AI songwriting

Where to Use AI in Making a Song: A Seven-Stage Map

 By the GLOBAL MUSIC PARTNERSHIP editorial team / Supervised by Naoto Tanaka (Tarnar sensei)

Decide stage by stage what you ask AI to do, what you hand over and what you keep. A seven-stage map makes a joint song with Japan calm and fun.

The best way to use AI in a song is to decide, stage by stage, what you will ask it to do, what you will hand over to it and what you will keep for yourself. A song passes through seven stages: lyrics, composition, arrangement, arranging the parts, recording, mixing and mastering, and each one can be treated differently. A short map of these decisions, written before the first session, makes a joint project with a partner in Japan calmer, faster and more fun.

Written by Tarnar-sensei (Naoto Tanaka): voice trainer for 30 years, has coached more than 50,000 people over 30 years. He wrote and composed the ending theme of the Japanese television programme Kaiun! Nandemo Kantei-dan in 2021 and 2025, and he has produced music for 119 projects or more.

Why does a stage map beat a single yes or no on AI?

Teams often begin with one big question: are we using AI, or not? That question is hard to answer because the answer differs at each step. A team may love AI-generated chord ideas and still want every word of the lyric to come from a human heart. Another team may write the melody by hand and enjoy letting AI propose ten arrangements to compare.

Splitting the question into stages gives you seven small, easy decisions instead of one large argument. It also gives your partner a clear document to react to, which saves time in the first online meeting.

One catalogue can hold both choices happily. The ending theme of Kaiun! Nandemo Kantei-dan was written and composed, in 2021 and again in 2025, without AI. The insert song for the 123 Musical, a community musical in Izusan, was made with AI from the lyrics through to the finished track. The producer is the same, the maps are different, and each was chosen to fit its project.

The approach here follows the way the Japan-side group behind this partnership window teaches it. Its online songwriting school, the AI Lyric and Composition School, trains students to treat each production step as a choice about how to ask AI, what to leave to it and where to decide personally. The school also teaches ways of making songs without AI, so the map works in whichever direction your team prefers.

What are the seven stages, and what question belongs to each?

Use the table as a worksheet. For every stage, mark one of three choices: ask AI for options, hand the stage to AI, or keep it with a person. Add a note on who holds the final decision, on your side or your partner's.

StageThe question to askTypical choice to mark
LyricsDo the words need to sound like a particular person?Ask AI for drafts, then rewrite in your own words
CompositionIs there a melody someone already hums?Keep a human melody, or ask AI for several seeds
ArrangementDo you know the sound you want?Ask AI for variations and compare
Arranging the partsWhich instruments carry the hook?Decide together after listening
RecordingWhose voice should the listener hear?Usually a human singer
MixingDo you have a reference mix to match?Try AI first, then adjust by ear
MasteringWhere will the song be heard?Choose by listening on the real playback device

Treat the third column as a starting suggestion that your team can overwrite.

How do you use AI for lyrics and still make the words yours?

Lyrics are the stage where the map matters most, because words carry identity. The school teaches two skills here. The first is how to ask: what to tell AI about the theme, the listener, the language and the feeling, so the draft arrives close to what you imagined. The second is how to treat what comes back. The words that appear are not used as they are. They are reworked into the writer's own language, line by line. The article on rewriting AI lyric drafts in your own words goes deeper into this skill.

A practical routine for a joint session looks like this.

  1. Write the theme in one sentence and name the listener in one phrase.
  2. Ask AI for three short drafts of the chorus only, not the whole song.
  3. Read the drafts aloud and underline two or three phrases that feel true.
  4. Rewrite the chorus using those phrases and your own images.
  5. Show the rewrite to your partner and explain, in one sentence, why you chose each line.

If your song will carry both Japanese and English, add a step: ask your partner which phrases translate well and which lose their colour. Whatever you decide about the lyric language, write it down in the stage map so both teams see the same answer.

A short lyric brief for the AI, kept beside the draft, makes the first step easier and shows your partner what you asked for.

