How to get your business recommended by ChatGPT and AI search
The practical version. What an AI assistant actually reads before it names a business, which parts you control, and the four things that are genuinely different from ordinary SEO.
- An assistant can only recommend a business it can find, read and trust. That means a fast site that plainly states what you do and where, real reviews, and the same contact details everywhere. This is the same groundwork that wins ordinary search.
- Four things are specific to AI answers: describing your business in linked structured data, putting the direct answer at the top of the page, publishing checkable specifics like prices and dated sources, and allowing the AI crawlers on purpose rather than by accident.
- A peer-reviewed 2024 study found that content changes alone lifted visibility in generative engine answers by up to 40%, with citing sources and adding statistics among the strategies tested.
- It is additive, not a replacement. Most local customers still arrive through ordinary search and the map, so do not stop doing the basics. Whether AI search is worth caring about yet is a separate, honest question.
Someone asks ChatGPT for a plumber in Croydon, or asks Google a question and reads the AI answer instead of the results. Two or three businesses get named. The rest are invisible, and they never find out it happened.
The good news is that getting named is not mysterious and it is not expensive. It is mostly the same groundwork that has always won local search, plus four specific things that make a site legible to software. This guide is the practical list, in the order worth doing them.
How an assistant actually picks a business
An AI assistant is not ranking pages the way a search engine does. It gathers a handful of sources, reads them, and writes an answer that quotes or paraphrases the most useful passages. Being named is less like ranking first and more like being quoted in an article.
That changes what matters. A page can rank respectably and still never be quoted, because it buries the answer under three paragraphs of preamble, or because nothing on it is specific enough to repeat. Conversely, a modest page that states a price plainly and cites where a number came from is very easy to quote.
Assistants also weigh what other sources say about you. Your Google Business Profile, consistent listings and genuine reviews do a lot of work here, which is why none of this is separate from ordinary local SEO.
Most of it is the groundwork you should be doing anyway
Before anything AI-specific, the basics have to be in place, because an assistant cannot recommend a business it cannot read. A site that loads slowly on a phone, hides its services behind clever wording, or has no consistent name, address and phone number is a bad source, and it will be skipped in favour of a competitor who is easier to understand.
| What both reward | What is specific to AI answers |
|---|---|
| A fast page, especially on mobile | Structured data linking the business, the person and the page as one entity |
| Plain wording that names the service and the area | The direct answer placed first, before the detail |
| A complete, verified Google Business Profile | Checkable specifics: real prices, named sources, honest dates |
| Real reviews and consistent listings | Crawler access granted deliberately, plus a plain-text site summary |
| Internal links that make sense | Content a machine can quote without rewriting it |
The four things that are genuinely different
These are the additions worth making, and between them they take hours rather than a retainer.
- Describe the business in structured data, and link it together. Most sites emit a scatter of unconnected fragments: a business name here, an author there, a price somewhere else. Give them shared identifiers so an assistant reads one confident entity instead of guessing whether three mentions are the same company. Google's structured data documentation is the reference.
- Put the answer first. Every page that answers a question should answer it in the opening lines, in plain language, with the detail underneath for humans who want it. If your page takes four paragraphs to get to the point, the assistant quotes whoever got there in one.
- Publish things that can be checked. Real prices instead of 'contact us for a quote', statistics with a named publisher and a date, and comparison tables. Vague marketing copy gives an assistant nothing to repeat, so it repeats someone else.
- Let the crawlers in on purpose. The assistants use their own crawlers, and a surprising number of sites block them by accident through a default setting or an over-eager security plugin.
That study is worth knowing about because it tested this properly rather than guessing. Among the strategies it evaluated, adding citations to authoritative sources and adding relevant statistics were among the effective ones. Keyword stuffing, predictably, was not.
Letting the crawlers in, deliberately
Each assistant uses named crawlers, and your robots.txt decides whether they may read you. One trap catches people out: robots.txt rules are not cumulative. If you give a named crawler its own group, it ignores the general rules entirely, so anything you meant to keep private has to be repeated inside every group.
