ChatGPT recommends businesses it can clearly identify and match to the question. State what you are, where you are and who you suit in a form machines can read; answer the specific questions customers ask on your own pages; keep your details identical everywhere; get onto the sources the assistant trusts for your area; then ask the same questions every month to see what changed.
How ChatGPT decides who to recommend
When someone asks ChatGPT a local question with search switched on, such as "best seafood restaurant in Dubai Marina" or "family hotel near Bondi Beach", it doesn't consult a hidden ranking of businesses. It runs one or more web searches, reads a handful of the pages that come back, and writes an answer from what those pages say. Other assistants, including Gemini, Perplexity and Google's AI Overviews, work in broadly the same way.
That has two practical consequences. First, you can only be recommended if the pages the assistant reads mention you. Second, those pages need to say something that matches the question: "has a play area", "dogs welcome", "ten minutes from the convention centre". Being good isn't enough. The evidence has to be on the page.
Step 1: Find out what it says today
Write down ten questions a customer would ask before choosing a business like yours, in their own words. Mix broad questions ("best boutique hotel in Dubai Marina") with specific ones ("hotel in Dubai Marina with a rooftop pool and parking"). Then ask ChatGPT each one with search on, and ideally Gemini and Perplexity too.
For every answer, note four things:
- whether you were named;
- your position in the list;
- who was named instead;
- which websites the assistant cited as sources.
Ask each question more than once. Answers vary between runs, so one answer is an anecdote and five are a pattern. This baseline is what you'll measure every later change against. Our live check runs a single question for you; the full free check runs ten across five assistants.
Step 2: Give machines the facts
Structured data is a small block of code, usually JSON-LD, that tells machines exactly what your business is. A hotel should describe itself as a Hotel with its address, map location, phone, star rating, check-in times and amenities. A restaurant should be a Restaurant with its cuisine, price range, opening hours and menu. A clinic, salon or firm uses LocalBusiness or one of its more specific types.
Many websites output only generic page data from an SEO plugin, which says nothing useful about the business. Search engines use structured data to understand pages, and assistants that search the web benefit from that understanding. Our hotel schema guide has a complete example you can copy.
Step 3: Answer the questions you're losing
Go back to your baseline. For each question where a competitor was named and you weren't, read what the assistant quoted from their page. Usually it's a plain, specific statement: "children under 12 stay free", "private dining for up to 20", "free parking on site".
Then add a short, specific answer to your own page. Two or three sentences under a clear heading is enough. Avoid marketing language; write the way you'd answer a guest at the front desk. If the information lives only in a PDF, a photo or your booking engine, the assistant probably can't use it.
Step 4: Make the site fast and readable
Keep the pages that matter light. The main content should appear within 2.5 seconds on a phone; Google calls anything over four seconds poor. Large photo sliders and background videos are the usual culprits.
Check that your robots.txt doesn't block the crawlers AI services use, such as OpenAI's OAI-SearchBot and GPTBot, Anthropic's ClaudeBot and PerplexityBot. Some security plugins and content-delivery settings block them by default. An llms.txt file that summarises your business and links to your key pages is a cheap extra.
Step 5: Keep your details identical everywhere
Use exactly the same business name, address, phone number and description on your website, Google Business Profile, booking and review platforms, and directories. Old phone numbers, previous names and different opening hours make it harder for any system to be sure that all these pages describe the same business.
Groups need to be explicit. If a hotel has a restaurant with a different name, or a pub was renamed last year, say so on the website, in plain words and in the structured data.
Step 6: Get onto the sources the assistant uses
The websites cited in your baseline are your target list. For local businesses they tend to be the same few kinds of site: city or tourism guides, food and travel publications, award lists, review platforms and booking sites.
If a guide is cited again and again for your area and you're not in it, that's often the single biggest gap. Approach it the way a journalist would want: a clear angle, real facts, good photos. Never buy fake reviews or paid "citations"; besides being against most platforms' rules, it tends to backfire.
Step 7: Measure monthly, and judge the pattern
Ask the same questions, the same way, every month, and record the same four things. Because answers vary, look for the trend across questions and months: more answers naming you, better positions, your own website appearing among the sources.
Website changes are usually live within weeks. Assistants and the search indexes behind them pick changes up at different speeds, so give it two or three monthly checks before deciding what's working.
Mistakes to avoid
- Publishing thin pages written only for machines. They rarely help, and they don't help your customers.
- Judging from one answer. Run each question several times.
- Ignoring your own website in favour of listings. Listings help, but your site is the one source you fully control.
- Expecting ads to help. Paid search placement doesn't make an assistant recommend you.