Google finally wrote it down. After a year of agencies selling "AI Mode optimization" packages built on guesswork, Google published an official guide to how content surfaces in AI Overviews and AI Mode — and it quietly demolishes half the advice being sold, while confirming a few levers most teams underuse. If you run organic growth or an e-commerce catalog, this is the document your 2026 roadmap should answer to.
At GEOly AI we track how AI Mode cites and recommends brands daily, so we read the guide the way practitioners should: not "what does Google say," but "what does this mean I should do differently on Monday." Here is that reading.
Key takeaways
- AI Overviews and AI Mode retrieve from the same Search index using the same ranking and quality systems. There is no separate AI algorithm to optimize for — ranking well remains the entry ticket.
- Retrieval runs through query fan-out: your page gets pulled for hidden sub-queries, not just the visible question. Covering a topic's sub-questions beats repeating one keyword.
- Google explicitly confirms it ignores llms.txt — the file "neither helps nor harms." Budget spent there is budget wasted.
- The don't list is pointed: don't chunk pages into snippet bait, don't rewrite content "for AI," don't chase inauthentic mentions.
- For merchants, the underused lever is feed quality: Merchant Center product data (price, availability, images, attributes) directly feeds AI shopping surfaces — the same data layer that decides agentic checkout eligibility elsewhere.
How AI Mode actually picks sources
The guide describes a retrieval-augmented pipeline. A user asks a question; the model issues query fan-out — multiple hidden sub-queries exploring facets of the question (we mapped this layer in detail in our grounding queries guide); retrieval pulls candidates from the standard Search index, ranked by the standard systems; for shopping intents, Merchant Center feeds and Business Profile data join the pool; the model synthesizes an answer with citations — "supporting links" chosen to back specific claims.
Two consequences follow directly. First, pages that rank for nothing get retrieved for nothing — classic SEO health is upstream of all AI visibility in Google. Second, a page can be retrieved for a sub-query it never visibly targeted; comprehensive topic coverage earns multiple fan-out entry points, which is why thin keyword-matched pages underperform in AI surfaces even when they rank.
