What is technical GEO? The checklist we run first.

Technical GEO is the engineering work that decides whether an AI engine can fetch a page, read it and quote it. It covers crawler access, the raw HTML a fetcher receives, structured data, passage structure, citation anchors and the snippet controls that limit quoting. None of it is writing, but the writing can't be quoted without it.

The rest of this page is the checklist itself, the same one behind our technical SEO and GEO work. Wherever it says what a crawler or a search system does, it links to that vendor's own documentation.

What does technical GEO include?

Seven checks, in the order we run them. You can verify each one with a plain request or by reading a single file.

1. AI crawler access, in robots.txt and at the CDN edge

An engine that can't fetch the page has nothing to quote. Robots.txt is half of this check. The other half is the firewall or CDN in front of the site, which can refuse a crawler that robots.txt allows.

The agent names do different jobs, so read each vendor's reference before writing a rule. OpenAI's bots reference says OAI-SearchBot surfaces sites in ChatGPT search while GPTBot gathers training data. Anthropic separates ClaudeBot from the agent that fetches pages for a user in its crawler article. Perplexity documents PerplexityBot in its bot guide, and Google lists Google-Extended in its crawlers reference.

When you read the edge logs, group requests by the CDN's verified-bot flag rather than the user-agent string. Many requests that claim to be a crawler are not, and our GEO essay has the numbers from this domain.

2. llms.txt and markdown negotiation

An llms.txt file is a plain-text map of a site written for language models. It is a proposal, not a standard, and Google's AI features guide says you don't need new machine-readable files or AI text files to appear in those features. We publish one anyway because it is cheap.

Markdown negotiation applies the same idea to each page. A request that sends Accept: text/markdown gets a clean markdown copy, and every other request gets the HTML. This site does it, so you can try it with curl -H "Accept: text/markdown" on this URL.

3. Server-side rendering

The words have to be in the HTML the server sends. Google's JavaScript SEO guide describes crawling, rendering and indexing as separate phases, with a headless Chromium rendering the page "once Google's resources allow". The same guide still calls server-side or pre-rendering a great idea, because not all bots can run JavaScript.

We don't assume any AI fetcher runs scripts. The test is a plain fetch: if the body text isn't in the raw response, the page isn't citable.

4. Structured data and sameAs

Schema tells a machine what the page is and who published it. Google lists the types it reads in its structured data documentation, and its AI features guide says no special schema is needed for those features. What it does ask for is markup that matches the visible text, so the work is getting it accurate rather than adding more of it.

sameAs is the property that ties an organization to its profiles elsewhere. Schema.org defines it as a reference page that unambiguously indicates the item's identity. It is how we tell an engine that the firm on this page and the one on GitHub are the same.

5. Passage-level structure

An engine that quotes from a page has to pick a passage, and a passage that only makes sense after the three paragraphs above it is a poor pick. So each section gets a heading phrased as the question it answers, a direct answer in its first sentence, and the detail after that.

This is where technical and editorial work meet, and it is the check that never quite finishes. Our GEO essay covers it in more depth.

6. Citation plumbing

Once a passage is worth quoting, make it easy to point at. Every heading gets a stable id, so a link can land on one section. A short cite block gives the page name, its dates and the canonical URL in a form someone can paste. A reuse note says what may be republished and how to credit it. This page has all three, and the cite block is at the bottom.

7. Opt-out levers: nosnippet, max-snippet and data-nosnippet

These controls work in the other direction. They limit what may be quoted. Google documents nosnippet, max-snippet and data-nosnippet in its robots meta tag reference, and its AI features guide names the same three, plus noindex, as the way to limit what Search shows from a page.

Check what your template sets. On pages we want quoted we use max-snippet:-1, and we keep data-nosnippet for parts that shouldn't be lifted out of context.

How is technical GEO different from technical SEO?

Most of the work overlaps. A page that can't be crawled or indexed won't be cited either, and Google's AI features guide says a page must be indexed and eligible to show with a snippet before it can appear as a supporting link. What changes is the goal and the unit of work.

How do we measure it?

Rank trackers don't see citations, so we track them separately: which pages get cited, for which prompts, on which engines, next to classic positions. That tracking runs in ConceptSEO, the platform we built for it.

For what this checklist produced when we ran it every week on live sites, including the checks that turned out to separate nothing, read the twenty-week AI search audit benchmark. The numbers live there, with their windows and sample sizes, so they aren't repeated here.

Questions about technical GEO

Do I need an llms.txt file to show up in AI answers?

No. Google says you don't need new machine-readable files, AI text files or special markup to appear in AI Overviews or AI Mode. We publish an llms.txt because it is cheap, not because any engine has said it reads one.

Does blocking GPTBot keep my site out of ChatGPT search?

OpenAI documents them as separate agents. GPTBot collects training data. OAI-SearchBot is the one that surfaces sites in ChatGPT search, and OpenAI says sites that disallow it will not be shown in ChatGPT search answers. Block the agent that matches what you want to stop.

Can our own team run this checklist?

Yes. Every check is a fetch or a file you can read, and none of them needs special tools. Where it usually stalls is after the audit, when rendering changes and edge rules need someone with the access to merge them.

Where this checklist comes from, and reusing it

The seven checks are the ones we run on our own sites and then on client work. Claims about what a crawler or search system does link to the vendor's documentation at the point they are made: Google for rendering, structured data, snippet controls and AI features, OpenAI, Anthropic and Perplexity for their crawlers, schema.org for sameAs, and llmstxt.org for the llms.txt proposal. We read the Google and OpenAI pages on October 26, 2026. Anything we observed ourselves links to the page that holds the data.

You may republish or adapt this checklist with attribution and a link to https://concept211.com/articles/what-is-technical-geo/. Every heading carries a stable id, so you can link to one check instead of the whole page.

Photo by Jakub Zerdzicki on Pexels