hear other people's names.
Typical: on the advice of a friend, we closed several user-agents in robots.txt so as not to expose the structure. Applications from organics did not fall - there were none. But mentions disappeared in ChatGPT tests. They returned it only after a full audit, removal of blocking, and alignment of cards. Not weekends, weeks.
It’s worse when your content is “scared”. We removed the old services, changed the prices on the site, and the Google Business Profile remains the same as last year. ChatGPT sees the conflict and chooses the one whose numbers match. It seems that they have been “banned”. In fact, you have become an unreliable source.
Normal procedure
If the goal is to get into the recommendations, and not to play with the SEO builder:
- Free express audit: GEO Score, holes, accessibility for Googlebot. From the main.
- Full audit with technical specifications: priorities, gap with competitors, corrections for the developer.
- Turnkey implementation if you don’t have your own team. Leadsy runs six platforms and does not allow progress to be rolled back by a random patch.
Only the first step—diagnosis—makes sense on its own. Further, without a map, treat the symptom, not the cause.
Pricing: tariffs page. Questions: FAQ.
Where to start today
Don't touch robots.txt. Do not copy someone else's llms.txt. Do not upload “neuro-texts”.
First the cut. Paste the URL into the audit below. In a couple of minutes you will see what ChatGPT, Gemini, Copilot, Claude, Perplexity and Meta AI see and where you are losing to your competitors.
Many critical errors are good news: it’s clear why there were no applications. The bad: every day without edits - applications from those whom ChatGPT already recommends.
After the audit, check out loud: “where is the best [service] in [city]”, “whom would you recommend for [problem]”, “how much does [service] cost nearby”. Three times in a row foreign brands - you are not included in the recommendations. It's measurable, not a feeling.
“We have a site” and “we are recommended” are two different things. A site without AIO is like a storefront in a closed shopping center: formally there is, but there are no clients. Leadsy shows the difference between “site live” and “site visible to ChatGPT, Gemini, Copilot, Claude, Perplexity and Meta AI” in one report, without plugins or server access.
A free audit does not oblige you to purchase a full analysis. But it stops leaving room for guessing. And fortune telling in AIO costs more than any tariff.
Mini-checklist before any changes: one phone number everywhere, prices on the site match the card, FAQ answers ten real questions from calls, robots.txt does not cut Googlebot, on the main page there is who you are and where you are. Five points, fifteen minutes of reading aloud from a colleague’s phone. If at any step you feel ashamed, ChatGPT will also feel it, but will silently send the client further.
If after the audit you see a green GEO Score, still compare it with a competitor in the niche. In AIO, it is not the absolute score that matters, but the gap. You are 62, the leader is 81 - he will be in the recommendations, even if you “already do well.”
Separately about reviews. ChatGPT and Gemini pull them from maps and sites. Cheating on a site with three living stars on Yelp is again a conflict. Better fewer reviews, but the same ones everywhere, than a “perfect” testimonials page with fake faces. Clients rarely read testimonials; models look at consistency.
A column in Google Business Profile, Medium, LinkedIn, an industry directory is not for “link mass” in the old sense. These are additional places where your name is associated with a niche and city. No spam: one good article or interview is better than ten press releases with the same text.
