GEO is the fastest-growing category of marketing advice with the thinnest evidential base. Vendors are selling optimisation techniques for systems whose operators have published documentation saying those techniques are unnecessary. This article separates what search engines have actually stated from what is being asserted, and is explicit about where the honest answer is that nobody outside the model providers knows.
What Google has actually documented
Google’s Search Central documentation on AI features is unusually direct. It states that you do not need to create new machine-readable files, AI text files, or markup to appear in these features, and that there is no special schema.org structured data you need to add. On eligibility it says that to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, meeting the ordinary Search technical requirements. It adds that there are no additional requirements and no other special optimisations necessary.
That is the primary source, and it undercuts a large share of what is currently sold as GEO. The controls Google does document are the familiar ones for limiting what appears: nosnippet, data-nosnippet, max-snippet, and noindex. Note the direct consequence of the snippet-eligibility rule: applying nosnippet to a page removes it from consideration as a supporting link.
The llms.txt question, answered plainly
llms.txt is a proposed convention for a markdown summary file that AI systems would read. It has been widely recommended and it is worth knowing where it stands. John Mueller of Google stated in June 2025 that no AI system currently uses llms.txt, and Gary Illyes said at Search Central Live in July 2025 that Google does not support it and does not plan to, comparing the idea to the meta keywords tag. Google’s AI features documentation makes no mention of it and explicitly says no AI text files are needed.
The file costs almost nothing to publish, so this is not an argument against having one. It is an argument against paying for one, or believing it will produce visibility.
Claims versus evidence
| Claim commonly made | Status |
|---|---|
| You need special schema markup for AI Overviews | Contradicted by Google, which states there is no special schema.org structured data required. |
| You need an llms.txt file | Google has said no AI system uses it and that it does not support it. No documented benefit. |
| Being indexed and snippet-eligible matters | Documented by Google as the actual eligibility requirement. |
| Clear, self-contained answers help | Plausible and consistent with how extractive systems work, but not quantified in any primary source. Treat as good writing practice rather than a mechanism. |
| Third-party mentions and corroboration help | Plausible, and consistent with how these systems are described as being trained and grounded, but no operator publishes a weighting. Unverified. |
| Vendor tools can measure your AI visibility share | They can sample prompts and count mentions. That is a sample, not a share of anything, and results vary by prompt phrasing, personalisation, and date. |
What to actually do, given the evidence
- Confirm your pages are indexed and not blocked, and that you have not applied nosnippet, since snippet eligibility is the documented requirement.
- Keep business facts identical everywhere. Consistency is the thing you control, it helps human buyers regardless, and it is what makes a description of your business unambiguous.
- Answer the question in the first sentence of the section, then support it. This is defensible on writing quality grounds alone, which means you are not betting on an unverified mechanism.
- Publish things worth citing: original data, documented decisions, specific numbers. Corroborated, specific, attributable material is the only durable asset in every version of this that anyone has proposed.
- Track brand-name queries and direct traffic alongside any prompt sampling, since those are measured on infrastructure you control.
Why this is unsatisfying, and why that is the right answer
The uncomfortable conclusion is that documented GEO is mostly ordinary quality and hygiene work with a new name attached, and the parts that would be genuinely novel are the parts nobody can verify. Anyone offering more certainty than that is selling confidence rather than evidence. The reasonable stance is to do the fundamentals well, keep a sceptical eye on the sampling data, and revisit when the operators publish something new rather than when a vendor publishes a case study.
Key takeaways
- Google documents that no new machine-readable files, AI text files, or special schema.org structured data are needed for AI Overviews or AI Mode.
- The documented eligibility rule is that a page must be indexed and eligible to be shown with a snippet. Applying nosnippet removes it from consideration.
- Google has publicly said no AI system uses llms.txt and that it does not support it. Publish one if you like, but do not pay for it.
- Claims that mentions and corroboration drive AI citation are plausible but unquantified by any operator. Label them as unverified.
- “Share of AI voice” is a sample of prompts on particular days, not a market share. Useful as a trend, misleading as a number.
- Most defensible GEO work is ordinary quality and consistency work, which pays off regardless of whether the AI mechanism turns out as advertised.
Related reading
Sources
- AI Features and Your Website, Google Search Central
- Google Says No AI System Currently Uses LLMs.txt, Search Engine Roundtable (2025)
- Google Says llms.txt Is Purely Speculative For Now, Search Engine Journal (2025)
- Improve your local ranking on Google, Google Business Profile Help
- Digital Brand
- GEO
- SEO
Written by
Mara Whitfield
Brand Strategy Lead
Mara Whitfield leads brand strategy at ThisCom, helping small and medium businesses build distinctive brands and consistent digital presences that earn trust and stand out.
All articles by Mara Whitfield →
