ChatCorpus

What is ChatCorpus?

ChatCorpus is a searchable archive of five million real human-AI conversations drawn from open research datasets. Instead of guessing how people talk to AI assistants, you can read complete chats end to end, search across real prompts, and compare chosen and rejected replies from preference data to see which responses people actually favoured.

Who is it for?

• Product teams studying how users phrase requests, where conversations break down, and what people expect from assistants.
• Prompt engineers mining real prompts for patterns, phrasings, and edge cases to improve their own prompts and evals.
• Researchers exploring large-scale conversational data without wrangling raw dataset dumps.
• Writers looking for authentic examples of how people actually converse with AI.

Key features

• Full-text search across five million conversations from open research datasets.
• Complete conversation views, not just isolated snippets.
• Preference-pair comparisons showing chosen versus rejected model replies.
• Filters to narrow results by dataset and conversation attributes.

Getting started

Visit chatcorpus.com to start searching. The archive runs entirely in the browser, needs no setup, and offers free access alongside paid plans for heavier research use.