How Google AI Overviews Choose What to Show
AI Overviews now sit above the results on a large share of searches. Here is what is known about how they pick sources.
The box at the top of a Google results page has quietly become the page for a lot of searches. Understanding what feeds it is now part of understanding search.
How often it appears
Tracking estimates put AI Overviews on roughly one in three US desktop queries, and far higher on informational ones, where some analyses put it near 60 percent.
That is not a feature appearing occasionally. For informational intent it is closer to the default.
What is known about source selection
Google has not published a ranking algorithm for Overviews, and anyone claiming otherwise is guessing. What the tracking data shows is consistent enough to be useful.
Overviews cite considerably more sources than a chat assistant does. Google's AI Mode attaches something in the range of 15 to 22 sources per answer depending on the vertical, roughly three times what ChatGPT, Gemini or Copilot return.
The sources skew toward established domains with existing search history, which is unsurprising given that the underlying index is Google's own. Pages that answer a specific question directly, near the top, are easier to extract from than pages that build toward an answer over 2,000 words.
What it changes
Two things, in opposite directions.
It is easier to be one of 20 cited sources than one of three, so the ceiling for inclusion is lower than with a chat assistant.
But inclusion converts worse. A user who reads a synthesised answer with 20 little source chips often does not click any of them. Being cited is not the same as being visited, and measuring GEO by traffic will understate it while measuring it by revenue may understate it further.
What to do about it
Answer real questions directly and early on the page, so a passage can be lifted cleanly. Keep the facts about your business consistent everywhere they appear. And be present on domains with the search history that Google already trusts, because the Overview is drawing from the same index that ranks them.
The honest caveat
This is a product under active development, and its behaviour has changed repeatedly since launch. Any specific number here is a snapshot. The direction, more synthesis and fewer clicks, has been consistent.
Getting started
Publisive places clients in established publications with the search history these systems draw on. We write the piece to your brief, you approve every word, and it stays indexed.
We write and place features in the publications our clients choose. About us
Frequently asked questions
01How often do AI Overviews appear?
Tracking estimates put them on roughly one in three US desktop queries overall, and considerably higher on informational searches, where some analyses put the figure near 60 percent.
02How many sources does an AI Overview cite?
Google's AI Mode attaches far more than the chat assistants, in the range of 15 to 22 per answer depending on the vertical. That is about three times what ChatGPT, Gemini or Copilot return.
03Does being cited in an AI Overview bring traffic?
Often less than you would expect. A user reading a synthesised answer with twenty source chips frequently clicks none of them, so inclusion and visits are not the same measurement.
