Do the words come first or the melody? How a song actually gets built

Here is a line out of a real first draft, written on paper, no tune anywhere in the room:

I have been thinking about the way you left the apartment that morning

Eighteen syllables. Try to sing it. You run out of breath somewhere around "apartment", and the melody has to sprint through the middle to arrive on time. Now the same line after it has been through a voice:

The morning you left, you did not look back

Ten syllables, four strong beats, and there is air left at the end of the phrase. Two things went: the apartment, and the clause that was only clearing its throat. Neither was carrying anything the listener needed.

That gap is the entire argument about what comes first. It is not a matter of temperament and it is not a question you get to answer with "whatever works for you". Each order protects one thing and quietly wrecks another, and the wreckage is predictable enough that you can plan around it.

The short answer, defended below: do not start from a finished lyric and do not start from a finished melody. Start from one line of the chorus that already has words and a rhythm at the same time, and grow the song outward from it. And if a machine is going to sing what you write, the choice is taken out of your hands anyway.

Three orders people actually work in

Words first

You write the lyric like a piece of writing. Verses, a chorus, maybe a bridge, all on the page before a single note exists. Then you sit down with the guitar or the keyboard and try to sing it.

What it protects: meaning. A lyric written this way says something. It has a subject, a turn, and usually one image nobody else would have found.

What it breaks: the body of the thing. Written language has no breath in it. On paper you can run a clause for twenty-two syllables and the reader just reads faster. A singer cannot read faster. So the melody ends up serving as a delivery truck for text, cramming syllables into bars, and every line arrives a little out of shape.

Music first

You find a melody by humming, by hitting chords until something repeats, by looping four bars until a shape appears. Words come later, poured into a mould that already exists.

What it protects: the rhythm, and it protects it completely. A lyric written into an existing melody scans by construction. It cannot not scan.

What it breaks: what the song is about. The melody hands you a mould with specific slots, and the fastest way to fill a slot is with a word that fits the vowel rather than a word that is true. Do that for three minutes and you end up with a song made of "stay", "away", "afraid", "the same", "escape". Every one of those opens the mouth on the same stressed vowel, which is why they turn up together. The song sounds finished and says nothing.

The hook arrived on its own

Third order, the one nobody plans. A line shows up whole in the car or the shower, with its rhythm already attached. You did not write it and you cannot remember deciding anything about it.

What it protects: the one thing that is hardest to manufacture, which is a line a stranger can repeat after one listen.

What it breaks: two things. The first is that everything around a gift line becomes scaffolding. You are no longer writing a song, you are building a delivery system for one line, and the verses come out as polite explanation. The second is worse. A line that arrived complete feels untouchable, so you keep writing songs around lines that do not deserve a song.

What this site does, before you plan the order around it

Worth saying plainly now rather than in a footnote at the bottom, because it changes the answer to the question in the title.

The lyrics generator on this site writes words. A title and a lyric, up to 20,000 characters, for 20 credits, which is 20 cents at 100 credits to the dollar. You can see the rest of the numbers on the pricing page.

It does not make sound. No vocal, no instruments, no mp3, nothing you can play. To hear the words you take them somewhere that generates audio, and the service most people use for that is Suno. Suno is a separate company. We do not own it, we do not resell it, and we do not control what it does with your text.

Which is exactly why the order matters here. Once the words and the audio live in two different places, the order stops being an artistic preference and becomes a fact of the plumbing: text is made first, audio second, always. The rest of this article treats that as the situation rather than an afterthought.

Why lyrics-first comes out unsingable, and the repair

The failure is not that the lines are too long. It is that written lines have the wrong shape: they carry their meaning in connective tissue rather than in stressed words.

Look at the first draft again. "I have been thinking about the way" spends nine syllables on a frame, and only one word inside them carries information. In speech that is fine, because speech has no bar lines. In a song, those nine syllables all have to land somewhere, and they will land on strong beats that then have nothing strong to say.

