Ask ChatGPT for a lyric "in the style of" your favourite artist and you tend to get the same thing every time: moonlit roads, broken hearts, fire and rain, and a chorus that says "I feel so alone" in so many words. A style label is too vague to steer anything, so the model falls back on its most familiar phrases. This guide swaps the label for a blueprint. You ask ChatGPT to take a reference song apart (its shape, its emotional journey, the kind of images it uses), then you write something entirely your own on that frame, and finally you make sure the words and the Suno style are pulling in the same direction.
One ground rule first. You never paste the reference song's lyrics into the chat. The guides we read (listed at the end) agree that doing so invites refusals, bland rewrites and copyright trouble. You don't need the words anyway. Everything worth borrowing is in the architecture, and architecture isn't something anyone owns.
Why "In the Style Of" Falls Flat
A style label asks the model to guess what you admire about an artist. ChatGPT can't know, so it averages, and the average of every song is a cliché. Naming the artist doesn't help much on the Suno side either. Blake Crosley's August 2026 guide to the Style field lists artist references as unreliable and possibly filtered, and recommends describing the sound with genre, tempo, instruments, vocal and mood instead. The fix in both places is the same: replace the name with a description of what you actually want.
Part One: Build the Blueprint
Step 1 — Pick a reference song you know well. Choose one whose effect you can describe: it builds slowly, it hides the main point until the bridge, it tells a story through objects in a room. If you can't say why it works, the analysis won't mean much to you.
Step 2 — Ask for an analysis, not a rewrite. Give ChatGPT the title and artist only and ask it to work from what it knows about the song. Something like:
Act as a songwriting teacher. Without quoting any lyrics, analyse "[song]" by [artist]. Describe: (1) the section order and roughly how many lines each section has; (2) the job each section does in the story; (3) the emotional arc from first line to last; (4) the types of imagery used (objects, places, weather, actions) without quoting them; (5) the rhyme approach (perfect, near rhymes, internal rhymes, how often); (6) point of view and who is being addressed.
Step 3 — Check it against your own ears. Language models can misremember songs, especially less famous ones. Listen once with the analysis beside you and correct anything wrong, such as a missing pre-chorus or a "hopeful ending" that is actually bitter.
Step 4 — Abstract it into a reusable blueprint. Now ask ChatGPT to strip out everything specific to the original song and keep only the pattern:
Turn that analysis into a blank blueprint I can reuse for a different subject. Remove the original song's topic, names and images. Keep: section order, line counts, the job of each section, the emotional arc as a sequence of steps, imagery categories, rhyme approach and point of view. Format it as a short template.
What comes back should read like a recipe, for example: Verse 1 (4 lines): ordinary scene, one object hints something is wrong. Verse 2 (4 lines): same place, later, object has changed. Chorus (4 lines, repeated exactly): one plain admission, near rhymes. Bridge (2 lines): shift of time or viewpoint. Arc: denial → noticing → admitting → letting go. Save it: a good blueprint can carry many different songs.
Part Two: Write Your Own Song on the Frame
Step 5 — Bring your own subject. The blueprint is the frame. The material has to be yours: a real place, a real habit, something that happened. Jack Righteous's 2026 lyric-writing guide suggests pinning down the speaker, who they're talking to, what happened and what the speaker won't admit before any lines get written. That last one is gold for a chorus.
Step 6 — Get ChatGPT's clichés on the table first. The aitooltips guide has a neat trick: ask the model which clichés it tends to use in lyrics, then paste that list back as a "do not use" instruction. Add your own pet hates too:
Before we write: list the 20 phrases and images you most often use in song lyrics. I'll ban them. Also avoid: "heart on fire", "lost in the night", "set me free", "falling apart", "shattered", rhyming "fire" with "desire".
Step 7 — Tell a small story instead of naming feelings. Rather than asking for "a sad verse about a break-up", ask for a scene. Feelings land harder when the listener works them out from evidence. Jack Righteous puts it as emotion coming through consequence, what changed because of what happened:
Using the blueprint, draft Verse 1 about [subject]. Don't name any emotion. Show it through one object, one place and one action. Four lines, near rhymes only.
