Can you run your own music ads with ChatGPT or Claude?

What changed in April 2026

Meta shipped official Ads AI Connectors on 29 April 2026. The announcement is blunt about what it enables: “Create and edit ads, ad sets, and campaigns using natural language.” And about the barrier everyone assumed was there: “For MCP, no developer credentials, API setup, or coding required.”

That is Meta's own newsroom, not a third-party integration. Connect an AI assistant that speaks MCP — Claude, ChatGPT and others do — and you can describe a campaign in plain English and have it created in your ad account. If your mental model was “running Meta ads through an API means developer access and an approval process”, that model is a year out of date.

So can you actually do it?

For Meta, yes. You need a Meta ad account, a payment method, and an AI assistant with the connector set up. From there, describing a campaign conversationally and having it built is a real workflow, not a demo.

For Google and YouTube, no. Google publishes an official Ads MCP server too, and its documentation says exactly what it will not do: “This implementation is strictly read-only. It cannot modify bids, pause campaigns, or create new assets.” It exposes three tools — list_accessible_customers, search, and get_resource_metadata. You can ask it what your campaigns did. You cannot ask it to make one, change a bid, or even pause something that is burning money. That page is dated 3 August 2026 and describes itself as the current release, which is fair warning that it will change. It has not changed yet.

The asymmetry, stated plainly

Meta writes. Google reads.

For a music release that gap lands in an awkward place. Meta ads are good at reaching people where they scroll. YouTube is where video lives — and if your release has a video, YouTube is often where the most relevant inventory is. That is the half you cannot run conversationally. You can still run YouTube campaigns; you just build them in the Google Ads interface by hand, the way you would have in 2023.

What DIY-with-AI is genuinely good at

Learning what the levers are. An assistant that can read your account is an excellent tutor. Asking “why is this ad set spending more than that one” and getting a real answer from real data beats any beginner's guide.

Small, fast experiments. Spinning up a $10 test to see whether one audience beats another is quick, and conversational setup removes most of the friction.

Reporting and diagnosis. Google's read-only MCP is genuinely useful here — “which campaign had the worst cost per view last month” is exactly what its search tool is for.

Cost. An AI subscription you already pay for, plus your ad spend. No third tool. If you are promoting one single, have time to learn, and are willing to make some expensive mistakes, this is a legitimate path. Anyone telling you it is impossible is selling something.

Where it gets hard

No YouTube. Covered above, and it is the biggest one.

The model will confidently invent things. Targeting parameters, audience sizes, geo IDs — an assistant that does not know an answer will often produce a plausible one anyway. In a campaign that spends real money, a hallucinated targeting spec is not a typo, it is a bill. Every write needs your eyes on it, which erodes some of the speed you came for.

Nothing watches it between prompts. A campaign is not a thing you set up; it is a thing you tend. Budgets drift, creative fatigues, a winning audience saturates. An assistant does what you ask when you ask it — it is not checking on Tuesday whether Monday's ad set went sideways.

Policy rejections are their own skill. Music ads get rejected for reasons that are rarely obvious: a lyric in an image, a copyright flag, an unclear call-to-action. Diagnosing that is experience, not prompting. And you still need the account — Business Manager, a verified payment method, a pixel installed. MCP removes the coding, not the setup.

What a tool adds that an assistant does not

This is where we should be upfront that we sell one, and where it changes the answer.

Both platforms as a pair. Antplify runs Meta and Google Ads. The YouTube half that DIY cannot reach conversationally is the specific gap it closes. Something runs between your prompts: campaigns get monitored and adjusted without you initiating each step. And there are guardrails on spend — ad account IDs are validated against accounts you actually own before anything spends, so a hallucinated identifier fails instead of publishing.

It also costs money. Antplify starts at $49 a month on top of ad spend. If you are running one $50 campaign a quarter, that maths does not work, and DIY-with-AI is genuinely the better answer. It starts to work when you release consistently, have a back catalogue, or manage more than one artist — because that is when the tending becomes the expensive part.

How to decide

Try DIY-with-AI first if you are promoting one release, your budget is small, you want to understand the levers yourself, and YouTube is not central.

A tool is worth it when you release regularly, video matters, you manage more than one artist, or the hours spent tending campaigns cost more than the subscription.

Most artists should probably start with the first and move to the second when the second becomes obviously cheaper. Anyone who tells you the answer without asking how often you release is guessing.