How every number is made

Version 0.1.0-draft. Artist audience data is from the snapshot of 2026-10-28. Playlist tracklistings were read on 2026-11-09. Playlists change week to week, so every playlist figure is a photo of that date.

Data sources

• Chatmu Big Data. 15,057,201 artists (name, country, city, genres, career stage), monthly audience snapshots from March to October 2026 (Spotify listeners and followers, Instagram, TikTok, YouTube and others) and 253,659 Spotify playlists.
• Playlist and catalog data. Track lists of editorial playlists, song and artist metadata and social profiles, used for the fake-artists analysis.
• Jeff (classification model). Jeff is an open, 0.8B-parameter model that makes narrow yes/no decisions with a calibrated confidence and the option to abstain, and runs locally in milliseconds. The published fake-artist figures use the transparent rules described below. Jeff is the planned second layer for ambiguous cases (empty or generic biographies, photos that do not show a person) and is not part of this edition's numbers.
• Public reports. IFPI (market size), Deezer (AI uploads and streams), Luminate (charts and musicians' attitudes), Spotify (AI policies). Each figure on the AI page names its source and date.
• Chatmu usage. What artists ask Chatmu's AI team, counted by distinct external users, without names. Team and test accounts are excluded.

The Industry Map

Chatmu indexes 15 million artists, but not all have the same detail. Around 2.1 million have a country, about 1.9 million have a genre and almost one million have complete audience data. The map shows each group at the level it supports, and says so.

• By country: artists sit around their country's centre, or on their city when the city is known and recognised (about 32,000 artists).
• By genre: genre families are placed by how often they share artists, and sub-genres are packed inside them.
• By career stage: one column per stage; the height is the artist's audience rank inside the stage. “Emerging” groups the artists that have no career stage in the source: they are very small (median 531 monthly listeners, 99% under 12,000) and have no country on record.
• By platform: one row per platform (25 platforms with audience data: Spotify, TIDAL, Deezer, YouTube, Amazon, Instagram, SoundCloud, TikTok, X and more). Each artist appears once, on the platform where they rank highest compared with the other artists there; the position is their audience on that platform (log scale).
• Social → streaming: social followers (Instagram + TikTok + YouTube) against Spotify listeners, both on a log scale.
• Not shown: an artist with no number for a view is left out of that view, never drawn at zero. Positions inside a country or genre are illustrative; the counts are real.

Fake artists (PFC)

Perfect Fit Content is music made to fit a playlist mood. There is no official count, so we measure a conservative floor with transparent rules over 545 editorial background playlists (lo-fi, sleep, focus, study, piano, ambient, instrumental, jazz and similar).

• High confidence: at least 50,000 monthly listeners, at least 100 listeners per follower, almost no biography, and either a link to a provider documented in public reporting or at least 500 listeners per follower.
• Medium confidence: at least 20,000 listeners, at least 30 listeners per follower, and either that provider link, no social profiles, or at least 100 listeners per follower.
• Excluded: major labels, and artists who do not appear mostly in instrumental playlists. The instrumental test uses audio analysis of the tracks.
• Estimated income: monthly listeners × 3 streams per listener × $0.0035 per stream, shown as a range (2–5 streams, $0.003–0.005). Yearly streams are the weekly figure × 52.

“PFC-like” means an artist matches the pattern. It is not a claim about any individual, a label or a distributor. Many labels and distributors also carry ordinary artists.

Limits to keep in mind

• Monthly listeners are not streams, and not unique people.
• Income is an estimate in a range, not accounting data. Rates vary by country, plan and fraud.
• Yearly streams are annualized from about four weeks of history.
• Some social profiles may be missing in the source, which can make an artist look like they have none.
• The audience snapshot covers about one million artists, not the whole index.
• AI figures come from third parties and measure different things (uploads, streams, charts). They are not directly comparable.

Corrections

If you see an error or want a correction, tell the Chatmu team through chatmu.io. We update the data and note the change.