WildChat-1M
Allen Institute for AI and University of Washington
A million real ChatGPT conversations with state, country, and hashed-IP fields attached.
Conversations between users and ChatGPT gathered through a free chatbot service, released with demographic fields. One row is one conversation with its timestamps, model version, language, and request headers. It supports studies of how real use differs by place and time.
Access
- Cost
- FreeNo charge.
- License
- Open licenseODC-By 1.0
- Sign-up
- No sign-up
- How you get it
- API and Bulk downloadParquet
What's in it
- One row is
- One row per conversation
- Covers
- Global users, with state and country fields
- Years
- 2023-2024 (see card)
- Updated
- One-time release
- Size
- About 1 million conversations (100K to 1M rows)
Quality
Documented in a paper with a companion search tool. Users knew chats might be shared for research, but the sample is self-selected.
- Contains toxic content, partly flagged
- Hashed IPs and locations are privacy-sensitive
- Users of a free ChatGPT wrapper are not representative of the public
Connects to
- LMSYS-Chat-1M: Comparable real-conversation corpus across many models
- Anthropic Interviewer: Complements logs with interview transcripts about AI use
Questions it could answer
- How do request types differ by country?
- What share of chats are coding, writing, or role play?