Research / Edition one / Dataset
Everything behind edition one. Published under CC BY 4.0 — use it, quote it, rescore it and publish a different answer.
| File | What it holds | Rows |
|---|---|---|
| …edition-one.pdf | The full report, 5 pages, as published | — |
| raw_publishable.csv | Every response in full, with prompt, engine, run, sources and classifications | 480 |
| warnings_classified.csv | Every warning sentence with its hand-assigned verdict and the reasoning | 497 |
| prompts.json | All 40 prompts, exactly as they were sent | 40 |
| surface_rates.csv | M1 and M2 per engine, stage and scenario | 80 |
| default_set.csv | M4, every product mention by engine | 239 |
| sources.csv | M6, every cited domain with its classification | 1108 |
| index-methodology.md | The study design, written before collection began | — |
| index_score.py | The scoring script, with every correction documented in comments | — |
| classify_warnings.py | The warning classification, as rules plus per-sentence overrides | — |
| FINDINGS.md | The definitive numbers, regenerated from the raw file | — |
If that redirects you back to the report, the gate has not seen you yet — leave an email there and it opens. One address, one report, unsubscribe in a click.
Citation: Kunzum, Crypto AI Visibility Index, edition one, September 2026.
https://kunzum.xyz/research/does-ai-recommend-crypto/