Step 1. Check that crawlers can reach your site at all
First, because if there is a block here nothing else matters. Type yourdomain.com/robots.txt into your browser and look for Disallow: /. If it sits next to GPTBot, OAI-SearchBot or under the wildcard, you are blocking assistants.
A file you can copy and the full crawler list are in the post on GPTBot and robots.txt. The check takes thirty seconds.
Step 2. Narrow the category on your Google profile
The primary category works as a filter. Asked about pizza, a model considers pizzerias, not every restaurant in town. If yours says "Restaurant" instead of "Pizzeria", you are out before anything starts.
Set the narrowest category that is still true. Add the rest as secondary.
Step 3. Fill in services and the questions section
A model needs sentences it can quote. The category alone is not enough. Under services, list what you actually sell, in the words customers use. Under questions, ask the five things you answer on the phone and reply in full sentences.
"Yes, there is free parking for eight cars behind the building" can be dropped into an assistant’s answer. A bare "Yes" cannot be used anywhere. More in the post on Google Business Profile and ChatGPT.
Step 4. Make your details match everywhere
Name, address and phone have to look identical across Google, Apple Business Connect, Bing Places, Facebook and your own site footer. Not similar, identical.
A model that sees three different phone numbers does not pick the best one. It skips the business and names a neighbour whose details line up. This is one of the three most common causes of absence we find in audits.
Step 5. Paste structured data onto your site
A dozen lines of code no customer will ever see, handing a machine your name, address, hours and trade without guesswork. Ready code with values to swap out is in the post on LocalBusiness structured data.
After pasting, run the page through Google’s Rich Results Test. It shows what was detected and whether anything is broken.
Step 6. Collect a few recent reviews with specifics
Reviews from three years ago are a weak signal. It is not about volume, it is about somebody having written something recently, and having written something concrete.
Instead of asking for a review, ask one thing: "what did you order and how was it?". A question about a specific gets an answer with a specific, and that is what a model quotes as a reason. We expand on this in the post on reviews and AI recommendations.
Step 7. Put answers to customer questions on your site
If customers ask about parking, card payments and Sunday hours, and your site says nothing about any of it, the model has nothing to quote. One page with ten questions and short answers solves it.
Write the way customers ask, not the way a keyword would read. A model matches the user’s question against sentences on your page, so the closer to natural language, the better.
What not to do
- Do not buy reviews. Google removes them, and a model reads wording, not star counts.
- Do not bolt your town and trade onto your business name. It breaks Google’s rules and corrupts measurement.
- Do not write pages for models. A page crammed with phrases reads badly, and the model extracts either a fact or nothing.
- Do not rely on one screenshot. The answer changes on every query.
How long it takes
| When | What happens |
|---|---|
| 1 to 3 days | Profile edits are approved and live |
| 1 to 2 weeks | Data spreads into maps and the services that copy them |
| 4 to 8 weeks | Models start quoting the corrected information |
To see a difference you need a starting point. Measure your share of answers before the fixes, with the same questions you will use afterwards. How to do it by hand is in the post on generative engine optimization, and if you would rather have it measured for you, the free scan takes two minutes.