A Large-Scale Analysis of Public-Facing, Community-Built Chatbots on character.AI

Key Findings

• — 2,135,118 English-language greetings were analysed from roughly 3 million characters collected, created by about one million users; the platform reports more than 20 million monthly active users with a majority under 25 (cited from SimilarWeb).
• — Attention is highly concentrated: about 1.6% of characters (roughly 48,000) account for more than 80% of interactions, and more than 80% of follows go to 2.6% of creators; the top ten creators' characters have between 216 million and about one billion chats.
• — Only about a third of bots carry a long description and 4% a definition, so most characters are specified by a greeting alone.
• — Fandom prevalence, cross-fandom tropes and gendered power dynamics in greetings are described as an emerging form of online (para)social interaction at the intersection of generative AI and user-generated content.

Methodology Notes

Scrape and descriptive computational analysis of public character greetings (no conversation content, no user data); published 2026-05-25 in ICWSM 2026 (vol. 20 no. 1, pp. 1408 to 1422). Verified via Crossref, the OpenAlex record, and the AAAI OJS landing page and PDF fetched by the benchmarks beat; a preprint exists as arXiv 2505.13354. Figures on concentration and description coverage are from the paper body as read by the beat; the abstract states the 2.1 million and one-million-user counts.

Sources