What has Google documented?
Google has documented that AI Overviews and AI Mode are built on the same core systems that power Search, including its ranking and quality systems, and that they show links to sources so users can explore further. Google has also described query fan-out in AI Mode: the system issues multiple related searches across subtopics and data sources at once and synthesises the results.
Google's published guidance for creators is that no special optimisation is needed beyond its existing guidance for helpful, reliable, people-first content, and that standard technical requirements such as indexability apply. These statements are the documented part; check Google's current documentation, since it changes.
What is observable in results?
Several patterns are observable by sampling AI Overviews across many queries, and are consistent enough to act on, with the caveat that they are observations rather than rules. Cited passages tend to be direct answers to the question or a sub-question, self-contained enough to read without the rest of the page. Cited pages tend to already rank in the top results for the query or a closely related one.
Cited content tends to name entities explicitly and to make specific, checkable claims rather than general ones. And the sources for one Overview tend to cover different sub-questions, consistent with fan-out assembling an answer from several angles.
- Cited passages are direct and self-contained
- Cited pages usually rank for the query or a related one
- Cited content is entity-explicit and specific
- Sources across an Overview tend to cover different sub-questions
What is unknown?
The exact weighting is unknown: how much ranking position matters versus passage quality, how sources are deduplicated, how freshness is treated, and how the synthesis chooses which claims to attribute. Anyone who tells you the formula is guessing, and any tactic that depends on a specific formula is fragile.
It is also unknown how stable any observation is over time. Google changes these systems frequently. The right response is to sample on a schedule, as the measuring AI citations guide describes, and to notice when patterns shift.
What can a publisher act on?
A publisher can act on the documented and observable parts together. Because Overviews use core ranking, the page has to be good by the usual standards: indexable, fast, helpful, authoritative on its topic, and ranking for related queries. Because fan-out assembles from sub-questions, the page should answer several sub-questions in self-contained sections with question headings.
Because cited passages are direct and specific, every section should open with its answer and make checkable claims. This is the SPEC-035 discipline used for every piece on ftes.ai, and it is derived from the documented and observable parts only.
What should a publisher ignore?
Ignore any tactic that claims a specific trick guarantees inclusion: a particular schema type, a magic word count, a hidden block of text for the AI. There is no documented basis for them, and some, such as hidden text, violate Google's guidelines and risk the page's ordinary ranking too.
Ignore anyone reporting precise citation percentages for tactics without showing their sampling method. The honest position is that structure and verifiability are observably associated with citation, that the weighting is unknown, and that a measured loop is how you find out what works for your pages.
How do you build your own evidence rather than relying on anyone's claims?
Build your own evidence with a sampling record. Choose a fixed set of prompts that matter to your business, including the primary questions and their likely sub-queries, and run them in Google search on a schedule, recording for each whether an AI Overview appeared, which sources it cited, in what order, and what passage from each source it appears to have used. Store the answer text with the date. Repeat weekly, because Overviews vary by session and change over time, and report ranges rather than single values.
After a quarter the record shows patterns specific to your queries: which of your pages are cited and for which sub-queries, which competitors are cited where you are not, and how the cited passages are structured. Those observations are worth more than any general claim, because they are about the queries you compete for, and they are what the Search Visibility FTE's measurement sweep produces. Treat every published pattern, including the ones in this guide, as a hypothesis to test against that record rather than a rule to follow.