The WashU study presented at the October 2026 ACM Internet Measurement Conference is the broadest independent audit of Google AI Overviews to date. Among other things, it tells us exactly when AI Overviews appear and when they do not.
The researchers ran 55,393 Google searches on trending topics over 40 days, from March 13 through April 21, 2026. Here is what they found about activation.
The headline number
AI Overviews appeared in 13.7% of all searches. That sounds low until you realize it covers every type of query, including navigational searches where an overview would make no sense.
How query format changes everything
The split by query type is where it gets interesting:
- 64.7% of question-form queries triggered an AI Overview.
- 9.5% of non-question queries triggered one.
That is a 6.8x difference. If you ask Google a question, you get an AI answer nearly two-thirds of the time. If you type keywords, you almost never do.
Query length matters too
Among non-question searches, activation rose with query length:
- One-word queries: 9.9% activation
- Six words or more: 38.7% activation
This pattern suggests that people seeking explanations in natural language are more likely to encounter AI-generated material. Keyword searches, which have been the foundation of SEO for two decades, are the queries least likely to trigger an AI Overview.
Topic matters most
The variation by topic is the most striking finding:
| Topic | AI Overview activation |
|---|---|
| Hobby and leisure | 46.1% |
| Science | 39.9% |
| Health | (not separately reported) |
| Law and government | 9.6% |
| Political | 7.5% |
Google's triggering logic applies undisclosed editorial discretion. Political searches are six times less likely to get an AI Overview than hobby searches. This is probably a safety decision: AI-generated answers on political topics carry more risk than answers about hobbies. But Google has not published the thresholds or the criteria.
What this means for content strategy
If you are producing content and want it to surface in AI Overviews, the activation data tells you where the opportunity is:
Answer questions. 64.7% of question-form queries get an AI Overview. If your content is structured around specific questions (how, what, why, when), it is more likely to be in the retrieval pool for an overview. FAQ-style content and question-driven headings are not just good UX, they match the query format that triggers AI generation.
Write for natural language, not keywords. Long-tail natural language queries (six words or more) trigger overviews at 38.7% versus 9.9% for single-word queries. Content that answers "how to configure Supabase auth with Next.js App Router" is more likely to be in an AI Overview than content targeting "Supabase auth."
Hobby and leisure topics are the most active surface. If your content is in these categories, AI Overviews are appearing for nearly half of all searches. The competition for citations is highest here, but so is the opportunity.
Political and legal content is mostly excluded. If you produce content in these areas, AI Overviews are not a significant channel. Traditional organic search remains the primary surface.
The fan-out connection
The activation data connects to the fan-out retrieval mechanism Google describes in its documentation. When an AI Overview is triggered, the model splits the query into related sub-queries. Question-form queries naturally decompose into more sub-queries than keyword queries, which may explain why they trigger overviews more often.
This also explains the topic variation. A hobby query like "best camera for beginner photography" generates productive sub-queries: "best beginner camera 2026," "camera for photography beginners," "entry-level camera comparison." A political query like "candidate positions on tax policy" generates sub-queries that are harder to answer neutrally, which may cause the system to suppress the overview.
What the data does not tell us
The study measured trending queries, not a representative sample of all Google searches. Trending queries may skew toward topics that are more likely to trigger overviews. The 13.7% overall activation rate should be read as a snapshot of trending search behavior, not as a universal average.
The study also did not measure what happens after the overview. Click-through rates, user satisfaction, and whether users trust the AI answer were outside its scope. The publisher revenue analysis in the same paper found that more than half of cited pages carry advertising, suggesting that publishers are losing ad impressions when overviews answer questions directly.
The practical takeaway
If you want your content cited in AI Overviews, the activation data gives you a targeting framework:
- Structure content around questions, not keywords.
- Target natural language queries of six words or more.
- Prioritize topics with high activation rates (hobby, leisure, science, health).
- Cover related sub-questions within the same page, because fan-out retrieval will surface your content for sub-queries you may not be tracking.
The SEO playbook of "target one keyword, rank for it, get traffic" is not dead, but it is no longer the only playbook. The AI Overview surface rewards breadth of coverage and question-oriented structure over single-keyword optimization.
