From our corpus
What the AI engines actually cite
Counted from the answers we have stored, per engine. Every figure carries the number of answers behind it and the dates they were sampled. A source domain is named only when an engine reached for it across several unrelated sites; one site’s rival appearing in one site’s answers is that site’s business, not the engine’s habit.
Generated 2026-10-22. Floors: an engine needs 100 stored answers; a domain is named at 20 answers across 3 sites. Download as JSON
ChatGPT
1497 answers · 5 sites · 2026-07-25 to 2026-10-21
- Answers that cite a source
- 91%
- of 1497 answers
- Citations per answer
- 6.6
- mean, over all 1497 answers
- Distinct domains cited
- 2,064
- across those answers
- Citations to the site being asked about
- 3.9%
- of all citations on this engine. The rest went to somebody else.
| Domain | Answers | Share | Sites |
|---|---|---|---|
| reddit.com | 233 | 16% | 5 |
| techradar.com | 109 | 7% | 4 |
| worldmetrics.org | 77 | 5% | 3 |
| itechguides.com | 60 | 4% | 3 |
| zipdo.co | 54 | 4% | 3 |
| toolradar.com | 53 | 4% | 3 |
| apple.com | 38 | 3% | 3 |
| g2.com | 37 | 3% | 4 |
| cruelx.com | 33 | 2% | 3 |
| arxiv.org | 32 | 2% | 4 |
| crowdreply.io | 30 | 2% | 3 |
| toolchase.com | 29 | 2% | 3 |
| cituna.com | 26 | 2% | 3 |
| google.com | 26 | 2% | 3 |
| axios.com | 25 | 2% | 5 |
Google AI Mode
1418 answers · 18 sites · 2026-08-11 to 2026-10-21
- Answers that cite a source
- 97%
- of 1418 answers
- Citations per answer
- 10.0
- mean, over all 1418 answers
- Distinct domains cited
- 3,311
- across those answers
- Citations to the site being asked about
- 2.2%
- of all citations on this engine. The rest went to somebody else.
| Domain | Answers | Share | Sites |
|---|---|---|---|
| youtube.com | 593 | 42% | 17 |
| reddit.com | 419 | 30% | 18 |
| google.com | 244 | 17% | 17 |
| linkedin.com | 144 | 10% | 15 |
| adobe.com | 143 | 10% | 5 |
| canva.com | 128 | 9% | 4 |
| medium.com | 116 | 8% | 14 |
| facebook.com | 111 | 8% | 13 |
| quora.com | 89 | 6% | 12 |
| apple.com | 85 | 6% | 9 |
| trustpilot.com | 72 | 5% | 9 |
| instagram.com | 68 | 5% | 14 |
| yahoo.com | 56 | 4% | 15 |
| g2.com | 46 | 3% | 8 |
| hubspot.com | 46 | 3% | 9 |
Claude
1384 answers · 5 sites · 2026-07-25 to 2026-10-21
- Answers that cite a source
- 98%
- of 1384 answers
- Citations per answer
- 8.9
- mean, over all 1384 answers
- Distinct domains cited
- 1,202
- across those answers
- Citations to the site being asked about
- 3.6%
- of all citations on this engine. The rest went to somebody else.
| Domain | Answers | Share | Sites |
|---|---|---|---|
| medium.com | 170 | 12% | 4 |
| alternativeto.net | 143 | 10% | 4 |
| trustpilot.com | 127 | 9% | 4 |
| g2.com | 96 | 7% | 3 |
| scam-detector.com | 64 | 5% | 4 |
| worldmetrics.org | 61 | 4% | 4 |
| wp101.com | 47 | 3% | 3 |
| seranking.com | 46 | 3% | 3 |
| aioseo.com | 45 | 3% | 3 |
| maxaeo.ai | 45 | 3% | 3 |
| frase.io | 44 | 3% | 3 |
| aeoengine.ai | 43 | 3% | 3 |
| therankmasters.com | 42 | 3% | 3 |
| gumroad.com | 39 | 3% | 3 |
| rankability.com | 39 | 3% | 3 |
Perplexity
1294 answers · 12 sites · 2026-07-13 to 2026-10-21
- Answers that cite a source
- 100%
- of 1294 answers
- Citations per answer
- 17.6
- mean, over all 1294 answers
- Distinct domains cited
- 2,129
- across those answers
- Citations to the site being asked about
- 2.9%
- of all citations on this engine. The rest went to somebody else.
| Domain | Answers | Share | Sites |
|---|---|---|---|
| reddit.com | 751 | 58% | 10 |
| youtube.com | 381 | 29% | 7 |
| linkedin.com | 320 | 25% | 10 |
| facebook.com | 163 | 13% | 4 |
| gitnux.org | 163 | 13% | 4 |
| adobe.com | 142 | 11% | 3 |
| worldmetrics.org | 139 | 11% | 4 |
| trustpilot.com | 117 | 9% | 6 |
| canva.com | 112 | 9% | 3 |
| google.com | 104 | 8% | 9 |
| f6s.com | 101 | 8% | 4 |
| medium.com | 100 | 8% | 7 |
| yahoo.com | 100 | 8% | 6 |
| zipdo.co | 98 | 8% | 3 |
| g2.com | 96 | 7% | 6 |
What this corpus cannot say
- The sites in this corpus are the ones our customers and we chose to measure. They are not a random sample of the web, and nothing here generalises to sites unlike them.
- Each answer was sampled for one site's own buyer questions, so the engines were asked what those sites' buyers ask, not what the public asks.