The studies read 2,989 citations on 1,615 domains across 30 niches and 14 markets. For each market, they show who the engines cite and how much of it is the business's own site.
The first study covers 30 niches. 27 of them got 4 buying questions each, and 3 were taken to 30 questions each in three bands: general, medium and specific. The second study covers 14 markets with 5 questions each. Both use ChatGPT and Google organic results, with 1 run per question in the 30-niche study. AI Overviews appeared on 12 of 198 queries and are left out of the shares.
First-party means a small provider's own site, up to an authority ceiling of 45. A vendor above the ceiling counts as an established vendor. Third-party means directories, review sites and publishers, and community means forums and social. The entry bar says how small a site can be and still get cited. For each question it is the tenth percentile score of the credible cited set (domains cited on more than one question, plus anything scoring at least as high as the weakest of those), and the bar shown is the middle value across questions. Scores come from the Peak Answer Link Ranking. Content share, used in the 14-market study, is the share of cited sources that are articles about the category. A vendor's own page does not count.
For background on how these engines pick their sources, see answer engine optimization and LLM SEO.
The engines change, so these are snapshots. ChatGPT's own repeat runs of one question overlap about 30% on cited domains. 5% of citations are on domains the authority index has not scored, and those are left out of the bars. The niches with 4 questions are directional only, so they are pooled by category and are not ranked.
| Study | Niches and markets | Questions | First-party | Third-party | Bar |
|---|---|---|---|---|---|
| Wedding photography | 1 | 30 | 58% | 13% | 47 |
| Email verification | 1 | 30 | 43% | 2% | 30 |
| Salesforce consultancies | 1 | 30 | 50% | 1% | 28 |
| Software and SaaS | 13 | 56 | 28% | 2% | 24 to 56 |
| Agencies and tech services | 6 | 26 | 53% | 4% | 32 to 95 |
| Professional services | 9 | 38 | 51% | 5% | 24 to 53 |
| Local and home services | 9 | 38 | 50% | 5% | 15 to 100 |
| Consumer brands | 3 | 15 | n/a | n/a | 16 to 56 |
Bars from the 30-niche study and the 14-market study use different question counts, so read gaps of a few points as noise. Niches with 4 questions are pooled by category.
Observational studies of what AI answers cite and how much they change. Each study has the data, the method and the limits.
A study is a snapshot from one day. The tracker asks your own buyer questions each week and shows who gets cited.