New Semrush research analyzed hundreds of prompts across ChatGPT's Instant and Thinking modes and found that only 25% of cited domains overlap between the two modes for the same prompts. That means 75% of sources ChatGPT cites in Thinking mode are entirely different from what it cites in Instant mode — for identical queries.

The behavioral difference is significant: Thinking mode performs 4.6x more web searches per query. It deprioritizes Reddit and user-generated content, and increases its reliance on official documentation, government sites, and academic sources (which grew from 12.4% to 17.5% of citations in reasoning mode).

The implication for brands: if you track ChatGPT as one channel, you are almost certainly overestimating visibility for casual browsers and missing the visibility gap with active buyers — who are far more likely to use Thinking mode for complex purchase decisions.

If you're tracking ChatGPT as one channel, you're measuring two completely different audiences as one number. A brand that's visible in Instant mode but invisible in Thinking mode is winning the casual browse and losing the buying decision. Thinking mode correlates with deeper purchase intent — it runs 4.6x more searches, queries more authoritative sources, and discards UGC signals that Instant mode values. That's not a minor variation. That's a completely different ranking contest. The immediate action: pull a representative set of your target queries and run them in both modes. If your brand disappears in Thinking mode, you need primary sources, official documentation, and entity consistency — not more blog posts.