The repair, in order:

1. Read the whole lyric out loud at speaking pace, then again at half speed. Half speed is roughly where a slow song sits. Any line you cannot deliver in one breath at half speed is not a long line, it is two lines pretending.
2. Write the syllable count in the margin of every line. Not to make them identical. Lines inside a section should be in a family, within about two of each other. A verse that goes 9, 10, 9, 17 tells you exactly which line to operate on.
3. Mark where the stresses actually fall when you speak the line. Then ask whether the words you care about are sitting on those stresses. In "The morning you left, you did not look back", the stresses land on MORning, LEFT, NOT and BACK. All four carry information. That is not luck, it is what survived the edit.
4. Cut hinges before you cut images. "I have been thinking about the way that" is pure hinge. "the apartment that morning" is two images competing for one slot, so one of them goes.
5. Delete the setup line. Almost every lyric written on paper has a line that explains the situation directly before the line that shows it. The explaining one is always the weaker one, and the song survives its removal every time.

There is a real counter to all of this, and it is worth stating rather than hiding. Some people genuinely write finished text first and it comes out singable. Pay attention to what those people are doing while they write: they are singing under their breath the whole time. The lyric only looks like it came first. In practice both arrived together, which is the order this article is arguing for.

If the line-level craft is what you actually need, the mechanics are worked through separately in the lyric-writing guide and this article will not repeat them.

Why melody-first turns into mumble

Melody-first fails in a way that is hard to see from inside, because the result sounds correct. The rhythm is right, the phrasing is right, nothing scans badly. The song is simply empty.

Here is the mechanism. You hum a melody, and to hold it in your head you fill it with placeholder syllables: nonsense, "na na", half words. Those placeholders are not neutral. They have vowels, and the vowels are chosen by your mouth for comfort, not by you for meaning. Then you write real words and, without ever deciding to, you choose words that preserve the vowel you have been humming for an hour. That is how a song ends up with four separate lines about being afraid to stay away from the same escape.

There is a second, uglier version: the placeholder that never leaves. A two-syllable open-vowel filler like "baby" survives into the final lyric because it fits perfectly, and it fits perfectly because you invented the melody around it. Now the hook of your song is a word that carries no information at all.

The fix is to separate the shape of the melody from the sound of it:

1. Record the hum before you write anything. Two reasons: the shape stops drifting, and it stops quietly bending to accommodate a weak line you have become attached to.
2. Transcribe the rhythm, not the syllables. Write it out as "da DUM da da DUM da", or as a row of syllable counts per phrase. What you want on paper is the pattern of long and short and stressed and unstressed. Nothing about vowels.
3. Write words to that pattern with the recording off. The vowel is the trap. With the audio silent, you are matching a rhythm rather than imitating a sound, and the rhyme family stops choosing your vocabulary.
4. Sing the new words against the recording once, at the end. If a word genuinely does not sit, change the melody. The melody is not sacred just because it came first.

The compromise that actually works: one line of the chorus

This is the position, stated flat: neither pure order is good, and the working method is to write one line that has both properties at once, then let everything else grow from it.

One line. Not a chorus, not a verse, not a concept. One line that could be repeated by somebody who heard it once.

You still have a key to a door I changed

Ten syllables. It contains a fact rather than a feeling, which is the test that matters: "I am not over you" is a feeling and can be written by anyone about anyone, while a key and a changed lock belong to one specific situation and imply the feeling without announcing it.

Once that line exists, it starts issuing orders, and the orders are useful:

• The syllable count becomes law for the section. Ten syllables for the hook means the other chorus lines want to be around ten. Not exactly ten. Around ten.
• The verse acquires a job. The verse is not more of the song, it is the evidence for the chorus. If the chorus is a verdict about a changed lock, the verse is the hallway, the coat that is still there, the day the locksmith came. Small and physical. No speeches.
• The second chorus gets its variation for free. Keep the syllable count, turn the end of the line: "You still have a key. I still have the door." Ten syllables again, and the relationship in the song has moved without a single new idea being introduced.
• The genre gets chosen last, and by the line. That sentence will not survive a stadium chorus; it is too quiet and too specific. It wants a small arrangement. Letting the line pick the genre instead of the other way around saves you from writing a country lyric over a house track.

If the hook itself is the part that will not come, that problem has its own article, and if you are trying to work out where the sections go in the first place, the structure templates cover that ground. This article is about sequence, not architecture.

When a machine makes the music, the order is not yours to choose

Here is where the question stops being philosophical. If the audio is going to be generated from a finished text, there is no order to decide. Words go first. You cannot hum at a text field.

That single constraint changes how the words have to be written, in five specific ways.

