
Can you trust the AI summaries that appear at the top of Google search results? A study by researchers at Washington University in St. Louis, presented in October 2026 at the ACM Internet Measurement Conference, gives a precise answer. Out of nearly 100,000 claims analyzed in Google AI Overviews, 11% are not confirmed by the very sources Google cites. For a small business leader who increasingly relies on AI to get informed or make decisions, that number deserves a closer look.
At a glance
- Researchers at Washington University in St. Louis (Jacob Montgomery, Umar Iqbal, Haofei Xu) analyzed 98,020 claims from Google AI Overviews over 40 days of searches.
- About 89% of verifiable claims are confirmed by their cited sources, but 11% are not: 7% could not be found in the cited page, 4% were directly contradicted.
- Only 41.9% of AI summaries are "fully grounded," meaning every claim in them is supported.
- AI summaries appear in 13.7% of Google searches, and up to 65% of searches phrased as a question.
- For a small business, the lesson is not to stop using AI to search for information, but to systematically verify any numeric, legal or technical claim before acting on it.
A large-scale study of Google's AI summaries
Google AI Overviews is the AI-generated summary that appears above classic search results for many queries, with links to the pages used as sources. For a business leader typing a question into Google, it is often the first thing they read.
The research team wanted to measure the real reliability of these summaries, not just their presence. Between March 13 and April 21, 2026, they ran 55,393 Google searches on trending topics, collected every AI summary produced, its citations, the regular results shown alongside it, and the full content of the cited pages. In total, the team examined 61,212 citations and broke the summaries down into 98,020 verifiable claims, each checked against the exact text of the cited page.
What the numbers show: high reliability, but not total
The headline result is reassuring, halfway. Nearly 89% of verifiable claims are clearly or broadly supported by the page Google cites as a source. That is a solid score. But the remaining 11% break down into two different problems: 7% are claims the researchers could not find anywhere in the text of the cited page, and 4% directly contradict what that page says.
The authors themselves note that this 11% figure should be read as a ceiling, not a final average: their automated method could not fully analyze user-generated content, such as Reddit discussions or YouTube videos, which sometimes serve as sources.
The most telling figure for a busy executive may not be that one. It is the share of fully grounded summaries: only 41.9% of the AI summaries analyzed contain zero unconfirmed claims. In other words, fewer than one in two AI summaries is reliable from top to bottom. A summary can be broadly accurate while still containing one invented or distorted detail, with nothing to flag it to the reader.
Behavior that varies by type of search
The study also shows that AI summaries do not appear at random. They are far more common on question-style searches (65% of these trigger one, versus 9.5% of other searches) and on longer queries. They also vary by topic: very common for hobbies (46.1%) or science (39.9%), much rarer for law (9.6%) or politics (7.5%).
Another detail matters for a small business working on its online visibility: 29.8% of domains cited in AI summaries did not even appear in the regular first-page results. Being cited in an AI summary and ranking well on Google are not the same battle.
Classic Google result
Citation in an AI summary
Key takeaway
A small business that wants to appear in AI summaries needs to work on its content differently from classic SEO: direct answers early on the page, named and sourced figures, sentences that still make sense out of context. This is exactly the principle behind GEO (Generative Engine Optimization), the counterpart to SEO for generative engines.
Why this matters directly to small business leaders
A business leader increasingly turns to ChatGPT, Gemini or a Google search to settle a quick question: a tax rule, a market figure, an HR best practice. The instinct is understandable, and the speed is real. But the study points to a structural limit: an AI summary can look built on solid sources while still containing one unverified claim, with no visual signal to set it apart from the rest.
The researchers also flag an economic issue beyond reliability alone: more than half of cited pages carry advertising, and Google sometimes places its own ad above the AI summary. Jacob Montgomery, one of the study's authors, sums up the stakes for content publishers: over time, Google and the sites it cites will need to work out a licensing or revenue-sharing arrangement, or the sites that AI summaries depend on may not survive.
| Risk for a small business | Concrete consequence | Fix |
|---|---|---|
| Unverified figure or rule taken from an AI summary | Decision made on a false basis (pricing, legal obligation, market data) | Verify through the cited source or an official source before any decision |
| Confusing AI summaries with classic search results | Wrong read on your own online visibility | Track SEO ranking and AI summary citations separately |
| Company content poorly reflected or missing from AI summaries | Lost visibility among prospects who go through AI | Structure site content as direct, sourced answers (GEO) |
How a small business can act on this today
Verify before deciding
Train teams on this reflex
Work on content for GEO
Track your presence in AI summaries
FAQ
What is Google AI Overviews?
Google AI Overviews is the AI-generated summary Google displays above classic search results for certain queries, with links to the web pages used as sources.
Are Google's AI summaries reliable?
According to the Washington University in St. Louis study, 89% of verifiable claims in AI summaries are confirmed by their cited sources, but only 41.9% of summaries are fully grounded, with zero unconfirmed claims.
Should a small business stop using AI to search for information?
No. AI remains a real time-saver for exploring a topic. Caution is mainly needed for numeric, legal or technical information used in a decision: it should be verified through the cited source before being used.
What is GEO and why does it matter to small businesses?
GEO (Generative Engine Optimization) is the practice of optimizing content so it gets picked up by generative engines like ChatGPT, Gemini or Google's AI summaries. It complements classic SEO by favoring direct answers, sourced figures and sentences that make sense out of context.
Conclusion
The Washington University in St. Louis study does not condemn Google's AI summaries: nine claims out of ten are reliable. But it is a reminder that a convincing summary is not always accurate from top to bottom, and that a small business relying on it to decide should keep the habit of checking the source. It is also a useful reminder for any company that wants to show up in these summaries: AI visibility is built on the same rules of rigor. To go further on AI assistant reliability, read our article on chatbot reliability against misinformation, or explore all of our AI resources for small businesses.


