AI search
How to get your business recommended by ChatGPT and AI search
To be recommended by ChatGPT, Gemini, Perplexity or Google's AI answers, a business needs four things: crawlable pages that are not blocked in robots.txt, clear factual statements about what it does, where and at what price, consistent mentions of those facts across third-party sites the models already trust, and content structured so a specific claim can be lifted without reading the whole page. Add an llms.txt file, keep answer-first paragraphs at the top of key pages, publish real prices, and make sure your Google Business Profile and major directory listings agree with each other. Nobody can guarantee an AI mention: there is no ranking to buy, no submission form, and the answer differs between users and between sessions.
A meaningful share of the people who used to type "web designer near me" into Google now ask an assistant instead, in a full sentence, and take the two or three names it gives them. That change is happening faster than most small businesses have noticed, and the mechanics of getting named are not the same as the mechanics of ranking.
This is being sold as a new discipline with a new acronym, usually GEO or AEO. Most of it is not new. But some of it genuinely is, and the parts that are new are cheap to do and almost nobody is doing them yet.
How an assistant actually decides who to name
Simplifying, but not much, there are two gates.
The first gate is being in the pool at all. When you ask an assistant for a recommendation it either draws on what it absorbed during training, or it runs a live search and reads the results, or both. Either way there is a retrieval step that assembles a set of candidate sources. If your site is not crawlable, or nothing anywhere says clearly what you do and where, you are not in the pool and nothing else you do matters.
The second gate is being the thing it picks out. Having read the candidates, the model composes an answer. It favours sources that state a specific, checkable claim in a form that can be lifted cleanly: a price, a location, a service, a limitation. Pages that circle a subject atmospherically for eight hundred words give it nothing to quote, so it quotes somebody else.
That is the whole game. Get into the pool, then be the most quotable thing in it.
What you can actually control
Let the AI crawlers in
The assistants use named crawlers, and a lot of sites block them without realising, either through a security plugin, a CDN setting, or a robots.txt copied from somewhere. Check that GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended and Applebot-Extended are not disallowed in your robots.txt, and that your firewall is not quietly rejecting them either.
There is a real decision here rather than an obvious one. Allowing these crawlers means your content may be used in training and in answers, sometimes without a click coming back to you. Blocking them protects the content and removes you from the answers entirely. For a small business trying to get found, being cited is worth more than the content is. For a publisher whose product is the words, the sum works differently.
Add an llms.txt
An llms.txt is a plain-text file at the root of your site that states, in ordinary language, what the business is, who it serves, what it charges and where the important pages are. It is a young convention rather than a standard, and no assistant is obliged to read it.
It is still worth twenty minutes, for a reason that has nothing to do with whether the file is fetched: writing it forces you to state your own facts plainly in one place, and those sentences are exactly what you then need on the pages themselves. Include the awkward parts. A line saying who you are not right for reads as credible and gets reproduced surprisingly often.
Write answer-first
Put a short, self-contained paragraph near the top of every important page that answers the obvious question completely, with the specifics in it. Not a teaser. The actual answer, including the price and the limitation.
The test is whether that paragraph, pulled out and read alone with no surrounding page, still makes sense and still says something checkable. If it does, an assistant can quote it. If it only makes sense in context, it will not be used. Every page on this site has one of these near the top, including the one you are reading.
Publish real numbers
This is the single biggest differentiator and most businesses refuse to do it. When somebody asks an assistant "how much does a website cost in Kent", the businesses that published a price get named and the ones hiding behind "contact us for a quote" do not, because there is nothing to say about them.
A range with an honest explanation of what moves it works perfectly well. You do not have to publish a single number. You have to publish something a model can repeat.
Be consistent everywhere else
Models weight corroboration heavily. If your website says you cover Sussex, your Google Business Profile says Brighton only, and three directories have an old address and a dead phone number, you look uncertain, and uncertain businesses do not get recommended.