The agents worth knowing by name, as of late 2026:
- OpenAI: GPTBot for training, OAI-SearchBot for search, ChatGPT-User for live browsing. OpenAI publishes the list and the IP ranges.
- Anthropic: ClaudeBot, Claude-User and Claude-SearchBot.
- Perplexity: PerplexityBot, with its own documentation.
- Google: Google-Extended controls use in Gemini and AI answers separately from ordinary Googlebot crawling.
- Also worth allowing: Applebot and Applebot-Extended, Amazonbot, Meta-ExternalAgent, and CCBot, which feeds Common Crawl and therefore a great deal of downstream training data.
The second thing worth doing is publishing a plain-text summary of your site for assistants to read, conventionally at /llms.txt. It is a short markdown file listing what you do, what you charge and where your main pages are, so a machine can get the facts without parsing your HTML. The convention is documented at llmstxt.org, it costs nothing to publish, and it removes any excuse for an assistant getting your prices wrong. This site publishes one of its own if you want to see the shape of it.
What you cannot control, and should be suspicious of
- Nobody can guarantee that an assistant will name you. The engines do not publish how they select sources and they change without notice.
- There is no equivalent of Search Console for AI answers yet, so measurement is indirect: you ask the assistants questions a customer would ask and see who gets named.
- Anyone selling a 'guaranteed ChatGPT ranking' is selling something that does not exist. Be as sceptical as you would be about a guaranteed Google position.
- Being cited does not always send a click. Some answers name you without linking, which is still worth having, because a recommendation by name is how people decide who to phone.
How to tell whether it is working
Pick the five questions a genuine customer would ask, phrased the way they would phrase them, and ask each of the major assistants once a month. Write down who gets named. It is crude, it is free, and it is currently a better signal than anything on a dashboard.
Then watch the ordinary numbers underneath it. If the groundwork is right, organic enquiries and map views tend to move first, because the same work feeds both. If nothing at all is moving after a few months, the problem is usually upstream: the site is too slow, too thin, or does not plainly say what you do and where. That is the part worth fixing first, and a free website audit will tell you which it is.
Common questions.
How do I get my business mentioned by ChatGPT?
By being an easy source to read and verify. In practice: a fast page that plainly names your service and your area, structured data describing the business, real prices rather than 'contact us', a complete Google Business Profile with genuine reviews, and robots.txt permitting OpenAI's crawlers. There is no submission form and no paid placement. You are making yourself the most quotable answer to the question, and the assistant does the rest.
Is GEO different from SEO, or is it the same thing rebranded?
Mostly the same, with real additions. Around four fifths of the work is shared: speed, plain wording, structured data, consistent details, reviews. What is genuinely new is that assistants quote passages rather than rank pages, so answer-first structure and checkable specifics matter more than they used to, and crawler access has to be granted deliberately. If someone is selling GEO as an entirely separate product with its own retainer, be sceptical.
Do I need an llms.txt file?
It is not required and no assistant guarantees it will read one, but it is cheap and low risk. It is a short plain-text file summarising what you do, what you charge and where your key pages are, which means a machine can get your facts without interpreting your page layout. For a small business with clear services and published prices it takes about twenty minutes to write and removes the chance of an assistant quoting a price that is three years out of date.
Will AI search replace Google for local businesses?
Not yet, and probably not entirely. Ordinary search and the map still send the overwhelming majority of local enquiries, so anyone telling you to stop doing local SEO is giving you bad advice. What has changed is that a second surface now exists alongside it, and it is won with largely the same work. Treat it as additive.
How long does it take to show up in AI answers?
Longer than a search ranking, and less predictably. Assistants draw on crawled content and on training data that can be months old, so a brand new page may be quotable within weeks by the live-browsing tools and take considerably longer to appear in the models' own knowledge. Treat it as a side effect of doing the groundwork well rather than a campaign with a launch date.