What to tell the AIExample of how to say it
ThemeA farewell to a summer we spent working side by side
ListenerStudents and parents in the audience on the last evening
LanguageSimple English with one short Japanese line in the chorus
FeelingWarm, hopeful and easy to sing along with
LengthFour lines for the chorus, written to be sung slowly

Keep every brief you write. After a few songs you will see which wording led to drafts close to your intention, and the next brief becomes quicker to write. For a choice of language, see Japanese, English or both.

How do you choose between composition and arrangement options?

In the school's lessons, composition and arrangement follow a clear rhythm. The AI produces options, the teacher and the student listen to them together, and the student puts into words why one option was chosen over another. That last step is the heart of the method. A reason such as "the second version leaves room for the vocal" teaches more than a vague feeling, and it lets your partner understand your taste.

You can borrow the same habit in a team meeting.

  • Play two or three options back to back, with no names attached.
  • Give each person one minute to write a single sentence about each option.
  • Read the sentences out, then choose together.
  • Record the winning reason in your decision log.

People who cannot read a score are welcome in this process. The school states that students can start without reading music and without any experience of singing or playing, so the listening method is open to everyone on your team. The decision log guide shows a simple layout for the record.

What decides recording, mixing and mastering?

Recording is usually the stage where a human voice enters, and many teams keep it that way because the singer is the face of the song. If your project involves vocals from both countries, plan the recording as a shared event, and use the decision log to note who sings which line.

Mixing and mastering are technical stages, so AI tools can be useful for a first pass. The final test is your ears on the device your listeners will use. If the song will be heard on phones at a festival, check it on phones. If it will be played in a hall, check it in the hall. Note the result next to the stage so the next song in your catalogue starts from a better place.

How do you fill in your own stage map?

Copy this template into a shared document and complete it before your first online meeting. It takes about as long as a cup of tea, and it gives both teams a common picture.

StageAsk AI / Hand to AI / Keep humanWho decidesReason in one sentence
Lyrics
Composition
Arrangement
Arrangement of parts
Recording
Mixing
Mastering

When both sides have filled it in, compare the sheets. Differences are welcome, because they show exactly what to talk about. Where the sheets agree, you can move quickly. Where they differ, a short conversation and a small trial of one chorus usually settle the matter.

Once the map is chosen, keep it with the song. A one-line note per stage, saying whether AI was used and by whom, is a kind gift to your future self and to anyone who later reuses the track. The same idea appears in A Making-of Note. It also fits the practice of recording who holds which rights in each work, which the partnership window asks both sides to do before a pilot begins.

How do you take the first step with a partner in Japan?

The Japan-side school, AI Lyric and Composition School (song.atono.jp), offers online individual lessons of 25 minutes, four times a month, at 22,000 yen including tax, with a single trial lesson available. The lyrics are guided by TARNAR and the composition by another teacher.

If you would like to try the map on a real song, the partnership window can help you shape a small pilot around one chorus or one track, in English or Japanese. Send a short message with your company, country, time zone, goal and preferred language, and mention that you would like to start from a stage map. You can reach the window from the top page: Global Music Partnership Japan.

Related reading: Co-production with Japanese artists: 5 things to agree on first.

FAQ

Where should I use AI when making a song?

Decide per stage. Lyrics, composition and arrangement suit AI options that you then choose from and rewrite, recording usually stays with a human singer, and mixing and mastering can start with AI and finish by ear.

How many stages does a song production map have?

Seven: lyrics, composition, arrangement, arranging the parts, recording, mixing and mastering. For each, mark whether you ask AI for options, hand the stage to AI or keep it with a person.

How do you keep AI-written lyrics from sounding generic?

Brief the AI on theme, listener, language, feeling and length, ask only for chorus drafts, then rewrite the drafts into your own words and explain each line choice to your partner.

Can people who cannot read music take part?

Yes. The Japan-side AI Lyric and Composition School says students can start without reading music and without experience of singing or playing, and the listening method in this guide suits the same people.

Does the Japan-side school also teach songwriting without AI?

Yes. The school teaches ways of making songs without AI as well, so the stage map works whichever direction your team prefers.