Make the name, address, phone number, service list and areas covered identical everywhere they appear. This is the same unglamorous work as local SEO, which is convenient, because you should be doing it anyway.
Get mentioned on sites that are not yours
This is the uncomfortable one. What an assistant knows about your industry comes mostly from sources that are not your website: trade bodies, local press, directories, forums, review platforms, supplier listings, community sites. A business that exists only on its own domain has one source vouching for it, and that source is you.
There is no trick here. Get listed in the real directories for your trade. Be findable in local press when you do something worth reporting. Answer questions in places where your customers already are. It is slow, and it is the part that compounds.
Use structured data properly
Schema markup in JSON-LD gives machines an unambiguous statement of what your business is, what it offers and what it costs. Organization or LocalBusiness, Service with priced offers, and Article on anything you publish covers most of it.
Two warnings. FAQPage rich results have been restricted to government and healthcare sites since 2023, so do not let anyone sell you FAQ schema as a ranking tactic on a commercial site. And HowTo rich results were removed entirely. Structured data is worth doing for clarity; it is not a magic ingredient.
What nobody can promise you
Be sceptical of anyone selling guaranteed AI visibility. Here is why it cannot be guaranteed.
- There is no ranking to buy and no form to submit. No inclusion process exists. Anyone claiming a relationship that gets you listed is describing something that does not exist.
- The answer is not stable. Ask the same assistant the same question twice and you can get different businesses. There is no position four to occupy.
- It varies by user. Location, phrasing, conversation history and which model version is serving the request all change the answer.
- Measurement is genuinely poor. You can see referral traffic from assistants in analytics, and you can test prompts by hand and log the results. There is no equivalent of Search Console. Anyone showing you a precise AI ranking report is showing you a sample of one, dressed up.
What you can do is check by hand, monthly. Write down ten questions a customer might ask, put them to ChatGPT, Gemini and Perplexity, and record who gets named. It is crude and it is honest, and it is better than the dashboards being sold.
The order I would do this in
| Step | Effort | Why it comes here |
|---|---|---|
| Check AI crawlers are not blocked | 20 minutes | Nothing else works if you are invisible. Free, and frequently the whole problem. |
| Fix Google Business Profile and directory consistency | A few hours | Corroboration across sources is heavily weighted, and it helps ordinary local search at the same time. |
| Add answer-first paragraphs to your main pages | A day | This is what actually gets quoted. Highest return per hour of anything on the list. |
| Publish prices, or honest ranges | An afternoon, plus a decision | The most common reason a business is not named. The hard part is the decision, not the work. |
| Write an llms.txt | An hour | Cheap, and forces you to state your facts plainly. |
| Add or fix structured data | Half a day | Clarity for machines. Useful, not magic. |
| Earn third-party mentions | Ongoing, forever | Slowest, compounds the most, cannot be shortcut. |
Does this replace normal SEO?
No, and treating it as a separate budget is how people waste money on it. Look again at that list: crawlability, consistent listings, clear structure, real content, third-party mentions. That is ordinary technical and local SEO with the emphasis moved.
The genuine additions are answer-first writing, publishing specifics you might have kept back, and llms.txt. Everything else is the same work you needed anyway, which is the good news. A site built properly for Google is already most of the way to being quotable, and a site that is invisible to Google is invisible to assistants for the same reasons.
One realistic note on timing. Broad questions can start naming a new site within days, because assistants running a live search are not waiting for the authority that Google's rankings require. Consistent appearances on competitive questions take months and depend mostly on the third-party mentions, which is the slow part. Anyone promising fast results on the competitive questions is selling you the easy half.
If you are weighing up whether to hire for this at all, I have compared the options honestly in choosing an AI SEO provider in the UK, including the cases where the answer is to buy a tool or do it yourself.
If you want to know why an assistant currently recommends somebody else in your town, send me the question you asked and your URL. I will tell you what they have that you do not. Ask me.