A commercial recommendation prompt. We size how many people ask ChatGPT this from real search demand, then run it through ChatGPT to see which websites it recommends.
One of the 5 times we asked ChatGPT this, verbatim. The leaderboard below aggregates all 5 runs.
Aggregated across 5 runs (min 2/5 appearances). Estimated visits split the prompt's demand by rank. Sites matching the prompt's own brand are marked branded and excluded from the leaderboard.
| Rank | Website | Appears | Est. visits / mo | Est. value / mo |
|---|---|---|---|---|
| 1 | polymarket.com | 5/5 | 151 | $344 |
| 2 | kalshi.com | 5/5 | 91 | $206 |
| 3 | manifold.markets | 5/5 | 61 | $137 |
| 4 | metaculus.com | 5/5 | 42 | $94 |
| 5 | predictit.org | 5/5 | 30 | $68 |
| 6 | augur.net | 5/5 | 23 | $51 |
| 7 | goodjudgment.com | 4/5 | 17 | $37 |
| 8 | hypermind.com | 2/5 | 13 | $29 |
| 9 | cftc.gov | 2/5 | 10 | $22 |
| 10 | betfair.com | 2/5 | 8 | $17 |
| 11 | smarkets.com | 2/5 | 6 | $13 |
| 12 | fivethirtyeight.com | 2/5 | 5 | $10 |
| 13 | lesswrong.com | 2/5 | 4 | $8 |
ChatGPT doesn't publish prompts, so we don't show logged conversations. We measure real demand for this topic from Google search volume, scale it to ChatGPT's user base (on average about 1 ChatGPT user for every 8 Google searchers, a ratio that varies by topic), and run the prompt through ChatGPT 5 times to record which websites it recommends and in what order. Estimated visits split that demand across the recommended sites by a click-through-by-rank curve. Every number is a modeled estimate, not measured analytics. Full methodology →