Anyone selling llms.txt, content chunking, or AI-specific schema as a path to AI Overview citations has been wrong for 18 months. Google says so.
But there is a wrinkle you have to iron out. “Not good for Google Search” is not the same as “not good for AI agents.”
In the section that answers whether SEO is still important for productive AI search, Google’s new optimization guide mentions the term AEO and GEO: “From Google Search’s point of view, optimizing for productive AI search improves the search experience, and thus still SEO.” Five tactics are named in the Mythbusting section as things to ignore: machine-readable AI files like llms.txt, content bundling, AI-specific content rewriting, misrepresentation, and structured data manipulation. That’s a contradiction, in Google’s own words.
Read those five again, once in a Google search, and once everywhere else.
The Scope Google Covers, and the Scope Is Not
Google’s guide, and the entire AEO and GEO playbook, is about one thing: getting your content cited within an AI-generated response. AI Overview, AI Mode, ChatGPT, and Perplexity all have the same shape. A really different scope is what happens when an independent agent doesn’t mention your website but actually does something on it.
The guide summarizes this. Under the “Agentic Experience” section, Google admits that “AI agents are autonomous systems that can perform tasks for humans, such as making reservations or comparing product information,” and that “browser agents can access your website to collect the data they need to complete these tasks, such as analyzing visual renderings (such as screenshots), inspecting and accessing the DOM tree.” Google points to a separate document on web.dev for agent-friendly UX patterns.
What the guide doesn’t talk about is that the five strategies it has withdrawn from the scope of the quote may still have an agent-acting-on-website scope service. That is an open question. Read each of these five strategies twice: once for the scope of the quote where Google’s output is correct, and once for the scope of the action where the answer varies by strategy and application.
LLMs.txt and AI Machine Readable Files
To be cited in Google Search, Googlebot reads your HTML and ignores llms.txt completely. The llms.txt file does not change the citations in AI Overview or AI Mode, and no consultant should bill you for it as a citation trick.
In the scope of action, the idea of ”website manual for AI agents” makes sense. An independent agent navigating your website to complete a task on behalf of a user can clearly benefit from a selective index of what capabilities are contained, what API endpoints are there, what workflows are written there. The goal of having a machine-readable map of agents who need to do something, not just retrieve, holds up.
But llms.txt itself is not yet a widely accepted standard for that. None of the major platforms whose agents will use it are committed to reading it as a discovery method. The concept may seem useful. A particular file format may end up being the standard, or another format may emerge, or the question may be resolved in another way entirely.
To be clear: Don’t put llms.txt on your website because someone told you it will help your AI Overview citations. The llms.txt file will not remove your AI overview citation count. If you have a different reason to publish a machine-readable manual for independent agents reading your documents, that’s a different decision, and the deployment data isn’t currently available to do it with confidence.
AI-Specific Content Rewriting Tells
For Google search citations, rewriting content aimed specifically at AI Overview is considered by Google’s quality systems as low-effort content. Rewriting AI is telling, not tricking.
For the scope of the action, the frame is wrong from the start. Directly labeling AI is the wrong framework. A good outline is writing clearly for any reader, person or machine. The content is designed to be released (answer-first, citable specifications, module blocks) to help every student, including the independent student. That’s the position of Machine-First Architecture, and it’s a content discipline that’s heavy on both scopes.
The same logic goes into the next three tricks on Google’s list.
Content Chunking, False Claims, and Structured Data Visualization
AI content classification follows an AI-specific rewriting concept. Breaking your content down into smaller AI-targeted chunks is a bad move, and building modular content blocks for retrievable output is a content behavior that helps any reader. Google systems handle multi-topic pages natively.
Lying works no matter how big it is. False product mentions, link buying, and altered quotes are not good for any reader or agent acquisition system. Google’s release here is closer to a statement of behavior than a question of scope. Manipulating returns with fake signals was a guideline violation for two decades before someone coined GEO to try to disrupt the SEO tools scene.
The structured-data obsession is easily misread among the five. Google has not said that it has stopped using the schema. The guide said that there is no special AI schema, and that focusing too much on schema as a citation lever is wrong. The standard schema.org markup still supports entity recognition, knowledge graph ownership, agent-readable product data for agent-consumer flows, and the overall machine-readable identity base. An Ahrefs study published on May 11, 2026 (1,885 pages adds schema, no meaningful citations to Google AI Overview, AI Mode, or ChatGPT) rated the query smaller than the title suggests. The schema is now the infrastructure for table ownership. What doesn’t work is to bolt it on for the sixth month and expect the quote to be raised.
What To Do With Google’s AI Optimization Guide
Ask yourself two questions after reading Google’s new guide.
Are you paying anyone for the tricks on the list released by Google? Wait.
Do you have visibility into how independent agents are reading your website outside of Google Search? You probably don’t, and neither does anyone else yet.
Read Google’s guide as the authority on what to include, and keep reading the rest of the web for what not to.
Additional resources:
This post was originally published on No Hacks.
Featured image: Roman Samborskii/Shutterstock
