No black box. Here's exactly what the model looks for, what it ignores, and why we believe the rankings — not the absolute scores — are where the real value is.
Why we built it
Most songwriters end up showing demos to friends, partners, and bandmates — people who love them, not their songs. The feedback is kind, vague, and useless for actually deciding what to pitch, record, or scrap.
SongVerdict exists to be the unkind friend. It doesn't care about your feelings. It picks a winner.
The four dimensions
Every song gets a 0–100 in each of the four dimensions below. The overall score is a weighted average — but the ranking is what you should actually act on.
Hook strength, repeatability, melodic accessibility, title presence in the chorus, and conformance to the conventions of your stated genre. We benchmark against current radio/streaming patterns — not against songs from 1997.
What it looks for
Specificity over abstraction. Concrete imagery, sensory detail, and an identifiable point of view tend to outperform generic feeling-words. We reward lines that earn their emotion through detail rather than declaring it.
What it looks for
How fresh the angle, imagery, and phrasing feel relative to common tropes in the genre. Originality isn't weirdness — it's the absence of cliché where cliché was easy.
What it looks for
Section pacing, prosody (do the syllables fit a singable melody?), rhyme integrity, and whether the song earns its length. A great idea poorly structured loses to a decent idea well structured.
What it looks for
Audio is optional but powerful. When you include it, SongVerdict adds a second pass — the producer's listening review — covering vocal delivery, melody, arrangement, energy, and production polish.
The audio review doesn't change the lyrical scores. It runs alongside them so you can see whether your delivery is helping or hurting an already-strong song.
Honest limitations
Nobody can. Hit songs depend on artist, label, timing, and luck. We score craft, not commercial outcomes.
If your gut and the verdict disagree, your gut might be right. Treat the score as a strong second opinion.
Your lyrics and audio are never used to train any model. Songs stay private to your account unless you choose to publish.
We don't sell, share, or train on your lyrics or audio. The only people who see your work are you and anyone you explicitly invite to a collab.
Free tier: 2 songs per analysis. No credit card.