Writing a song sounds romantic until you’re actually doing it. You’ve got a feeling, maybe a loose idea, and then, nothing. The words don’t come. The melody won’t sit right. You open a blank doc, stare at it for twenty minutes, and close it.

That’s the exact problem AI solves well. Not by writing your song for you, but by getting you unstuck fast, at every stage of the process. Here’s how to use it properly, step by step.

Step 1: Get Your Idea on Paper First

Before touching any tool, write one or two sentences about what your song is actually about. Not the genre, the meaning. Something like:

“This song is about pretending you’re fine with someone moving on when you’re really not.”

Or:

“It’s about the weird relief you feel after something finally ends, even if it hurt.”

This is your creative brief. Everything you prompt AI with later will be built on this. Without it, AI gives you technically correct lyrics that feel like they belong to nobody.

Step 2: Write Your Hook Yourself

The hook is the one line or two, that the whole song orbits. It needs to come from you because it needs to feel real. Don’t overthink it. Write something specific and true. Specific always beats poetic.

Weak hook: “I miss you every day”

Stronger hook: “I still check my phone like you’re going to text first”

The second one has an image. It puts the listener somewhere. That’s what AI can’t manufacture, it can only build around what you give it. Once you have a rough hook, you’re ready to bring AI in.

Step 3: Use AI to Build the Lyrics Around Your Hook

This is where tools like ChatGPT, Claude, or any AI writing assistant actually earn their place. The key is knowing how to prompt them so you get something usable, not something that sounds like a template.

Here are real prompts that work:

For writing a verse:

“Write a first verse for a song. The theme is about realizing you’ve been holding on to a relationship that’s already over. The hook at the end of the chorus is: ‘I still check my phone like you’re going to text first.’ The verse should build toward that feeling without stating it directly. Keep it conversational, no forced rhymes, no clichés. 6–8 lines, natural flow.”

For getting chorus options:

“Give me 4 different chorus variations for a song about quietly outgrowing someone you used to love. Each chorus should be 3-4 lines. Make them feel personal, not like a motivational quote.”

For a bridge:

“Write a bridge for this song. Verse is about missing someone. Chorus is about accepting the distance. The bridge should be the moment of honesty, where the singer admits they’re not actually okay. Short, direct, no more than 6 lines.”

Run these, read what comes back, and treat it as a first draft, not a finished product. Some lines will be good. Some will be off. That’s normal.

Step 4: Edit Like a Writer, Not a User

This is the step most people skip, and it’s the most important one.

Go through every AI-generated line and ask two questions: Does this sound like something a real person would say? Does it belong to this specific song or could it go in any song?

Anything generic gets cut or rewritten. Anything that doesn’t match your brief gets cut. What you’re doing here is curating, using AI to generate raw material, then shaping it with your own judgment until it sounds like it came from one person with one clear point of view.

A good edit pass takes 20-30 minutes. A bad one takes 2 minutes and it shows.

Step 5: Turn Your Lyrics Into a Vocal

Once your lyrics are done, you have two paths. Choose the one that fits where you are right now.

Option A: Record Your Own Voice

This is the most direct route and the one that makes the song feel most like yours. You don’t need equipment, your phone’s voice memo app records clean enough audio to work with at this stage.

Hit record, sing your lyrics straight through. Don’t aim for a perfect take. Aim to find the melody, where each line naturally wants to sit, how fast the words want to move, where you breathe. That scratch recording tells you more about your song than anything you wrote on paper.

Once you’re happy with the melody and timing, record a cleaner version. If you want to go further, adjust pitch, clean up the tone, or hear your voice in a different style, tools like Kits.ai let you upload your own vocal recording and convert it using any voice from their licensed library. So you could record your rough voice memo, run it through Kits, and hear it come back polished, pitch-corrected, and in a completely different vocal style. Still your performance, just cleaner or different in character.

Option B: Use AI to Generate the Vocal From Your Lyrics

No mic. No recording. You paste your lyrics in and an AI singer delivers them back to you as a clean acapella file. Here are three tools that do this well:

Kits.ai: Paste your lyrics, pick a voice from their ethically licensed artist library, and generate. You get three unique vocal outputs to choose from per generation. If you want to tweak the melody after, their built-in Pitch Editor lets you reshape individual notes with MIDI-style control. Free plan available. Paid plans start at $10/month.

ElevenLabs: Better known for voiceovers, but their singing voice library is genuinely capable for song demos. You can pick from hundreds of vocal styles, adjust pitch and tone, or clone your own voice if you want the AI to sing in a version of your voice. The output is a clean vocal file, no backing track.

Soundverse: Specifically built for lyric-to-vocal generation. Paste your lyrics, choose a style (pop, rock, rap, opera), and it generates a clean acapella ready for mixing. Supports reference audio too, so if you have a vocal tone in mind, you can upload a reference and the AI will match it.

All three export as WAV or MP3. Download the vocal file and keep it separate, you’ll need it clean for the next step.

Step 6: Add the Music

Now you have a vocal. What you don’t have yet is the music behind it. And this is exactly the step where most creators either give up or make poor decisions, they grab a random track from a stock library, throw it under their vocal, and wonder why it sounds like a mismatch.

The right approach is to build the music around the vocal you already have.

Before generating anything, read your lyrics out loud over silence and pay attention to:

  • Natural tempo: how fast do the words want to move? That’s your BPM direction. Conversational and close-up usually lands between 70–90 BPM. Punchy and energetic pushes 100–120.
  • Emotional weight: heavy and raw calls for minimal instrumentation. Bright and open wants more texture.
  • Genre territory: what world does the subject matter of your song live in? Don’t pick a genre for how it sounds. Pick it for whether it matches what the lyrics are saying.

Write those three things down. Now head to Mubert Render. If you’re not sure about the genre direction yet, spend a few minutes in the genre library 100+ categories across every style, which makes it much easier to identify the sonic territory you’re actually going for.

Generate a track. Then immediately play your vocal file alongside it, sing along, speak your lines over it, whatever you have. Don’t evaluate the music in isolation. Evaluate it with your words on top. That live test tells you instantly whether the energy is right or working against you.

Run multiple generations. Mubert creates a unique track every time, you’re genuinely auditioning different compositions, not cycling through the same loop. Give yourself at least 5-8 options before deciding. The one that makes your lyrics feel at home is the right one.

If you’re curious about the breadth of artists whose sounds are shaping what Mubert generates, the artist community is worth exploring, real producers and musicians contribute the samples and loops that the AI draws from, which is a big part of why the output sounds less mechanical than most generators.

Step 7: Bring It All Together

Now you have two files: your vocal and your backing track. Bring them into any free audio editor. Stack the vocal on one track, the music on another.

Make sure the vocal sits clearly above the music in the mix, it should never feel buried. Do a basic volume balance, and if the backing instruments are clashing with your voice in the mid-range, pull those frequencies back slightly in the music track. You don’t need to be an audio engineer to do this. Trust your ears, if the vocal feels clear and the music feels like it’s supporting rather than competing, you’re done.

Export as a high-quality MP3 or WAV and you have a finished song.

The Honest Reality

AI won’t write a great song without you. What it will do is get you past every point where you would have given up, the blank verse, the chorus that won’t come, the lyric that’s almost right but not quite. It handles the friction. You handle the taste.

That combination, done properly, produces something real. And in 2026, the tools to do it are free, fast, and genuinely good.