Datasets
Each study publishes an anonymized dataset of derived features where possible. Datasets contain no customer prompts, AI responses, or customer identifiers — see the data policy. Row counts reflect the citations evaluated in each study.
A synthetic panel matched to human sub-intent reproduces the consideration set — but not the share of voice.
- Sub-intent-matched vs human prompt panels on ChatGPT (derived features) (CC BY 4.0) — 1,230 citations evaluated · datasheet
- Sub-intent-matched and unstratified scenario panels — full prompt text (CC BY 4.0) — 95 citations evaluated · datasheet
We tested our own prompt generator against 143 humans. It measures your home field, and we found what a market-matching panel needs.
- Synthetic vs human prompt panels on ChatGPT (derived features) (CC BY 4.0) — 1,325 citations evaluated · datasheet
- Synthetic prompt panels — full prompt text (CC BY 4.0) — 114 citations evaluated · datasheet
A reader asked if 'rank is unstable' was measured or folklore. So we measured it.
- Prompt-phrasing consistency of ChatGPT recommendations (derived features) (CC BY 4.0) — 1,144 citations evaluated · datasheet
We pre-registered 'prompt phrasing barely matters.' ChatGPT proved us wrong.
- Prompt-phrasing consistency of ChatGPT recommendations (derived features) (CC BY 4.0) — 1,144 citations evaluated · datasheet
Only one AI surface deep-links to the moment inside a YouTube video
- YouTube moment citations across AI answer surfaces (derived features) (CC BY 4.0) — 5,978 citations evaluated · datasheet