Designing for AI answers: how a 2027 website gets read by machines and people

A growing share of people ask an AI assistant who to call before they ever open a search result. The assistant reads some websites, builds an answer, and maybe links one or two of them. Your website now has two audiences: the person, and the machine that summarizes you for the person.

The good news… almost everything that helps the machine also helps the person. Here’s what we do on our own site, and what I’d do on yours.

1. Real HTML, rendered on the server

If the text of your page only appears after a pile of JavaScript runs, some tools will see a blank page. Ours is built with Astro and ships plain HTML from the server, so the content is there on the first byte. I wrote about that stack in The modern web stack for 2026.

You don’t need our stack to get this right. You do need to check. View the page source (not the inspector, the actual source) and search for a sentence from your page. If it isn’t there, a lot of tools won’t see it either.

2. Headings that match the questions people ask

An assistant answering “how much does a kitchen remodel cost in Roseville” looks for a page that clearly answers it. A heading that says “What a kitchen remodel costs in Roseville” and a paragraph with an actual range under it beats a beautiful page that says “Crafting Spaces You Love.”

That’s also just good writing for people. Say what you do, where, and roughly for how much, in plain words.

3. Structured data that states the facts

Structured data is markup (from schema.org) that states facts in a form a machine doesn’t have to guess at. Our Lab articles carry Article, FAQ and breadcrumb markup, and the site carries our business details. For a local business the important ones are your business type, name, address, phone, hours, service area and services.

The point is consistency. If your address on the site, on your Google profile and in your markup all match, every tool reading you agrees on who you are. Google Business Profile for local SEO covers the profile side.

4. An llms.txt file

llms.txt is a proposal for a short plain-text file at the root of a site that lists the important pages with a one-line description each (llmstxt.org). Ours is at greyskymedia.com/llms.txt and it’s generated on every build from the same content the pages use, so it can’t fall out of date.

I want to be honest about this one. It is a proposal, not a standard, and I don’t know of any assistant that has said it ranks sites higher because of it. It costs almost nothing and it makes the site easier to read, so we do it.

5. A plain-text twin of every page

We go one step further. After each build, a script writes a Markdown version of every page next to the HTML: same words, no navigation, no styling. The llms.txt file links to those. Because the Markdown is made from the built page, it can’t drift from what visitors see.

This is probably more than a typical small-business site needs. I mention it because it shows the principle: give machines a clean copy of the same facts people see, never a different story.

6. Don’t hide the important stuff

Things that make a page hard for machines to read:

• Prices and service lists inside images.
• Key details behind tabs or accordions that only load on click.
• PDFs as the only place your menu or price list lives.
• Pop-ups that cover the content on load.

Every one of those also annoys people. Fix them for the people and the machines come along.

7. Speed still counts

Crawlers have budgets too. A slow page is a page that may get read less often. Everything in Why your website should be boring and fast still applies.

Where to start

Ask two or three AI assistants about your own business: what you do, where, and how to reach you. Write down what they get wrong. That list is your to-do list, and it usually points straight at the page or profile that needs fixing.

If you’d like help working through it, let me know. Thanks!