The first thing you write is not a line, it is a brief

The generator has exactly one free field, theme, and it takes 300 characters. That is it. Everything else is closed lists you pick from: point of view, song language, audience age, structure, hook direction, vocal, musical style, and one to three emotions.

Three hundred characters is roughly one paragraph. So before a single lyric line exists, you have to compress the entire situation into a text message. Most people waste it on adjectives. A better use looks like this:

She left in January. Her coat is still in the hall and I changed the lock the week after. I want the chorus to land on this line: you still have a key to a door I changed. Verse should be small and physical, no speeches, just the hallway and the coat.

Notice what that does. It spends its budget on concrete facts, one instruction about register, and the hook line itself. The one-line method from the section above survives intact; it just moves into the brief. You can try that in the generator and see how differently a brief like that behaves compared to "a sad song about a breakup".

Stress is the thing that will betray you

A human singer reads your line, understands it, and puts the emphasis where the meaning is. A generator does not have that luxury in the same way. It works from the text, and the safest assumption is that your lines will be sung with whatever stress the words naturally carry when spoken.

Practical consequence: write lines whose spoken stress already falls where you want the beat. If the important word in a line sits on a weak syllable, it will not be rescued in performance. Rewrite the line so the word lands on its own.

Line length has to be a family, not a range

This one is mechanical and easy to check before you submit anything. Take the verse and count the lines. If they run 8, 9, 8, 10, the result has a chance of sounding composed. If they run 8, 9, 8, 19, the nineteen has to be delivered somehow, and the usual outcome is that it gets crammed and the phrasing goes flat right where your best line was.

The fix is not to pad the short lines. It is to split the long one.

Repeats are written, not felt

Humming a hook teaches your body that it repeats. A text field knows nothing about that. If the hook should come back four times, write it four times. Do not write it once and assume anything will happen. The same applies to a tag at the end of a chorus, an echoed last word, a line that returns half-length in the outro: if it is not in the text, it does not exist.

Things you cannot ask for, so do not write them

The vocal field offers three choices, and a duet is not among them. Which means a lyric built as an exchange between two named voices, verse one from her and verse two from him, comes out of the audio service sung by one voice doing both parts. It reads as confusion, not as a conversation.

Check the closed lists before you write, not after. The structure list has six options, the style list has 53 curated roots, the emotion picker takes one to three of twelve. Writing a lyric that needs something outside those lists is the most common way to waste 20 cents and twenty minutes. What happens after the text exists, the handoff into an audio service, is covered in its own guide.

Where this advice does not apply

Five situations where the order is decided by circumstance and arguing about it is pointless.

Rap over an existing beat. The beat is first and it is not negotiable. Flow is a response to drums, so the words are written against a rhythm that already exists in the room. This is melody-first with the melody removed, and it does not suffer the mumble problem because rap carries meaning in consonants and rhythm rather than in sustained vowels.

Topline work. Somebody hands you a finished instrumental and asks for a melody and words over it. Music is first by contract. Your only freedom is inside the bars you were given.

New words to an existing tune. Parody, a hymn text, a birthday version of something everyone knows. The melody is fixed down to the syllable, so you are not writing, you are fitting, and the syllable count stops being a guideline and becomes an exact requirement.

Two people in a room. Co-writing has no order. Whoever opens their mouth first sets it, and the useful skill is noticing within ten minutes which of you is driving so the other one can stop fighting the direction.

Writing in a language you do not speak fluently. Here the order should flip harder toward melody, because your instinct for which syllables are heavy is unreliable. Establish the rhythm first and fill it carefully, and check your stresses against someone who speaks it. There is a separate piece on that.

The order, compressed

If you want it as a sequence rather than an argument:

1. Get one repeatable line with a fact in it. Say it out loud until a rhythm settles on it.
2. Count its syllables. That count is the law for the section it belongs to.
3. Build the chorus around it, then write the verse as evidence for it.
4. Pick the genre and the arrangement last, and let the line choose them.
5. If a machine is doing the audio, compress steps 1 to 3 into a 300 character brief, check your line lengths, write the repeats out in full, and only then spend the 20 cents.

And if none of this is the problem yet, if the actual issue is that you have never finished a song at all, the order question is premature. Start here instead, finish something bad, and come back when you have a draft that will not sing.