My Friend Is AI — aggregate dataset
The numbers behind the charts at myfriendisai.com, as plain CSV and JSON. Data through 2026-09; the in-progress calendar month is excluded.
Derived counts only — no post text, no titles, no usernames, no post IDs.
| File | Rows | What it is |
|---|---|---|
| monthly_theme_counts.csv · .json | 250 | Per theme per month: published keyword count, denominator, rate per 1,000. |
| monthly_community_volumes.csv · .json | 1,755 | Per tracked community per month: post volume, with tier. |
| README.md | — | Column-by-column schema, provenance, and how to read the numbers. |
| METHODOLOGY.md | — | Standalone statement of scope, method, validation, and limits. |
| manifest.json | — | Version, generation timestamp, row counts, SHA-256 hashes. |
Read these as a floor, not a ceiling. The keyword instrument is precision-first: a hand-coded audit put per-theme recall between 0% and 32%. Shape and timing are approximately honest; magnitude is an undercount, and it is uneven across themes, so theme heights are not comparable to each other. METHODOLOGY.md and the site's About page state the limits in full.
Licensed CC BY 4.0. Cite as: My Friend Is AI (2026). Reddit discourse tracker for AI companionship communities — public aggregate dataset v1. myfriendisai.com. Code and validation records: github.com/hopeshub/myfriendisai.