ChatGPT retrieves far more pages than it actually cites: Study

AI citations in ChatGPT are far more concentrated than citation distributions in traditional search. Roughly 30 domains capture 67% of citations within a topic.

• That’s according to Kevin Indig’s latest study, which also found that broad topical coverage, long-form pages, and cluster-based models outperform the old “one keyword, one page” approach.

The details. Citation visibility wasn’t evenly distributed. In product comparison topics, the top 10 domains accounted for 46% of citations; the top 30, 67%.

• AI visibility was slightly less concentrated than classic organic search, but still highly centralized.
• Indig’s conclusion: you’re effectively shut out unless you build enough authority to win one of a limited number of citation “seats.”

What changed. Ranking No. 1 in Google still matters, but it’s not enough. Of pages ranking No. 1, 43.2% were cited by ChatGPT — 3.5x more often than pages beyond the top 20.

• ChatGPT retrieved far more pages than it cited. AirOps found that it retrieved ~6x as many pages as it cited, and 85% of the retrieved pages were never cited.
• A third of the cited pages came from fan-out queries, and 95% of those had zero search volume.

Why we care. Publishing the “best answer” for one keyword isn’t enough. ChatGPT rewards domains that cover a topic from multiple angles, not pages optimized for isolated terms. And discovery often happens outside the keyword universe you track.

The patterns. Longer pages generally earned more citations, with variation by vertical. The biggest lift appeared between 5,000 to 10,000 characters. Pages above 20,000 characters averaged 10.18 citations vs. 2.39 for pages under 500.

• This pattern broke in Finance, where shorter, denser pages often outperformed long guides. In Education, Crypto, and Product Analytics, longer pages continued to gain citation value with little drop-off.
• 58% of cited URLs were cited only once. Pages that recurred across prompts were usually category roundups, comparison pages, or broad guides answering multiple related questions.

On-page behavior. ChatGPT cited heavily from the upper part of a page. The 10% to 20% section performed best across all industries.

• The bottom 10% earned just 2.4% to 4.4% of citations. Conclusions were largely ignored.
• Finance had the steepest ramp, with 43.7% of citations in the first 30%.
• Healthcare and HR Tech were flatter.
• Education peaked later, around 30% to 40%.

About the data. Indig analyzed ~98,000 citation rows from ~1.2 million ChatGPT responses (Gauge), isolating seven verticals. The study used structural page parsing, positional mapping, and entity and sentiment analysis to identify which pages earned citations and where they come from.

The study. The science of how AI picks its sources

View the full article