What should I put on my website so AI assistants can read it?

Write for a reader that cannot see

An AI assistant reads the text of your page. It does not look at the design. A phone number that only exists inside a banner image, a list of towns that is only a map graphic, or prices in a photographed flyer may never reach it.

The fix is free. Put the fact in a sentence. If a blind customer using a screen reader could get every important detail from your page, an assistant probably can too.

The facts to state outright

Think about the questions a customer asks an assistant, then make sure your site contains the sentence that answers each one.

• What you do, one service per page, named the way a customer would name it.
• Where you do it: a written list of towns, neighborhoods or ZIP codes.
• When: opening hours, weekend hours, and whether you take emergency calls.
• How much: a call-out fee, a starting price, a range, or what the estimate depends on.
• Who: the business name, how long you have traded, licenses and insurance.
• How to reach you: phone, email, form, and the business address if you show one.

Add an FAQ to each service page

Customers ask assistants questions, so pages written as questions and answers match well. Put three to six real questions on each service page, the ones you actually hear on the phone, with a two or three sentence answer under each.

Keep the answers specific to your business. "Do you repair or only replace?" and "Do you charge to come out and look?" are useful. A generic paragraph about why maintenance is important is not.

Schema markup, in plain words

Schema markup, also called structured data, is a block of labeled facts in the page's code that visitors never see. It says, in a fixed vocabulary from schema.org, "this is a business, this is its name, this is its phone number, these are its hours, these are the areas it serves".

There is a general LocalBusiness type and more specific ones such as Plumber, Electrician, RoofingContractor, HVACBusiness and Dentist. Use the most specific one that fits. The markup must match what the page visibly says. You can test it with Google's Rich Results Test or the validator on schema.org. Markup does not guarantee anything shows up differently in search. It removes guesswork for the machines reading the page.

Do not lock the door

Your site has, or can have, a robots.txt file that tells automated visitors what they may read. Some site platforms and security plugins now block AI crawlers by default. If you want to be read, check that yours does not.

The crawler names to look for include GPTBot and OAI-SearchBot from OpenAI, ClaudeBot from Anthropic, and PerplexityBot from Perplexity. Blocking them is a legitimate choice for a publisher protecting its writing. For a local business that wants to be recommended, it works against you.

What llms.txt is, honestly

llms.txt is a proposed convention: a plain text file at the top level of your site that summarizes who you are and links to your most important pages, written for language models. It takes a few minutes to make.

It is not an official standard, and the major AI companies have not said their assistants rely on it. I keep one on my own site because it is cheap and harmless, not because it has been shown to change anything. Do it after the items above, not instead of them.

A fifteen-minute check you can do yourself

You do not need a developer to find out how your site reads to a machine. You need a browser and a notes app.

• Open your home page, select everything, copy it and paste it into a plain notes app. What lands there is close to what a text reader gets. Look for your phone number, your towns and your services in it.
• Do the same for your most important service page. If the paste is mostly menu items and a slogan, the page is thin.
• Type your address followed by /robots.txt into the browser. If you see "Disallow: /" under "User-agent: *" or under one of the AI crawler names, that visitor is being told to stay out.
• Paste a page address into Google's Rich Results Test. It shows whether the page has structured data and whether any of it is broken.
• Ask an assistant that can browse the web what your business does, which towns it covers and when it is open. Compare the answer with the truth, and note which pages it cites.

One service page, before and after

Here is a page I see often. The heading says "Our Services". Under it is a photo with "Quality you can trust" written across it, then a row of icons labeled Repair, Install and Maintain. The phone number is part of the header image. To a person it looks finished. To a text reader it says almost nothing: three single words and a slogan.

Now the same business with the facts written down. The heading says "Water heater repair and replacement" with the town. The first paragraph says they repair and replace tank and tankless heaters, in which towns, and that they take calls on weekends. A short paragraph says what a visit involves and what the estimate depends on. Four questions follow, each with a two-sentence answer. The phone number is text you can tap.

Nothing about the second page is technical. It is what the owner says on the phone, typed out. An assistant can quote the second page. It cannot quote the first.

Common mistakes

These undo good work, and most of them are easy to miss because the site still looks fine.

• Facts that only exist in an image, a video or a downloadable flyer.
• Hours on the website that differ from the hours on the Google Business Profile.
• Schema markup copied from a template with another business's details or the wrong business type.
• An FAQ stuffed with town names and search phrases in place of real answers.
• A security setting that blocks AI crawlers, turned on by default and never looked at.

Questions to ask whoever looks after your site

If somebody else built or manages the site, four questions tell you where you stand.

• Does the page content arrive as text in the page itself, or is it filled in by script after the page loads?
• What structured data does each page carry, and does it pass a testing tool?
• Does our robots.txt, our host or any plugin block AI crawlers?
• Who updates the facts when our hours, prices or areas change, and how fast?

What order to do it in

If you only have an hour, spend it on words. Put the phone number, hours and towns in text on every page. Then give each main service its own page with a short FAQ. Then check robots.txt. Schema markup comes after that, and llms.txt last. The first two steps help every visitor, human or machine. The later ones only help if the earlier ones are done.

Keep it true and keep it current

An assistant that repeats an old price or a town you stopped serving sends you the wrong customers. Put a date on pages that carry prices, and review your facts a couple of times a year. Sites I build at Novyx include the structured data and plain-text facts described here.

Questions people ask