Google AI Overviews and AI Mode: official recommendations for websites
Google states that being included in AI Overviews and AI Mode requires no special technical requirements beyond standard Search. A page must be indexed, eligible to appear with a snippet, and meet the usual technical and content guidelines. The biggest long-term advantage still belongs to original content with real value.
What AI Overviews and AI Mode are
AI Overviews generate a summary directly within Google's search results and supplement it with links to supporting web sources.
AI Mode offers a more conversational way of working with complex queries and follow-up questions. It also draws on web results and links to relevant pages.
Both features are part of Google Search. They are not a separate index that a website registers for through a special GEO file.
How query fan-out works
Google describes a technique called "query fan-out." From the original query, the system generates multiple related searches covering different subtopics and data sources. This lets it surface supporting pages that might not appear in the classic top ten results for the exact original phrase.
The practical implications:
- mechanically repeating a single keyword is not enough;
- it helps to cover the topic in depth;
- a single page doesn't need to answer every variant of the query;
- internal architecture should help connect the main topic with its subtopics;
- original experience can set a page apart from commodity summaries.
Don't create a separate page for every possible fan-out variant, though. Google warns against scaled content produced primarily to manipulate results.
Technical requirements
A page must:
- meet Google Search's technical requirements;
- be accessible to Googlebot;
- be indexed;
- be eligible to display a snippet;
- have its main content available on the mobile version too;
- have no critical rendering errors;
- provide a good experience for visitors.
Appearing in these features isn't guaranteed even after meeting all the recommendations.
Content with lasting value
Google's current official guidance emphasizes "non-commodity content" — content that doesn't merely recycle information that's already widely available.
Add:
- your own first-hand experience;
- original photos and video;
- real-world data;
- testing methodology;
- specific recommendations and limitations;
- a unique expert perspective;
- tools, calculators, and templates;
- up-to-date product or local information.
A generic article built by rewriting the top search results can easily be replaced. Your own measurements or experience cannot.
What Google explicitly does not require
llms.txt
Google states that it does not use llms.txt for Search. The file neither helps nor hurts ranking and visibility in its generative features.
Special AI schema
No special Schema.org type is required for AI Overviews or AI Mode. Standard structured data can still help Search understand content and unlock supported rich results.
Artificial "chunking"
There is no requirement to split text into tiny blocks just for AI. Choose length and structure based on the topic and the user's needs.
Rewriting content solely for AI
Google understands synonyms and meaning. There's no need to produce a version of a page for every phrase or rewrite natural text into some imagined "LLM language."
Inauthentic mentions
Artificially publishing brand mentions on low-quality sites is not a reliable approach. Search and spam systems evaluate the quality and authenticity of sources.
Products and local businesses
Google recommends keeping data current in:
- Google Merchant Center;
- product feeds;
- Google Business Profile;
- supported business and merchant features.
For an online store, an accurate feed, stock availability, shipping, and returns information often deliver more value than another generic blog post.
Structured data
Use it where:
- a matching type exists;
- the properties match the visible content;
- the data is up to date;
- the implementation is valid;
- it's meaningful for Search or other consumers.
Don't add schema just to increase the number of types. "More is better" doesn't apply here.
How to measure performance
Google recommends using Search Console. Its current documentation describes a performance report for generative AI features. Track:
- pages gaining visibility;
- queries and topics;
- clicks and conversions;
- changes following content and technical updates;
- the split between classic and generative search, where the report exposes it.
Alongside this, measure the quality of the visits you get: engagement, inquiries, sign-ups, and revenue. A citation alone, without business impact, may not be the goal.
Practical action plan
- Verify the indexing of key pages.
- Check mobile content and JavaScript rendering.
- Remove duplicates and weak variants.
- Identify questions that need a deeper explanation.
- Add your own experience, data, and media.
- Update Merchant Center or Business Profile.
- Add meaningful structured data.
- Monitor Search Console and conversions.
- Update the article only for substantive changes, not just to bump the date.
Frequently asked questions
Do I need a GEO agency to appear in AI Overviews?
No. You need solid SEO, a technically accessible website, and content with real value. Outside help can speed up analysis and execution, but no agency can guarantee inclusion.
Does FAQ schema help?
It can describe an actual FAQ section, but it's not a requirement for AI Overviews and it doesn't guarantee a citation on its own.
Does the article need to be long?
There's no ideal length. It needs to sufficiently satisfy the user's intent without unnecessary filler.