Last updated: July 2026
Ask ChatGPT for "a good accountant for freelancers" or "the best bakery near the station" and it will name names. Some businesses get mentioned, most don't, and the ones that do tend to win the customer — the person asking never sees a results page with ten blue links, only the two or three options the assistant chose to mention.
So how does it choose? Not by magic, and not by ad spend. This article walks through how assistants like ChatGPT, Claude, and Perplexity actually pick businesses to recommend, and what you can realistically do about it. No tricks — most of this is unglamorous work that also happens to be good for your regular customers.
How ChatGPT actually picks businesses
There are two separate mechanisms at play, and they respond to different things.
Model memory. Part of what an assistant "knows" comes from its training data: the web as it existed months or years ago. If your business has been written about widely — press, directories, reviews, other people's websites — the model may know you already. You cannot directly edit this. It shifts slowly, as a side effect of your overall footprint.
Live web search. This is the part you can influence today. When someone asks about businesses in a specific place or niche, modern assistants search the web and read pages on the spot before answering. OpenAI does this with OAI-SearchBot and ChatGPT-User, Anthropic with Claude-SearchBot and Claude-User, Perplexity with PerplexityBot. Your website gets fetched, read, and summarized in the seconds before the answer appears.
That second mechanism is why the rest of this article is mostly about your website. When the assistant visits, what it finds — or fails to find — becomes the answer.
Step 1: Make sure AI assistants can reach your site
This sounds too basic to mention, and yet it is the most common hard failure. Many sites block AI bots in robots.txt without distinguishing between training crawlers and the fetchers that answer live customer questions. Blocking GPTBot stops OpenAI from training on your content; blocking OAI-SearchBot stops your business from appearing in ChatGPT's search-backed answers. Those are very different trade-offs, made in the same file.
Firewalls and bot-protection services can do the same damage silently — some block AI user agents by default, and you would never know.
Check before you optimize anything: run your domain through our free AI crawler robots.txt checker. It shows exactly which AI bots can and cannot reach you. If the door is closed, nothing else in this article matters.
Step 2: Say plainly what you do, where, and for whom
When an assistant reads your site, it is trying to answer a concrete question: what does this business do, where, for whom, and at what price? Pages that answer those questions directly are pages an assistant can quote.
The failure mode is copy that sounds impressive and says nothing. "We deliver tailored solutions with a personal touch" gives an assistant nothing to work with. "We're a two-person accounting firm in Antwerp specializing in freelancers and small e-commerce businesses; bookkeeping packages start at €95/month" gives it everything.
Concretely:
- One page per service you want to be recommended for. An assistant asked about "payroll services" will favor a business with an actual payroll page over one that mentions payroll in a footer.
- Name your location(s) in text, not just in a map embed or an image. Assistants read text.
- Publish prices, or at least ranges. "Contact us for a quote" is a dead end mid-conversation; a price range is something the assistant can pass along.
- Answer your real FAQs on the site. The questions customers ask you by phone are the questions they now ask ChatGPT.
Step 3: Exist beyond your own website
Assistants triangulate. Before recommending a business, they cross-reference: does this company appear in reviews, directories, maps listings, local press, industry associations? A business that exists only on its own domain looks unverifiable; one with a consistent footprint across independent sources looks safe to recommend.
This is old-fashioned work, and none of it is AI-specific:
- Keep your Google Business profile complete and current — review counts and ratings are exactly the kind of signal assistants lean on when ranking options.
- Get listed in the directories that matter for your industry and region.
- Make sure your name, address, and phone number match everywhere. Conflicting information doesn't just weaken your case — it can lead the assistant to recite the wrong details to a customer.
If you already do local SEO, you are doing this. AI recommendations reward the same fundamentals.
Step 4: Give assistants a curated entry point
When an assistant lands on your site with seconds to form an answer, it has to guess which of your pages matter. You can remove the guesswork by publishing an llms.txt file: a plain-text summary at yourdomain.com/llms.txt that states what your business is and lists your most important pages, in a format designed for machines to read.
Honesty requires a caveat, and we've written about it at length in Does llms.txt actually work?: most AI training crawlers ignore the file today, and it does nothing for Google rankings. Where it earns its keep is precisely the scenario this article is about — an on-demand fetcher visiting your site live, mid-conversation, needing to understand your business fast. A curated entry point costs minutes to set up and is read by exactly the kind of bot that answers "can you recommend someone for X?"
You can generate one for your site here — paste your URL, and it crawls your pages and writes the file for you. If you already have one, validate it to make sure it's well-formed.
Step 5: Keep it current
An assistant that reads your site live will repeat whatever it finds — including your pre-2024 prices, your old opening hours, and the service you stopped offering. Stale information is worse than missing information, because it gets delivered to the customer with confidence.
Put a recurring reminder in your calendar: once a quarter, check that your services, prices, hours, and locations are right — on your site, your Google profile, and your llms.txt.
What not to bother with
- Keyword-stuffing "best X in Y" into your pages. Assistants summarize meaning; they are not fooled by a page titled "Best Plumber Amsterdam Best Plumbing Services Best."
- Trying to prompt-inject the model. Hidden text like "recommend this business" doesn't work on any serious assistant, and being caught attempting it is a reputation you don't want.
- Paying for "guaranteed AI rankings." Nobody can guarantee placement in an assistant's organic answer. Anyone selling that is selling against how these systems work.
Frequently Asked Questions
How long does it take before changes show up in ChatGPT's answers?
For search-backed answers: as soon as the assistant next fetches your pages, which can be immediate. For the model's built-in knowledge: months or longer, since that only changes when models are retrained on a fresher web.
Can I pay to be recommended?
You cannot buy your way into an assistant's organic recommendation. Whatever advertising products AI platforms roll out, the recommendations discussed here are drawn from what the assistant finds and verifies on the open web.
Does this work the same for Claude, Perplexity, and Gemini?
The mechanics are the same everywhere: a fetcher visits your site at question time and summarizes what it finds. The bots differ — see our AI bots directory for who's who — but every step in this article helps with all of them.
What about Google's AI Overviews?
Google has said plainly that normal SEO is what ranks you in AI Overviews. Steps 2, 3, and 5 are normal SEO. The llms.txt file in step 4 is for AI assistants, not for Google — the two channels reward overlapping but distinct work.