"Two mugs on the draining board, I only washed the one" says more than "I miss you so much", and it's yours in a way no style label could produce.
Step 8 — Generate options, then choose. Ask for three versions of each section, each built around a different central image, and lift the strongest lines into your own draft. Robin Frederick at My Song Coach makes the point that AI output works best as a springboard, and that line lengths straight out of a chatbot are often too uniform. Vary them: a tighter verse, a longer chorus line, a short punch at the end.
Step 9 — Rewrite by hand. Read the draft aloud and change any line that could be in anyone's song. This is where it becomes yours, and it matters beyond taste: copyright guidance in places such as the US currently looks for meaningful human authorship, so your own selection, rewriting and new lines are worth keeping a record of.
Part Three: Make the Words and the Sound Agree
Step 10 — Name the lyric's mood in one or two words. Decide what your finished lyric actually feels like (wistful, defiant, relieved) and whether it moves, for instance "starts numb, ends hopeful".
Step 11 — Build the Suno Style prompt from that mood. Describe the sound without artist names, using genre, tempo, instruments, vocal and mood. Blake Crosley suggests around four to seven descriptors; fewer leaves gaps that Suno fills with defaults, and more starts to contradict itself. For the mug lyric above, something like:
Indie folk, slow tempo, fingerpicked acoustic guitar, soft brushed drums, intimate male vocal, wistful, sparse
Step 12 — Watch for accidental mismatches. A grief lyric over upbeat dance pop, euphoric, festival energy will usually sound confused rather than clever. Soundverse's prompt guide warns that conflicting tones in one prompt tend to produce inconsistent melodies, and suggests describing a mood shift as an arc instead, such as "builds from loneliness to hope". If you do want the bittersweet sad-words-happy-beat effect, make it a deliberate choice and say so plainly in the style, for example bright synth-pop, melancholic vocal, rather than letting it happen by accident.
Step 13 — Generate, listen, adjust one thing. If the music fights the words, change the style prompt first and leave the lyric alone. If the music is right but a line sounds wrong, change that line only. One change per attempt is the only way to learn what fixed it.
Common Mistakes
Pasting the reference lyrics: you don't need them, and it causes more problems than it solves.
Trusting the analysis blindly: check it against the actual recording.
Keeping the original's images: the blueprint should hold categories ("an object that changes"), not the original song's objects.
Naming every feeling: show the evidence and let the listener feel it.
Artist names in the Style box: describe the sound instead.
Mood by accident: if the words and the music disagree, make sure that was the plan.
A Quick Checklist
1. Reference chosen, no lyrics pasted2. Analysis checked by ear and corrected3. Blueprint saved with no original images4. Own subject, speaker and secret defined5. Cliché ban list in the prompt6. Verses show scenes, not named feelings7. Draft rewritten by hand, line lengths varied8. Lyric mood named; Style prompt built from it9. No artist names; any mood contrast is deliberate
Where That Leaves Us
The songs we love are worth studying, not copying. Take the shape, the arc and the way they choose images, bring your own story, and make sure the sound you ask Suno for tells the same story as your words. That's how a human songwriter and an AI collaborator end up with something that sounds like nobody else. If you'd like to hear how that works for us, our songs are on the homepage, and our Suno Guide has the prompts, style tags and how-to guides we keep coming back to.
Sources
A.I. Tooltips — How to use ChatGPT to write lyrics (analysing a reference song without pasting lyrics, why "in the style of" is too vague, the cliché ban-list prompt; December 2025)
Jack Righteous — How to write song lyrics with AI: a practical 2026 guide (speaker, listener and what they won't admit; each section has a job; emotion through consequence; several options around different images; updated August 2026)
Blake Crosley — Suno's Style field and Style Influence slider, explained (artist references unreliable, four to seven descriptors; August 2026)
Soundverse — How to write effective Suno AI prompts (conflicting tones, describing mood as an arc; January 2026)
Robin Frederick, My Song Coach — Can you use AI for writing lyrics? (AI drafts as springboards, uniform line lengths, human contribution and copyright)
