How to get your business recommended by ChatGPT and Google's AI answers

ChatGPT can use information learned during training and current information retrieved through search. Keep business details on accessible, up-to-date pages, and check the sources cited in answers about your company. OpenAI explains how ChatGPT Search works.

For the search half, OpenAI runs its own crawler, OAI-SearchBot, and sends rewritten queries to outside search providers. Its help page links the privacy policies of two: Microsoft and Shopify.

Microsoft runs Bing, which is why registering a site in Bing Webmaster Tools is the first thing I do on any AI-search project.

Yes. Google says any page that is indexed and eligible to show a snippet in ordinary Search can appear as a link in AI Overviews and AI Mode, with no extra technical requirement. Both features can run several related searches across subtopics before answering, which Google calls query fan-out, so the links come from more results than one search returns.

Three settings catch people out. A nosnippet or max-snippet rule also limits what the AI features can show from a page. Since 31 August 2026 every Search Console property has had a Search generative AI control under Settings. Include is the default; Exclude takes your links and content out of AI Overviews and AI Mode, so check that nobody has switched it. The third is Google-Extended. People block this robots.txt token to keep content away from Google's AI, but it is not a crawler and changes nothing in Search: Google's crawler documentation says it affects neither inclusion nor ranking. It governs Gemini's training and whether the Gemini app can ground its answers in your pages.

Google's guide to its generative AI features names two business tools that help products and services show up in AI responses: Merchant Center feeds and the Business Profile. The profile has a guide of its own.

AI crawlers can read a site when three layers let them through: robots.txt, the bot settings of whatever sits in front of the site, and the HTML each page sends. Each AI company documents a search crawler that decides whether your pages appear in its answers. At OpenAI and Anthropic the training crawler is a separate setting, and blocking it leaves search alone.

One robots.txt rule trips up careful people. Under the robots.txt standard, RFC 9309, a crawler that finds a group with its own name obeys only that group, so a group allowing OAI-SearchBot frees it from every Disallow line under User-agent: *; repeat those lines in each named group. And check any block list pasted in from a forum: one that names OAI-SearchBot, PerplexityBot or Claude-SearchBot keeps your pages out of those assistants' search answers.

Then the layer in front of the site. OpenAI asks site owners to check that their host or CDN accepts traffic from its published crawler IP addresses, and Google puts the same check first among its best practices for AI features: a firewall, bot filter or security plugin that turns a crawler away overrides anything robots.txt allows.

Last, the text itself. Vercel's December 2024 study caught OpenAI's crawlers and Anthropic's ClaudeBot downloading JavaScript files, 11.50% and 23.84% of their requests, and never running them; PerplexityBot did not run them either. Googlebot, whose infrastructure Gemini shares, renders pages. Nearly two years on, the crawler pages of OpenAI, Anthropic and Perplexity still say nothing about JavaScript, so I build as if nothing has changed: prices, address and services go into the HTML the server sends. The article on a website that isn't showing up on Google has the one-minute check in Chrome.

Type the search ChatGPT would write into Bing and Google, “best dentist Manchester” rather than the question a patient asks, and note what comes back: list articles, directories, review sites, local press, a trade body's member search. Those results are the nearest public view of what ChatGPT reads before it answers. Getting your business onto them is most of the work.

Ahrefs' numbers point the same way. In its study of 75,000 brands, published in May 2025, the quarter mentioned most often around the web, with or without a link, averaged 169 appearances in Google's AI Overviews. The quarter below averaged 14. And 26% of the brands never appeared at all. A December 2025 follow-up measured ChatGPT directly: there, web mentions correlated with visibility at 0.664 and YouTube mentions at about 0.737.

Where I would start, in this order, for a business whose customers are nearby:

• The Business Profile and every other listing. Same name, address, phone number and hours everywhere: an assistant that finds two phone numbers has to pick one.
• The platforms your trade books through. For a restaurant, OpenTable comes first: ChatGPT shows its reservation times in every country, while Resy and Yelp are limited to North America.
• The “best of” lists that rank for your search; find who updates each one and send them your details and photographs.
• Trade bodies, suppliers and partners: member directories, stockist pages, case studies. Local and trade press too, when there is real news.
• Forums and Reddit, where people ask for recommendations in their own words. A partnership announced in May 2024 gave OpenAI access to Reddit's Data API to bring Reddit content into ChatGPT; answer as yourself, under your own name, where the question is about what you do.
• YouTube: a short video of the work, with your name said aloud and written in the description.

