How to Use AI for Music Research (Curators, Sync, and Market Scans)

Scaffolding vs sourcing: the distinction that matters

There are two ways to use AI for research and they produce very different results. The first is asking it to summarize or structure something you already have: a piece of music journalism, a label’s roster page, a set of playlist descriptions you pulled yourself. Feed it real content and ask it to find patterns, generate questions, or turn a scattered set of notes into something coherent. This it does well.

The second is asking it to provide facts about the music industry from its own knowledge: who the curators are, what their emails are, which sync supervisors are looking for what, how many followers a given playlist has. This is where it gets dangerous. The model was trained on a snapshot of the internet and fills gaps by generating text that sounds plausible. There is no flag when it crosses from memory to fabrication. A playlist name it invented sounds the same as one that exists. A supervisor’s email it made up reads the same as a real one.

The rule is simple: use AI to build a list of things to investigate, not as the investigation itself. Every specific name, email, and number it gives you is a lead. Treat it as something to check before you ever use it.

Why this matters more in music research than most places

The hallucination problem is a real risk in any AI research context, but it bites harder in music industry research for a specific reason: the stakes on a bad contact are higher than they look.

Pitching to a curator whose email address the model invented means your email goes nowhere, or worse, to the wrong person. Pitching to a sync supervisor with a job title or company that no longer exists signals that you didn’t do your homework. Music communities are smaller than they appear. One sloppy, obvious pitch can follow your name around. The cost of checking a contact before sending is thirty seconds. The cost of sending a pitch to a hallucinated contact can be harder to undo.

This is not a reason to avoid using AI for research entirely. It’s a reason to use it correctly: build the scaffold first, verify the specifics after. The scaffold itself, the structure, the angles, the questions to ask, is genuinely useful and saves real time.

Using AI for playlist curator research

Playlist curator outreach is one of the most time-consuming parts of an independent release campaign. You need to find playlists that fit your genre and audience, figure out whether they take submissions, find out how, and write a pitch that sounds like a person. AI can take some of that load, in the right places.

Where it helps: generating a list of playlist types and scenes to investigate for a given genre, drafting a framework for evaluating whether a playlist is worth pitching (size relative to genre, how recently it’s been updated, whether it takes indie submissions), and drafting the outreach message once you have a real target. Give it the curator’s actual playlist name and description, say what your track is, and ask for a first draft of a pitch. Then edit to make it sound like you.

Where it fails: as a source of curator names, emails, and submission methods. Those have to come from the actual platform. The playlist discovery work, building a list of real playlists that fit your track, has to happen on Spotify, on SubmitHub or Groover’s actual catalog, or via the real websites of independent curators. AI can structure what you do with that list. It cannot build the list reliably from scratch.

For the mechanics of pitching to independent curators and press, the full outreach playbook is in music PR without a publicist. That guide covers where to find real contacts and how to pitch them without sounding like mass outreach.

draft your Spotify editorial pitch from the real track details with the free pitch generator →

Sync licensing research with AI

Sync is a real income stream for independent artists, and the research side of it is genuinely time-consuming. There are hundreds of music supervision companies, licensing agencies, and individual supervisors across TV, film, advertising, and games. Figuring out who is looking for what, and how to reach them, is work.

AI is useful for understanding the structure of the sync market: how licensing agencies differ from supervision companies, what a sync license covers versus a master use license, what kinds of briefs tend to come up for a given genre. That conceptual scaffolding is solid ground to work from and does not require the model to recall current industry facts.

For specific supervisor contacts, the same guardrail applies as with curators. Supervisors change jobs. Companies rename. Agencies open and close. The contact information a model has in its training data may be years out of date, and it will not tell you that. For current contacts, start with actual industry directories: MUSIC x FILM’s supervisor database, Music Bed’s licensing side, the IMDbPro credits for shows you want to pitch, LinkedIn for current job titles. Build the list yourself, then use AI to draft the outreach.

For the full sync licensing picture, including what rights you need to have in order to even submit, see sync licensing for independent artists.

Scene and market research: feed it what you find

One of the more useful things AI can do for a release campaign is help you understand a scene or market you’re trying to enter. Maybe you make music in a genre adjacent to what’s breaking right now and you want to understand the landscape. Maybe you’re pitching a track for a specific sync context and want to know what’s already placed there.

The move is to bring the raw material and ask for the summary. Pull a handful of editorial playlist descriptions in that genre. Find a recent piece of music journalism about the scene. Pull a few label roster pages. Paste all of it in and ask the model to identify patterns: what do these releases have in common, what language do the curators use to describe this sound, what seems to be getting traction. That’s a legitimate and useful application.

What does not work: asking the model to describe the scene or market from its own knowledge. Its training data has a cutoff, and it tends to present its training-era picture as current. The music industry moves fast enough that a model’s internal picture of a scene can be meaningfully out of date. Feed it fresh content and ask it to work with that; don’t ask it to report on something it may not have current information about.

For the adjacent work of planning what to do with that research once you have it, including how to structure pitching across curator, editorial, and sync channels in a release campaign, see pitching beyond Spotify.

A note on contracts and legal language

AI is genuinely useful for one specific part of contract reading: plain- language explanation of what a clause means. If you paste in a paragraph from a distribution agreement and ask what it says in plain terms, a good model will give you a clear answer. That is useful. It’s faster than digging through a glossary, and it can surface the actual implications of standard boilerplate that is otherwise designed to be unreadable.

It is not a music attorney, and it cannot tell you whether to sign something. The comprehension help is real. The judgment call about any deal that involves publishing rights, master ownership, a label agreement, or an advance is a decision for a music lawyer. The cost of that call is almost always worth it on anything material.

One caution on the data side: be deliberate about pasting unsigned contracts or financial terms into third-party AI tools. Depending on the tool and its data settings, that content may be retained or used for training. Know the tool’s privacy terms before you paste anything sensitive.

when the research is done, draft your Spotify pitch from the facts with the free pitch generator →