Skip to content
The Chat Archive

Contribute a research collection

Research groups can contribute consented, appropriately licensed AI conversations while preserving provenance and contributor choices.

Prepare a collection

  1. Document who collected the conversations, the collection dates, source systems and exact model names where known. Leave unidentified models unknown.
  2. Provide the license, evidence of redistribution permission, participant consent, and any restrictions. An export or a public share link alone does not establish redistribution rights.
  3. Remove personal information, credentials, restricted material and third-party content you cannot redistribute. Include a description of privacy checks and known limitations.
  4. Use JSONL with one conversation per line and role/content messages. Include stable source identifiers. Keep different licenses in separate collections.
  5. For a pilot, use the normal private-draft import and consent flow. For bulk collections, open a proposal in the project repository containing metadata and a public license link only; an archive administrator will review provenance before scheduling any import. Do not attach private transcripts or consent records to a public issue.

Review and maintenance

Collections undergo license, privacy, duplicate and quality review. Imported public datasets do not earn individual Research Points. Keep an institutional contact and a process for upstream removals. Stable source IDs let us suppress removed records and exclude them from later releases; copies already downloaded cannot be recalled.