How to get your business recommended by ChatGPT (and what you can’t control)
No one can make an assistant recommend you. Here is where they look, what you can check this week, and what you can safely ignore.
Perhaps a new client mentions that they asked ChatGPT who to call. Or you try it yourself one evening. You type the question your customers would ask, and you read the answer. Three names. None of them yours.
Here’s the honest answer first. No one can make ChatGPT, or any other AI assistant, recommend your business. Not an agency, not me, not you. As of October 2026, OpenAI says placement in ChatGPT’s search results isn’t guaranteed. What you can do is smaller and more useful. ChatGPT, Google’s AI answers, Claude, Perplexity and Copilot can all search the web before they answer, and they work from what they find. You can make sure they can reach your website. You can make your site say plainly what you do. You can make sure the rest of the web agrees with it. And before any of that, you can check what they say about you today.
That’s the order I’d follow. Look first. Then fix what the evidence points to.
Where the assistants get their answers.
An assistant can answer from what it learned in training. Or it can search the web first and write its answer from the pages it finds. The second route is the one that matters to you, because that’s when a current page about your business can be read and used.
To search, each company runs a crawler: a program that visits websites and reads them for its search service. Some also run a separate crawler, or offer a setting, for collecting pages to train their models. Different job. That difference matters later.
Here’s which door is which, as of October 2026.
| Assistant | Where it looks | What controls access | What the company says about choosing |
|---|---|---|---|
| Google AI Overviews and AI Mode | Google’s own search index, rooted in the same core ranking and quality systems as ordinary Google Search | Googlebot. The page has to be indexed and eligible to show a snippet, and your Search Console AI setting has to stay on “include”, which is the default | Google says there are no extra requirements and no special optimisation. Useful, original content matters most (Google’s guide1) |
| Gemini app | Can draw on Google Search | Not documented in the sources I checked | Nothing published that I found |
| ChatGPT | ChatGPT search, which OpenAI says uses third-party search providers it doesn’t name | OAI-SearchBot. Block it and your site won’t be shown in ChatGPT’s search answers. GPTBot is the separate training crawler (OpenAI’s crawler page2) | OpenAI says it ranks results on several factors aimed at relevance and reliability, and that placement isn’t guaranteed |
| Claude | Web search, when it searches | Claude-SearchBot. Anthropic says blocking it may reduce how visible and accurate your site is in search results. ClaudeBot is the training crawler (Anthropic’s crawler page3) | Nothing published that I found |
| Perplexity | Web search | PerplexityBot, which Perplexity says is for its search results and not for training models (Perplexity’s crawler page4) | Nothing published that I found |
| Microsoft Copilot and Bing | Bing | Bingbot, Bing’s crawler | Bing’s advice: depth and expertise, clear structure including FAQ sections, fresh content, and current Bing Places details |
Each company’s advice applies to its own product. Assuming ChatGPT works the way Google describes would be a guess. And look at the last column. For three of the six, I found nothing published about how they choose which business to name. Anyone who tells you exactly how ChatGPT picks a dentist is telling you more than OpenAI has said publicly.
You’ll still read that ChatGPT doesn’t search the internet in real time. That’s out of date. OpenAI documents both ChatGPT’s web search and the crawler it uses for it.
First, ask them what your customers ask.
This is the step most worth doing properly, because done badly it misleads you.
1. Write down five to ten real customer questions. Use the words customers use, not the words in your brochure. Think about what people ask before they choose. A dental clinic in Cork, say, might write “Which dentist in Cork is good with nervous patients?” A small hotel in the Lake District might write “Quiet hotel near Keswick that takes dogs.” Add one question with your own name in it, such as “What is [your business] known for?”, so you can see how you’re described.
You’re done when the list is written down and someone else in the business agrees those are the questions customers really ask.
2. Ask each question in two or three assistants. Choose the ones your customers mention, or ChatGPT and Google’s AI answers if you don’t know. Type the question exactly as written. Don’t rephrase it to help yourself.
3. Note where you’re asking from. As of October 2026, OpenAI says ChatGPT may use an approximate location, based on your internet connection, for local questions. If you ask from home and your customers are in another city or country, you may not see what they see. Put the place in the question, as in the examples above. I’d also note whether you were signed in, so you can compare like with like next time.
4. Record every answer. A simple spreadsheet is enough. For each answer, note:
- the date, the assistant and the question;
- whether you were named;
- whether you were described correctly (services, location, hours, who you’re for);
- which sources the assistant showed, if any;
- who was named instead.
