/

Research

/

3 min read

What AI engines cite: findings from 1.2M prompts in Q2

We tracked 1.2M prompts across six engines in Q2. Most citations went to third-party pages, and engines agreed on sources far less often than expected.

Priya Raman

Head of research

Low-angle view of a grey building facade

Across 1.2M prompts we tracked in Q2, answers from the six engines in Cited cited 612,000 unique domains. Each prompt ran three times a week on ChatGPT, Perplexity, Gemini, Claude, Copilot and Google AI Overviews, across B2B software, commerce, finance, health and travel. This report covers where the citations came from, how often the engines agree, and what that means for brands.

Most citations point to someone else’s page

For category prompts, such as “best expense software for a small team,” 71% of citations went to third-party pages. Brand-owned pages took the remaining 29%. On brand prompts the split reversed, with 64% of citations pointing to the brand’s own site.

  • Review and comparison sites: 27% of category-prompt citations

  • Publisher and media articles: 19%

  • Forums and community threads: 14%

  • Documentation, video and other sources: 11%

  • Brand-owned pages: 29%

The engines rarely agree

For the same prompt in the same week, any two engines cited at least one domain in common 38% of the time. ChatGPT and Perplexity overlapped most, at 46%, while Google AI Overviews averaged 29% overlap with the other five. A page cited on one engine is a poor proxy for visibility on the others, and a single blended score can hide a brand that is strong on two engines and absent from four.

Recent pages are cited more often

Of the pages cited, 58% had been published or visibly updated in the previous 12 months, including for evergreen questions. The skew was strongest on Perplexity and Copilot. It was weakest on Claude, where older reference pages appeared more often.

What to do with this

  1. Map the third-party pages cited for your category prompts, and start with the ones where you’re missing or described inaccurately.

  2. Track Share of Answer per engine before you blend it.

  3. Show a genuine last-updated date on pages you expect to be cited, and change it only when the content changes.

The full breakdown by industry and engine is available to Cited customers in Citation Sources. We’ll publish the Q3 update once the quarter closes.

Run these methods on your own prompts

Start a 14-day trial and track 150 prompts across ChatGPT, Perplexity and Gemini.

Run these methods on your own prompts

Start a 14-day trial and track 150 prompts across ChatGPT, Perplexity and Gemini.

© 2026 Cited

Built with Framer

Create a free website with Framer, the website builder loved by startups, designers and agencies.