How to get your business recommended by ChatGPT

Ask an AI assistant for a shortlist of businesses like yours. If your name is missing, the tempting response is to buy an "AI visibility" package. First check whether your own website answers the questions a customer would use to build that shortlist.

Who do you help? What does the service include? Where do you operate? What can a buyer verify? A homepage that says "we deliver excellence" leaves all four unanswered.

The demand is real, with a geographic caveat. BrightLocal's 2026 survey of 1,002 US adults found that 45% had used AI tools for local business recommendations in the previous 12 months. That describes US consumers, not adoption in India or the UAE. BrightLocal's findings.

What you can and can't control#

OpenAI says ChatGPT search uses multiple factors to find relevant, reliable information and doesn't guarantee placement. OpenAI's ChatGPT search guidance.

I'd start with the pages and records a customer can inspect today. Those improvements help a buyer even when an assistant chooses another source.

There are also different outcomes to measure. An answer might name your company, cite a page explaining a topic, or link to your service as an option. A citation of your article on appointment booking doesn't mean the assistant recommends you to build the reader's booking system.

Search access and training permission are separate decisions#

Blocking every AI crawler can also block search discovery; allowing every crawler may grant more than you intended. OpenAI and Anthropic document distinct agents with different purposes. OpenAI crawlers, Anthropic crawlers.

A site that wants ChatGPT search access while declining OpenAI's training crawler could use these rules, after reviewing the rest of its robots.txt:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

That's a permissions example, not a complete configuration. Private customer records still need authentication and access controls.

Check the hosting layer too. A robots.txt allowance can't get past a firewall or bot challenge that blocks the request, and OpenAI recommends allowing its published search-bot IP ranges alongside crawler access. Verify real traffic against the current documentation before making exceptions.

Give each important service a page with usable details#

Imagine an accounting firm that offers bookkeeping, GST filing and company incorporation. Its homepage lists them, but every button opens the same enquiry form, so a customer comparing GST providers still has to call to learn what the package covers.

A useful GST service page would explain:

• which kinds of businesses the firm accepts;
• the filings and ongoing support included;
• the documents needed to begin;
• how fees are calculated, with real figures if the firm approves them;
• the region served and whether work happens remotely;
• who reviews the work and how a buyer can verify their qualifications.

These are recommendations for a clearer buying decision, not a published list of ChatGPT ranking factors. Be precise about exclusions too: if a package covers routine filing but excludes a tax dispute, say so.

I prefer one substantial page per real service to dozens of near-identical pages for different keyword spellings. Google's AI search guide encourages useful, distinctive content and warns against producing variations mainly to manipulate rankings or generated answers. Google's generative AI guidance.

Publish evidence with enough context to mean something#

An unsupported "40% improvement" makes a weak case study. What improved? Over what period? Compared with which baseline? Was the system live, or was someone estimating the benefit of a proposal?

For a real project, document the buyer's problem, the change you made and the limits of the result, linking to approved evidence where you can. For a prototype, explain what it demonstrates and what remains untested. Keep client identities private until the client authorises publication.

A labelled worked example can still teach. A booking business taking a $5 deposit on 20 appointments collects $100 in deposits. That arithmetic says nothing about reduced cancellations unless someone measures cancellations before and after.

Put the accountable person on the page. A named author can explain why they prefer an approach, what they assumed and when they last checked a changing rule. Personality helps when it shows judgement; an invented anecdote supplies no evidence.

Keep local business information accurate#

If you're eligible for a Google Business Profile, check it against the website: the actual business name, service categories, contact details and hours. Describe real service areas and premises, and never add a fictitious office to look local.

Google documents relevance, distance and prominence as factors in its local results. That describes Google's local results only. Complete, accurate records still help people checking an AI-generated shortlist. Google's local ranking guidance.

Look beyond your website for contradictions. An old directory may list a disconnected number; an association profile may mention a service you've dropped. Fix the records you control and request corrections to consequential errors elsewhere.

For a store, product data needs the same care: descriptive names, current availability and explicit delivery and return policies. If you want transactions inside AI tools, read our guide to Shopify and AI shopping channels; discovery and a working purchase integration are different jobs.

Use structured data to describe what's on the page#

Structured data can describe your organisation, products or articles in a standard format. Keep it consistent with the visible content: no hidden ratings, invented addresses or prices readers can't find.

Google says its AI search features need no special schema markup and no special AI text files. An llms.txt file is therefore no substitute for a useful service page. Those statements concern Google's systems, not every AI product. Google's AI features documentation.

Before buying a technical add-on, ask what documented problem it solves and how the vendor will show that it works on your site.

Build a repeatable prompt check#

One favourable screenshot tells you what one answer said. For a more useful check, keep a small set of customer questions and run them on a schedule. This is a proposed method, not research we've already run.

Start with eight questions across your main services, including location and buying constraints. Test in fresh conversations and note whether web search ran. Save the exact wording, the date, the assistant, your location settings if known, the answer and the links it cited. Try this without your company name:

Search the web for businesses that provide [specific service] for
[customer type] in [location]. I need [constraint]. Compare suitable
options using current public information. Link to the sources supporting
each option and distinguish confirmed facts from details I need to ask
the provider about.

Then inspect the sources yourself. Does your business appear? Is the description correct? Does the answer confuse you with another company? Does a cited competitor page answer a buyer question your site ignores?

Report results at the size of the sample. Eight prompts run twice give 16 answers; if you appear in five, write "5 of 16 answers in this prompt set mentioned us", not "a 31% share of AI search".

Measure enquiries alongside visibility#

OpenAI's publisher guidance says ChatGPT referral links include utm_source=chatgpt.com. Use that in your analytics, then look at what those visitors do: do they find the right service page and send a suitable enquiry? OpenAI publisher FAQ.

Google now documents a Search Console report for generative AI performance; its 3 June 2026 announcement says it rolled out worldwide by 31 August 2026, with impressions broken down by page, country and date. Google's announcement.

Keep those numbers separate from your prompt checks and sales records. Someone can read an answer without clicking, or hear your name from a colleague first, so ask new leads how they found you and allow more than one answer.

For a small business, I'd review this monthly: fix access failures first, then inaccurate facts, then unanswered buyer questions. Keep earlier results so you can compare without claiming causes too early.

Make the first improvement concrete#

Choose one valuable service and walk the whole path. Can a search system reach the page? Can a buyer identify the offer and verify the provider? Can they contact you without repeating what they already told you?

If your website needs work on those questions, talk to us. Share the service page and the kind of enquiry you want it to attract. We'll work on the page, its evidence and its route to an enquiry, without promising a position in someone else's search results.

Sources#

Checked 4 October 2026.

• BrightLocal, 2026 research on AI business recommendations.
• OpenAI, ChatGPT search, crawlers and publisher FAQ.
• Anthropic, Claude crawler purposes and controls.
• Google, generative AI search guidance, AI features, local ranking and the AI performance report.