Google's AI guide warns that chasing inauthentic mentions helps less than it seems, since other Google systems exist to block spam. A mention counts when a person chose to write your name.

For a shop, product data does the job listings do for a café. ChatGPT builds product results from structured data supplied by merchants and third parties, and OpenAI's shopping help says they are not ads. Shopify stores are already connected through Shopify Catalog; other merchants can apply to send OpenAI a product feed.

The timescales are uneven. A robots.txt fix is read within about a day. The lists, the press and the forum answers take months, and I plan three to six months before a local business turns up for a category search in its own town, longer where the lists are already crowded. It is the part of SEO that runs longest.

The evidence has limits. OpenAI does not guarantee placement, and the Ahrefs figures describe correlation rather than cause. Treat them as context for testing, not a formula for securing a recommendation.

Each service you want to be recommended for needs a page of its own that says, in plain text near the top, what you do, where, for whom and how the price works. OpenAI's example searches run to three to five words, and the page that repeats the customer's words in its first lines is the easiest to quote.

Google's guide says nobody needs to write in a special way for its AI features. Taken literally, that is right. I still put a question-shaped heading over each section with the answer in the two sentences under it, because that is how people read on a phone, thumb on the screen, skipping everything that isn't the answer. Opening hours, the towns you cover and starting prices belong in the text; in a photo of the menu board they are not text at all. When we built SnapQio, a screen-capture tool for Windows, and the site that sells it, the selling points went onto the page in the words a buyer would use: “No account needed”, “works offline”, “x64 + ARM64”. Each is a condition someone can add to a question for an assistant, and each has a line on the page that answers it.

One more figure from Vercel's study, about redesigns rather than writing: 34.82% of ChatGPT's fetches and 34.16% of Claude's ended on a 404 error page. Redirect every old address.

Organization structured data helps Google tell your business apart from others and takes about an hour to add, but in a 2026 Ahrefs study pages that added structured data saw no measurable rise in AI citations. Google Search ignores llms.txt, and no crawler page from OpenAI, Anthropic or Perplexity says their search crawler reads it.

Google's Organization markup goes on the home page or the about page and takes the name, logo, address, phone number and sameAs links that tie the site to its listings. When Ahrefs tracked 1,885 pages that added structured data between August 2025 and March 2026 against about 4,000 that did not, citations rose 2.4% in AI Mode and 2.2% in ChatGPT, both within the noise, and fell 4.6% in AI Overviews. Those pages were already cited heavily, so the study says little about a site nobody cites yet.

llms.txt, proposed by Jeremy Howard in September 2024 and revised in August 2026, is a Markdown summary of a site for language models. Chrome's Lighthouse checks for one only to flag a server error, and a site without the file is marked not applicable. OpenAI, Anthropic and Perplexity each keep one for their developer documentation. For a business site, I would prioritise accessible pages and accurate content before commissioning a separate file.

In GA4, open Reports, then Acquisition, then Traffic acquisition, and find the AI Assistant row. Google added the channel on 13 May 2026 for visits from assistants such as ChatGPT, Gemini and Claude, and its definition also names Copilot, DeepSeek and Grok. Clicks from Google's own AI Overviews and AI Mode are not in that row; GA4 counts them as Organic Search.

ChatGPT adds utm_source=chatgpt.com to the links it cites, OpenAI says, so its visits carry a label in any analytics tool. Add Session source as a secondary dimension to see which assistant sent each visit. Where one is missing from the row, a custom channel group catches it with a regular expression and applies to past data too; Google's own example pattern includes Perplexity.

Search Console's Generative AI performance report, which Google says it rolled out to every website on 31 August 2026, counts impressions in AI Overviews and AI Mode by page, country, date and device. Bing goes further. Its Webmaster Tools added an AI Performance report in February 2026 that counts citations in Microsoft Copilot and Bing's AI summaries and lists a sample of grounding queries, the phrases the AI searched with when it used your page. Those queries are as close as anyone outside Microsoft gets to seeing how an assistant rewrote a customer's question.

None of these reports counts the customer who reads an answer, remembers your name and types it into Google the next day. Add “ChatGPT or another AI assistant” to the options under “How did you hear about us?” on your enquiry form, and once a month ask ChatGPT, Gemini, Perplexity and Copilot the five questions your customers ask, signed out, noting who gets named and which pages are cited.