5. Repeat it on two or three different days. This sounds like overkill. It isn’t. In a study by SparkToro and Gumshoe, about 600 US volunteers ran the same 12 recommendation prompts through ChatGPT, Claude and Google’s AI nearly 3,000 times at the end of 2025. They found less than a 1 in 100 chance5 that ChatGPT or Google’s AI would give the same list of brands in any two answers. The authors note the study’s limits, and Gumshoe sells AI tracking. One answer tells you very little. How often you appear across many answers tells you more.
You’re done when you can say something like “named in this many of that many answers” for each assistant, and whether the descriptions were right. A hypothetical example: eight questions, two assistants, three days. That’s 48 answers. Named in five, all in one assistant, twice with the wrong opening hours. That’s a result you can act on.
What your results are telling you.
Look for four patterns. Each points somewhere different.
You weren’t mentioned at all. Start with access. If the assistants can’t read your website, nothing else you do on it counts for much. Then look at whether your own site says what you do.
You were mentioned, but described wrongly. Wrong details come from somewhere. Check the sources the assistant showed. It might be an old page on your own site, a directory listing nobody updated, or a profile with last year’s hours. Fix it at the source.
Someone else was named instead. Open the sources shown next to their name. Compare what they publish with what you publish. Do they have a page that answers the exact question? A profile with current details? Reviews that describe the thing the customer asked about? That tells you which of the next sections to work on.
You appear in one assistant and not another. Go back to the table. Different assistants use different search services and different crawlers. A site can be open to one and closed to another without anyone meaning it to be. Start with access.
Sometimes the answer names no business at all, only a directory or a booking site. That’s useful too. It tells you where the assistant looks for that question.
Can they reach your website?
This is a job for whoever looks after your website. Ask them to check three things and give you the answers in writing.
First, the robots.txt file. Most websites have a small file at yourdomain.com/robots.txt that tells crawlers which parts of the site they may visit. Ask that it doesn’t block the search crawlers: Googlebot, Bingbot, OAI-SearchBot, Claude-SearchBot and PerplexityBot. OpenAI says changes to this file take about a day to reach its crawler.
One warning before anyone edits it. You may read advice to “allow GPTBot” so that ChatGPT can recommend you. GPTBot is OpenAI’s training crawler. The one that affects ChatGPT’s search answers is OAI-SearchBot. Training crawlers (GPTBot, ClaudeBot and Google’s Google-Extended setting) are a separate choice, and OpenAI’s documentation says a site can allow its search crawler while blocking its training one. Whether your pages are used for training is your decision. It isn’t the same decision as being findable.
Second, the Search Console setting. Search Console is Google’s free dashboard for website owners. As of 31 August 2026, every site has a setting for Google’s generative AI features6. It’s on by default. Switching it off removes your pages from AI Overviews and AI Mode, though Google says it doesn’t affect your ranking elsewhere. Ask your web person to confirm nobody has changed it, and that your important pages are indexed and allowed to show a snippet. Google says a page needs both to appear as a link in its AI answers.
Third, the words themselves. This is my own rule of thumb, not something the companies publish. I’d want the important facts written as ordinary text on the page: what you do, where, your hours, how to book. Not only inside an image, a menu PDF or a brochure download.
You’re done when your web person has confirmed that the search crawlers are allowed, the Search Console setting is on “include”, and your main pages are indexed.
Does your own site say what you do?
An assistant can only repeat what it can find. If your homepage says “bespoke solutions for discerning clients”, there’s very little to repeat.
You know what the business does and why people choose it. The website often says less than you think. That’s the gap between how you see the business and how customers meet it, and assistants meet it from the outside too.
Google’s guide says content people find unique, compelling and useful is likely to shape your presence in its AI search more than any of its other suggestions. It asks for first-hand content, not the generic text any business in your sector could publish.
So check that your site says, in plain words:
- what you do, in the words customers use;
- who it’s for, and who it isn’t for;
- where: the towns or areas you serve, and whether you travel;
- how it works: what happens at a first appointment, how booking works, what happens next;
- proof: real reviews, qualifications, projects, the people who do the work.
Then go back to your customer questions. For each one, can you point to a page that answers it?
The answers that help most are the ones only you can give. A clinic can say how it actually handles nervous patients. A hotel can say how far it is from the station and which rooms are step-free. That’s the first-hand detail Google says it wants, and nobody else can write it for you.
One limit. Don’t respond by producing dozens of near-identical pages for every variation of every question. Google warns that mass-producing pages to catch AI search queries can breach its spam policy on scaled content.
Does the outside evidence agree?
Assistants don’t only read your website. If the rest of the web disagrees with your site, an assistant may repeat either version.
Start with the profiles the search companies run themselves. Google says a Google Business Profile can help your services be visible in its AI responses as well as ordinary search results. Bing advises local businesses to keep their Bing Places details (address, hours, contact) current. Check both are claimed, accurate and complete.
