Checklist · 7 min read

GEO Checklist 2026: How to Make a Page AI-Readable and Citable

GEO (generative engine optimization) is the practice of making a page easy for AI assistants to understand, quote and act on. In 2026 that comes down to a short, concrete list: answer first, structure for questions, publish the machine-readable files, keep facts verifiable, and let the right crawlers in. This page is that list, and every item links to a deeper guide.

It is written for people who build and run small sites, not for a marketing department. Every item is something you can check yourself in an afternoon, and nothing here depends on knowing how any particular model works internally.

What is GEO, and how is it different from SEO?

Search engine optimization competes for a position in a list of links. Generative engine optimization competes to be the source: the page an assistant pulls a fact from, links to, or calls a tool on when a person asks a question. That changes what you optimise for. A ranked page can be long and pleasant; a cited page has to be extractable — one sentence that answers the question, headings that match how people ask it, and claims that can be checked.

The two are not enemies. Crawlable pages, fast pages, clear titles and genuine content help both. GEO simply adds a second, machine-facing layer on top: structured data, a plain-text map for models, and declarations that tell an assistant what your page can do, not just what it says. If you have never seen what that layer looks like, start with what WebMCP is — it is the clearest example of a page exposing tools to an assistant.

The 2026 GEO checklist

Work top to bottom. The first six items are content and structure; the last four are the machine layer most sites skip entirely.

1. Answer in the first paragraph. Give a direct, one or two sentence answer before any preamble. Assistants quote the answer first, and a definition-shaped opening is the easiest thing to lift.
2. Use question-shaped headings. Write H2s the way a person would ask the question ("What is GEO?", "Do I need llms.txt?"), so the heading itself is a match.
3. Add FAQPage schema that matches the visible text. The structured Q&A and the on-page Q&A must be identical, word for word. Mismatched schema is worse than none.
4. Show your work. Put a section that only you could write — a measurement, a test result, a screenshot with a date. Unique, verifiable evidence is the strongest reason to be cited. Our CSV test with a chat model is an example of this pattern.
5. Never invent numbers. If you do not have a source, do not state a statistic. An unsourced figure is a reason to be skipped, not a reason to rank.
6. Write scannable structure. Short paragraphs, real subheadings, tables where a comparison helps, and links to the pages that go deeper.
7. Publish an llms.txt. A plain-text map of your site for language models, served at /llms.txt. It is optional and cheap — and one of the few levers you fully control. See llms.txt explained.
8. Declare your tools in agent-tools.json. If your page offers a converter, a calculator or a lookup, describe it in a machine-readable file so an assistant can find and call it. See agent-tools.json explained.
9. Mark up interactive forms with WebMCP. Attribute-level declarations let an assistant register your form fields as callable tools. See can AI fill out a form? and WebMCP vs MCP.
10. Let the right AI crawlers in. Check that your robots.txt does not accidentally block the assistants you want. See AI crawlers in robots.txt.

The part most sites miss: the agent layer

Items one to six are content discipline. Items seven to ten are a different layer, and they are the reason a page can be used rather than just read. Three files and one markup convention do the work:

None of this is guesswork on a live site: on our own pages a browser assistant can read the tool list back through navigator.modelContext.getTools(), and we have published the exact results — including the fields an agent still cannot use. If you want the shipped version of this layer rather than the theory, it is the service we run at /agent-ready.

What we measured on our own sites

This checklist is not a repackaged list of tips; each point comes from something we had to fix on pages we run. The useful, honest parts are the limits:

• How many tools a live page will actually advertise, and one entry an anonymous agent could not reach — see agent-tools.json explained.
• Which HTML field types become agent parameters, and the file-input gap we measured across live pages — see can AI fill out a form?.
• Why a parser check on robots.txt can read "zero crawlers" when the file is fine — see AI crawlers in robots.txt.
• What an llms.txt file really contains, measured by size on three sites — see llms.txt explained.

Writing the limits down is deliberate. A page that claims everything works is easier to dismiss than one that says where the gaps are, and the gaps are often the most useful thing a reader takes away.

How to test your site against this checklist

Reading a checklist is not the same as passing it. Test the live page, not the draft:

1. Fetch the page and confirm it returns HTTP 200 (a soft-404 that returns 200 with the home page is a common trap).
2. Open the page source and read the JSON-LD. Check the FAQ answers match the visible Q&A word for word.
3. Request /llms.txt and /agent-tools.json and confirm both return real content, not a 404.
4. If the page has tools, check the declared selectors against the real DOM — a selector that points at an element that does not exist is invisible to an agent.
5. Re-read robots.txt and confirm the crawlers you want are not blocked.

That is the same sequence we run in is your website agent-ready?, with the commands and the measured output on three live sites.

Frequently Asked Questions

What is GEO (generative engine optimization)?

GEO is the practice of making a web page easy for AI assistants to understand, quote and act on. Classic SEO optimises for a ranked list of links; GEO optimises for being the source an assistant reads and cites when it answers a question, and for being a page an assistant can call a tool on.

Is GEO the same as SEO?

No, but they overlap. Good SEO still helps: crawlable pages, clear titles, real content. GEO adds structure that machines reward directly, such as a definition in the first paragraph, question-shaped headings, FAQ schema that matches the visible text, an llms.txt file, and machine-readable declarations of any tools the page offers.

Do I need a separate llms.txt file?

It is optional but cheap, and it is one of the few GEO levers you fully control. An llms.txt is a plain-text map of your site written for language models, and it is served at /llms.txt. It does not replace robots.txt or sitemap.xml; it complements them.

How do I know if my page is ready for AI assistants?

Work the checklist above and then test the live page. Confirm the page returns HTTP 200, that your FAQ schema matches the visible Q&A word for word, that /llms.txt is reachable, and, if the page has interactive tools, that they are declared and reachable by an anonymous agent.

Can I guarantee an AI assistant will cite my page?

No. No one can promise that a model will quote a specific page. What you can do is remove the reasons a page gets skipped: unclear answers, missing structure, blocked crawlers, unverifiable claims, and no machine-readable description of what the page can do.