Key point. We took seven claims that are often repeated about AI search and checked each one against our own measurements or the study behind it. None of the seven holds as usually stated. Each page gives the claim, our verdict, the numbers with their sample sizes and intervals, and how far the result reaches.
Advice about AI search spreads faster than evidence. Many of the claims below are reasonable guesses, and some are half true. We picked seven we hear often, where we had a measurement of our own or could read the original study in full, and wrote one short page for each.
Which claims did we test, and what did we find?
Each claim gets one of three verdicts, and the table gives the short version. Follow the link for the numbers, the method and the scope.
| Claim | Our verdict | The short reason |
|---|---|---|
| If you rank first on Google, the AI Overview will cite you | Not supported | The first result was cited more often than lower results, and still missed often |
| Check your AI visibility once and you know where you stand | Not supported | The same searches returned different cited sites from one check to the next |
| AI can write your thought leadership | Not supported for facts only you hold | Asked to write about facts that exist only in unpublished data, models got none right |
| Most companies block AI crawlers | Not supported for startup websites | Almost no startup sites block the crawlers AI search engines use |
| Precise statistics make content look new to AI | Supported for simple automated scores, which is the problem | An invented two-decimal figure fooled a simple novelty score |
| Social posts get you into AI Overviews | Not supported for posts on X | YouTube shows up often as a source; posts on X almost never do |
| Add quotes and statistics, as the GEO paper says, and AI engines will cite you | Could not confirm for live AI engines | The paper tested a lab engine that wrote answers from five fixed search results |
What do the three verdicts mean?
The verdicts describe the evidence we hold, not the world in general. We use three:
- Supported: our measurement agrees with the claim, within its scope.
- Not supported: our measurement contradicts the claim as it is usually stated.
- Could not confirm: we looked, and neither our data nor the study behind the claim settles it either way. Could not confirm is not the same as false.
A verdict covers only what we measured: one engine, one kind of search, one sample. Every page says which. A result on Google AI Overviews says nothing about ChatGPT, and a result on startup websites says nothing about news publishers.
How do we test a claim?
We test a claim when we hold a measurement that speaks to it, made under our standing rules, or when the claim rests on a study we can read in full; each page says which. In our research measurements we ask each question more than once, we never pool engines, we publish a finding only with its scope, and we record which model answered.
Every number on these pages is quoted in the exact wording we published it in. Each one also has a methodology card with its sample, its 95% interval where we publish one, its dates and its source page.
Frequently asked questions
Why only seven claims?
We only test a claim where we have a measurement that answers it or can read the study it rests on. Many popular claims are about things we have not measured, and we do not give a verdict without evidence.
Do these verdicts apply to every AI engine?
No. Each verdict covers the engine and the kind of search we measured, and each page names them. We never move a result from one engine to another.
Can I quote these results?
Yes. Quote the sentence as written on the page and link to it, or to its methodology card, so readers can see the sample and the scope behind the number.
What to do next
- Pick the claim that drives the most decisions on your team and read its page.
- Check whether your own AI visibility reports repeat each question before they report a change.
- Use the methodology cards when you quote a number, so the scope travels with it.
Work with us
If you want a team to measure and improve how AI search treats your company, our managed AEO services do that work with you: Book a call. If you would rather see where you stand first, Lenz gives you a free analysis of how ChatGPT and Gemini answer your buyers' questions: Get your free AI visibility report.