Then everything else. A directory listing with an old phone number. A professional body that lists your previous address. A booking platform that describes rooms you no longer offer. The aim is the same name, address, phone number, hours and services everywhere.
Reviews are part of this evidence. I haven’t found OpenAI, Anthropic or Perplexity saying that reviews decide which business gets named. But reviews are public, they describe what it’s like to be your customer, and an assistant can find them. Genuine reviews, and thoughtful replies to them, are worth having for their own sake.
Real mentions count too: partners who link to you, local press, professional bodies. A 2025 vendor study by Ahrefs, across 75,000 brands, found that how often a brand was mentioned across the web7 tracked its appearance in Google’s AI Overviews more closely than links to its site did. That’s a correlation, not proof of cause, and the sample was larger brands.
Genuine is the important part. Google lists chasing inauthentic mentions among the things that aren’t as helpful as they might seem. Paid “best of” lists and fake reviews are evidence of the wrong kind.
You’re done when your core details match on your site, your Google Business Profile, Bing Places and the main directories in your sector.
What you can ignore.
A lot of what’s sold under this heading isn’t needed. Each point is scoped to the company that said it.
- An llms.txt file, for Google. Google says its search doesn’t use llms.txt or similar AI text files. What the other assistants do with it isn’t documented in the sources I checked.
- Special “AI schema” markup, for Google. Google says structured data isn’t required for its generative AI search and there’s no special markup to add. Bing, separately, does recommend clear structure, including FAQ sections. So FAQs on your site are fine. Just don’t expect a magic tag.
- Chunking or rewriting your site for AI. Google lists both among the things you can ignore.
- Anyone who guarantees an AI ranking or recommendation. OpenAI says placement isn’t guaranteed. Google says no special optimisation is needed. A guarantee promises something the companies themselves don’t.
- Paying to be inside the answer. OpenAI began testing ads in ChatGPT in early 2026, starting in the US. OpenAI says the ads are labelled, separate from the answer, and don’t influence it. I couldn’t confirm which countries they run in as of October 2026. Nothing I found from OpenAI or Google offers a way to pay to be placed inside the answer itself.
You’ll also meet the term GEO, short for generative engine optimisation. It came from a 2024 academic paper that tested changes to web pages in a lab benchmark, not real assistants recommending a clinic or a hotel. If someone quotes its numbers to you as a promise, be careful.
How to tell if anything changes.
Change will be hard to read at first, because answers vary so much. Plan to watch over months, not days.
- Repeat your check roughly once a month: same questions, same assistants, same place. Count how often you appear.
- Look at Search Console’s generative AI report, available to all sites since 31 August 2026. It shows impressions in AI Overviews and AI Mode by page, country, device and date, though it may not appear if your site has very few impressions.
- Look at Bing Webmaster Tools, Bing’s free equivalent of Search Console. Its AI Performance report8, in public preview since February 2026, shows how often your content is cited in Copilot, Bing’s AI summaries and some partner services.
- Check your analytics. As of October 2026, OpenAI says ChatGPT adds
utm_source=chatgpt.comto the links it sends people through, so those visits can show up labelled. - Ask new customers how they found you. It costs nothing. Write the answer down.
What nobody outside these companies knows yet.
Some of this I can’t tell you, because it isn’t published.
- Which search providers ChatGPT uses. OpenAI says third-party providers. It doesn’t name them.
- How ChatGPT, Claude, Perplexity and the Gemini app decide which business to name.
- Whether reviews, the number of reviews or directory listings change the answer in any assistant. Google and Bing advise on their own profiles, and there are correlation studies. There’s no published proof of cause.
- Which countries see ads in ChatGPT today.
- How long a change on your site takes to show up in answers, if it does.
Everything here is as of October 2026. These products change, so check the date on anything you read about them, including this.
Look first. Then decide.
None of this guarantees a recommendation. What it does is make the true picture of your business easy to find, easy to read and consistent wherever an assistant looks. That’s worth doing anyway. It’s the same picture a careful customer sees.
If you’d like a second pair of eyes, I can run this check from the outside and show you what I find, what it’s based on and how sure I am. It’s the start of how I research a business from the outside. You decide whether any of it matters, and there’s no obligation. Ask me to look.
Sources in this note
- Google Search Central Google’s guide
- OpenAI OpenAI’s crawler page
- Anthropic Anthropic’s crawler page
- Perplexity Perplexity’s crawler page
- SparkToro Less than a 1 in 100 chance
- Search Console Help Setting for Google’s generative AI features
- Ahrefs How often a brand was mentioned across the web
- Bing Webmaster Blog AI Performance report