I counted the sources in 13 of Google's AI answers. 168 citations, and not one domain appeared twice.

I keep a daily probe that asks the assistants the buying questions in my category and records every source behind every answer. Last week I pointed it at Google's two AI surfaces, the AI Overview box and the fuller AI Mode, and counted.

Thirteen answers. 168 citations. Every single answer used a different domain for each of its sources. Not one domain was cited twice, in any answer, the whole week. Thirteen of thirteen, no repeats.

That stopped me, because it is the opposite of how a normal search results page behaves. On the ten blue links, one strong site often takes three or four of the top spots. A page that ranks well ranks well repeatedly. Here, ranking well once seemed to use up the domain for that answer.

Why an AI answer looks like this
An assistant does not hand you a ranked list. It breaks your question into several smaller ones, runs those, and assembles an answer from the best source it found for each. Google's own patent on query variants (US11663201B2) describes dynamic control that "can often lead to a relatively large (e.g., more than 5, more than 10, or more than 15) quantity of variants."

So an answer is not "the top pages for your query." It is "the one best page for each of a dozen sub-questions, gathered from wherever each happens to live." One page per sub-question. One slot per domain.

For scale: the AI Mode answers pulled 14.75 sources on average, as many as 24. The AI Overview box drew on far fewer, 6.25 on average, sometimes zero. Different surface, different appetite, same one-slot pattern.

What this changes about what you build
If you get at most one slot in an answer, then the usual SEO instinct, publish many pages on one keyword and let the strongest rank, does close to nothing here. Ten pages about "best CRM" do not buy you ten slots. They compete for the same one.

The move that fits the mechanism is the other one: be the single best page for one sub-question, and to appear in more answers, cover more distinct sub-questions. Breadth of genuinely-different pages beats depth on one term. A page that owns "how to migrate from X to Y without downtime" can win the slot for that sub-question in a way your tenth "best tool" page never will.

The honest limits, because they matter
Thirteen answers is a small sample, one question set, one week. And I cannot tell from the outside whether the one-per-domain rule is Google selecting one page per domain, or Google deduplicating what it displays after selecting more. Those are different mechanisms.

Here is why I am publishing it anyway: for someone deciding where to spend next week, the consequence is identical either way. Whether it is selection or display, you still get one slot, so you still win by covering more sub-questions rather than stacking pages on one. When the decision does not change, the caveat is worth stating out loud rather than sitting on the finding.

I found this by running the measurement on my own site, which is the only way I trust a number. If you want to check me, the quantity to reproduce is simple: take any AI answer, list its sources, and see how many share a domain. I would genuinely like to know if you get repeats where I got none.

---

I found this running my own tool, Glotier, which checks whether ChatGPT, Gemini and Perplexity name your product and what on your pages stops them from citing you. The first check is free.

Comments

[Dmitry Sergeev]

that domain diversity is wild. makes me wonder if they're just trying to avoid looking biased or if the algo is actually hitting a wider range of sites now