Methodology

How SongVerdict actually scores a song

No black box. This is the exact framework: the four lyric dimensions, how genre changes the weighting, and how the hit-probability number is derived.

1. The four lyric dimensions

Every song is scored 0–100 on four dimensions. The model is instructed to use the full range — most songs are not 80+. A weak song will score in the 30s–50s.

  • Commercial appeal — hook strength, repeatability, title presence, conformance to current genre conventions.
  • Emotional impact — specificity over abstraction. Concrete imagery and lived-in detail beat generic feeling-words.
  • Originality — fresh angle, imagery, and phrasing relative to genre tropes. Originality isn't weirdness — it's the absence of cliché where cliché was easy.
  • Structure & craft — section pacing, prosody, rhyme integrity, no filler verses.

2. Genre-aware weights

A country ballad isn't judged like a trap song. When you tag a genre on upload, SongVerdict shifts the weighting. Below are the actual relative weights used to compute the overall score and the hit-probability number.

GenreCommercialEmotionalOriginalityStructureProduction
Pop38%18%8%28%8%
Hip-Hop / Rap28%18%22%18%14%
R&B / Soul22%34%12%18%14%
Country22%34%10%26%8%
Rock / Alt22%24%20%20%14%
EDM / Dance30%10%14%22%24%
Indie / Folk16%30%26%18%10%
Latin30%22%12%22%14%
K-Pop34%16%10%30%10%

Weights are normalized — they sum to 1.0 per genre. The bolded number in each row is the dimension that genre weights heaviest.

3. What the AI is told to look for, by genre

Beyond weights, each genre carries explicit critique guidance. This is verbatim from the system prompt:

Pop

Score for pop: hook is king. Punish forgettable hooks hard. Reward earworm payoff, repeated phrases that demand a singalong, and clean structure. Lyrical depth matters less than memorability.

Hip-Hop / Rap

Score for hip-hop: punchlines, flow, distinctive POV, and bar-by-bar craft matter most. Reward originality and specific imagery. Punish generic flexes and lazy rhymes (cat/hat, pain/rain). Hook still matters but verses carry weight.

R&B / Soul

Score for R&B/Soul: emotional honesty and vocal-first writing carry more weight than commercial polish. Reward intimate detail, melodic phrasing, and slow-build payoff. Punish over-stated emotion without lived-in detail.

Country

Score for country: storytelling is the spine. Reward concrete narrative detail (place names, objects, specific scenes). Punish vague emotion or list-songs without a story. Hook should land like a movie title.

Rock / Alt

Score for rock/alt: attitude, originality, and a bold POV outweigh commercial polish. Reward strong imagery and conviction. Punish over-produced pop-by-numbers writing.

EDM / Dance

Score for EDM/dance: the drop is the chorus. Lyrics carry less weight; energy build, hook chant-ability, and a singable post-drop motif carry most. Punish wordy verses and lyrics that fight the drop.

Indie / Folk

Score for indie/folk: originality, voice, and emotional specificity matter most. Don't penalize lack of commercial appeal. Reward unusual angles, fresh imagery, and a clear authorial voice. Punish generic singer-songwriter clichés.

Latin

Score for Latin: rhythm and melodic hook drive replay. Reward danceable phrasing and a chant-able chorus. Hook payoff matters more than verse complexity.

K-Pop

Score for K-Pop: structural perfection and section-by-section payoff matter most (pre-chorus → chorus → post-chorus). Reward earworm chorus and tight bridge. Punish flat dynamics or weak post-chorus.

4. The hit-probability number

Hit probability is a single 0–100 number that combines the four lyric dimensions (and audio dimensions if you uploaded audio), weighted by genre. The math:

hit_probability =
    commercial  * w_commercial
  + emotional   * w_emotional
  + originality * w_originality
  + structure   * w_structure
  + production  * w_production    // only if audio was uploaded

(weights from the genre table above, sum-normalized)

We label the result as Long shot (under 50), Worth a push (50–74), or Real shot (75+). These are calibrated for craft, not chart prediction. Nothing predicts a chart hit — that depends on artist, label, timing, and luck.

5. What this score is NOT

  • — It's not a prediction of chart performance. Hits depend on factors no model can see.
  • — It's not a substitute for taste. If your gut and the verdict disagree, your gut might be right.
  • — It's not trained on your work. Your lyrics and audio are never used to train any model.
  • — It's not absolute. The ranking (which song beats which in your batch) is more reliable than the absolute score.

Your songs stay yours

We don't sell, share, or train on your lyrics or audio. The only people who see your work are you and anyone you invite to a collab.