We analyzed 700,000 ChatGPT conversations containing web citations from Q4 2025. The dataset spans millions of individual citations and covers every major topic category. Here's what we found.
Turn 1 dominates citation activity
The vast majority of web citations in ChatGPT conversations occur in Turn 1 — the AI's first response to the user's initial question. As conversations continue, citation rates drop off sharply. By Turn 3, most responses contain no web citations at all. This means your most important optimization target is the cold-start query: the first question a user types, before any context has been established.
Wikipedia is the default knowledge layer
Wikipedia remains the single most-cited source across nearly every topic category. For factual, encyclopedic information, ChatGPT treats Wikipedia as the ground truth. Brands that want to appear in AI-generated factual summaries should ensure their Wikipedia presence is accurate, comprehensive, and properly sourced.
Sources cluster tightly by topic
For specialized topics, ChatGPT draws from a surprisingly small pool of high-authority sources. In fintech, the same 15-20 domains account for the majority of citations. In healthcare, a similar pattern emerges around major health publishers and research institutions. This means breaking into the citation pool for your category is achievable — the pool isn't as large as it might seem — but requires establishing genuine topical authority.
What this means for your AEO strategy
- Optimize for cold-start queries — the most common search patterns in your category
- Build topical authority in your specific niche to enter the high-authority citation cluster
- Invest in Wikipedia presence for foundational brand and category information
- Fresh, frequently updated content gets cited more in ongoing conversations