How to Make a Songwriting Assistant with AI with YOUR ideas and content

My last video, on using AI to rebuild a paid plugin with stock plugins, got over 50,000 views and more than 100 comments. A lot of you told me the same thing: AI gives wrong answers. You're right, it often does. This video is about why, and how to set it up so it actually helps.

One thing up front. This isn't about letting AI make music for you. Tools like Suno do that, and that's not what I'm after. This is about building a Claude project that works like a mixing and songwriting buddy. It asks questions, pushes back on your ideas, and helps you make decisions, while you stay in control of the music.

Why does AI give wrong answers?

Because it isn't thinking the way you and I do. It's pattern matching. It takes what you give it and produces the most likely next thing. If what you give it is vague, generic, or missing the details of your rig and your song, you get the generic answer, and the generic answer is often wrong for you.

That's the same problem I had with YouTube advice for years. The answer wasn't wrong for the person giving it. It was wrong for my gear, my room, and my songs.

What decides whether the answer is good or bad?

In the video I break it down to four cornerstones. All four have to be right.

• Brain. Which model you're using. They are not all the same.
• Context. What it knows about you, your gear, and the song in front of it.
• Prompt. What you actually asked, and how.
• Instructions. The standing rules it follows every time, before you type a word.

Most people only touch the prompt. Context and instructions are where the real difference is.

How is my Claude project set up?

The video walks through the whole thing. The short version:

• Instructions that describe who I am, what I'm making, and how I want it to behave.
• Skills, which are reusable jobs I've written up once, like the lyric assistant I use in the demo.
• Plugin manuals loaded in as context, so when I ask about a compressor it answers about the one I actually own, not a generic one.

How do you stop AI from just agreeing with you?

This was the biggest complaint from the last video, and it's fixable. Left alone, AI tends to tell you your idea is great. The fix is in the instructions. You tell it, in plain words, to ask questions before answering, to push back when something doesn't hold up, and to give you options with a recommendation instead of a pat on the back. I show the exact wording in the video.

What does it look like in practice?

The last part of the video is a live demo. I run my lyric assistant skill on 10 pages of my own lyrics and let you watch what it does with them: what it asks, where it pushes back, and where it leaves the call to me.

Huge thanks to everyone who commented on the last video. A lot of your comments shaped this one.

FAQ

Does AI write the music for you?

No. That's a different kind of tool, like Suno. The setup in this video is a buddy for mixing and songwriting decisions. You still write, play, and decide.

Why does AI give wrong answers about mixing?

It's pattern matching on what you give it. Without your gear, your song, and clear instructions, you get the generic answer, which is often wrong for your situation.

What are the four things that decide if AI gives a good answer?

Brain (the model), context (what it knows about you and the song), prompt (what you asked), and instructions (the standing rules it follows every time).

How do you make AI push back instead of agreeing?

Put it in the instructions. Tell it to ask questions first, challenge ideas that don't hold up, and give options with a recommendation. The video shows the wording I use.

Bottom line

AI is only as good as what you set it up with. Get the brain, context, prompt, and instructions right, tell it to push back, and it becomes a useful second opinion that still leaves the music to you. The gear it knows about is all on the gear page.