Even Nike Gets Out-Cited: What Five Categories Show About Where AI Gets Its Answers
Published October 12, 2026 · Adam Rock
The short version
We ran 30 commercial prompts across five categories for five days and measured every source the AI engines cited. The pattern from our earlier cross-brand study holds: the sources AI cites most are not brand websites. In running shoes, two review sites out-cited Nike by more than ten to one.
Even Nike gets out-cited
| Source domain | Citations |
|---|---|
| runnersworld.com | 143 |
| runrepeat.com | 131 |
| nike.com | 11 |
This is the part people push back on. Our buyers are different, they say. In a category where the buyer is a runner rather than a procurement team, the same thing happens: the AI cites the people who tested the shoe, not the company that made it.
Why this happens
An AI answer is assembled from sources that already corroborate a claim in public. A product page asserts. A review, a test thread or a comparison evidences. When the engine chooses what to stand behind, it leans on the source that is not the interested party.
The engines do not agree with each other
| Engine | Distinct domains cited |
|---|---|
| Perplexity | 349 |
| Gemini | 282 |
| ChatGPT | 190 |
For identical prompts, Perplexity surfaced about 1.8 times as many distinct domains as ChatGPT. If you watch only one engine, you are watching a fraction of the sources that decide the answer.
What to do with it
- Get named in the reviews, tests and roundups that already rank for your category. That is the evidence layer the engines reuse.
- Build a footprint where the citing happens: video, forums and honest comparisons, not just owned pages.
- Make your own pages machine-readable, so when an engine does cite you it can quote you cleanly.
- Track it per engine, because coverage differs and one blended number hides where you are missing.
What this does not claim
This is our measurement, not a survey of the whole market. The window is five days, so treat it as a first look. The eight-engine figures apply to one CRM prompt. A citation is not an endorsement and not a recommendation.
The dataset
The full tables, method and limits are published at the five-category citation-sources dataset
Frequently Asked Questions
Does this mean I should stop writing for my own site?
No. Your own pages still matter, especially when an engine cites you and needs to quote you cleanly. The point is that your site is not usually what gets you cited in the first place.
Which categories did you measure?
Five: CRM, AI visibility tools, running shoes, e-commerce and project management. Each category ran six prompts.
Is the sample big enough to trust?
It is a first look, not a long-term trend. The window is five days and the panel is 30 prompts, so treat the direction as the signal rather than the exact counts.