Until this month, “using AI for playlist pitching” mostly meant asking ChatGPT to write a nicer outreach email. That changed when AI assistants learned to use real tools. Claude, ChatGPT and other agents can now connect to live music data and do the research part of pitching for you: find playlists that actually fit your song, check whether they’re healthy, and hand you a ranked shortlist.<\/p>\n
This guide shows the exact setup we built at PlaylistMap, what a real session looks like, and where the limits are. Total setup time is about a minute, and the research layer is free.<\/p>\n
Two things, concretely:<\/p>\n
Here is a real exchange from our own test account, unedited:<\/p>\n
\nUs:<\/em> Which Spotify playlists should I pitch my dreamy bedroom-pop single to?<\/p>\n
Agent:<\/em> Found 4,932 active playlist matches. Top 3: “absolute bangers” (14.9K followers, strong fit: dream pop + indie pop match, features girl in red, updated this month), “clairo vibes” (5.3K followers), “sad sad girl” (346 followers).<\/p>\n
Us:<\/em> Before I pitch, run an integrity check on “absolute bangers”.<\/p>\n
Agent:<\/em> “absolute bangers” scores 49\/100 (Proceed with caution, medium confidence). A research signal, not a guarantee.<\/p>\n<\/blockquote>\n
Notice the second answer. The biggest playlist in the match list came back with a caution flag. That is precisely the point of putting an agent on this job: the match told us where the audience is, the check told us to look closer before investing a pitch. A human doing this manually would need to eyeball the playlist’s update history, its artist mix and its follower-to-engagement pattern. The agent did it in one message.<\/p>\n
The one-minute setup<\/h2>\n
The connection standard behind this is called MCP (Model Context Protocol), an open standard introduced by Anthropic in 2024 that lets AI assistants call external tools. You don’t need to understand it. You need one URL:<\/p>\n
https:\/\/playlistmap.com\/mcp\/server<\/pre>\nIn Claude:<\/strong> Settings \u2192 Connectors \u2192 Add custom connector \u2192 paste the URL. Done.<\/p>\n
In ChatGPT:<\/strong> Settings \u2192 Connectors (or Developer mode \u2192 custom connector) \u2192 add an MCP server \u2192 paste the same URL.<\/p>\n
No API key and no account are needed for the research itself. The full walkthrough with copy-paste configs for every client, including older desktop builds, lives on our MCP setup page<\/a>.<\/p>\n
Prompts that work<\/h2>\n
Once connected, plain language is enough. These four patterns cover most sessions:<\/p>\n
\n
- “Here’s my Spotify link: [track URL]. Which active playlists fit this song, and how big are they?”<\/li>\n
- “I make [genre] like [comparable artist]. Find playlists worth pitching and rank them.”<\/li>\n
- “Someone offered me a paid placement on this playlist: . Run an integrity check before I answer them.”<\/li>\n
- “Match every track on my EP to playlists and rank the best opportunities first.”<\/li>\n<\/ol>\n
A tip from our own testing: give the agent a comparable artist, not just a genre. “Indie pop” matches thousands of playlists; “indie pop like Men I Trust” pulls the ones whose track lists actually sit next to your sound.<\/p>\n
What live data still cannot judge<\/h2>\n
The matcher cannot determine whether your recording is competitively mixed, and catalog data cannot capture every local scene preference or curator quirk. Before pitching, listen to the shortlisted playlists yourself and compare your finished track with what they actually add.<\/p>\n
What the agent can’t do (and why that’s deliberate)<\/h2>\n
The agent never sees curator contact details, and it cannot send pitches. That boundary is enforced in the server code, not in a policy document. Two reasons. First, curator emails are the paid product that funds the free layer, and leaking them to every AI agent on the internet would kill the dataset. Second, and more importantly: many curators dislike automated outreach. A pitch that lands is short, personal and clearly written by someone who listened to the playlist. The agent’s job ends at a ranked, vetted shortlist; the pitch is yours to send from inside PlaylistMap<\/a> after you unlock the contacts you chose.<\/p>\n
Treat any tool that promises fully automated mass pitching with suspicion. Mass automated outreach is one reason curators ignore generic inbox pitches.<\/p>\n
Does the free tier have a catch?<\/h2>\n
Rate limits, and that’s it. Matching and integrity checks are capped per hour so the free layer stays free for everyone (the underlying analysis costs us real compute per request). For a working musician researching one release, the caps are hard to hit. If you want the agent to follow a complete research workflow instead of ad-hoc questions, we also publish a free downloadable skill for Claude<\/a> that scripts the whole flow: collect the song, match, vet, shortlist, hand off.<\/p>\n
The bottom line<\/h2>\n
AI agents are genuinely good at the part of playlist pitching musicians hate: the research. They are still the wrong tool for the part curators care about: the personal pitch. Wire the two together in that order and you get the 2026 version of a workflow that used to eat a weekend: a shorter research pass, then a handful of pitches you wrote yourself, sent to playlists you have screened for fit, freshness and integrity risk.<\/p>\n