Skip to content
All posts
Songwriting2026-04-247 min readBy Todd Nigro

What "specificity" actually means in a lyric

The word every AI evaluator over-uses and under-defines. A working writer's version: specificity is not detail. It is the detail only this narrator, in this song, could have given you.

If you read enough AI feedback on song lyrics, you learn that "be more specific" is the advice the model gives when it doesn't know what else to say.

A line scores a 62. The model says: make it more specific. Another line scores an 88. Same advice. Someone asks why the drafts feel the same shape as each other, and the answer comes back, again: specificity.

The word has been ground into dust by repetition. But the thing it originally pointed at is real, and it is the single most important concept in lyric writing that almost nobody gets right on the first pass — including, until recently, our own evaluator.

This post is what we mean by specificity when we score lyrics on the Lyric Scoring Standard. It is also, I think, what songwriters should mean when they use the word.

Specificity is not detail

Most people treat "specific" as a synonym for "detailed." It isn't.

A lyric can be extraordinarily detailed and completely generic. "I sat alone in a dimly lit bar on a rainy Tuesday evening and sipped a glass of red wine" is detailed. It is not specific. Nothing in that sentence could have come only from this particular narrator. Anyone who has ever written a country song has written that sentence. The details are fine. They are also furniture.

Specificity is the detail only this person, in this song, could have given you.

It is not the bar. It is the name of the bar, because the narrator has been there enough to know it by name. It is not the rain. It is the rain hitting the metal awning he always parks under so he can keep the windows down, because that's the detail that places him inside a life, not inside a setting. It is not the red wine. It is the red wine because she used to order it and he hates it and he's ordering it anyway, because that is what grief does.

That's the difference. Detail is noun + adjective. Specificity is detail that carries a reason.

The Stolpe test

Pat Pattison teaches something he attributes to Joni Mitchell and refines with students at Berklee: a lyric is strong when the external world the narrator describes and the internal world the narrator reveals match each other so tightly that you cannot separate them. Verne Stolpe, who taught him, put it a different way: verses lean external, choruses lean internal, and each one carries the weight of the other.

The practical test is this. Read a line. Ask: does this line give me information about the world, information about the narrator, or both? If it's only one, it is a detail. If it's both, it is specificity. "Dimly lit bar" is only about the world. "The bar where I always took you on a Tuesday" is about the world and the narrator in the same breath.

The Lyric Scoring Standard has a metric called Sensory Specificity. When the evaluator is scoring it, this is the test: can you point at a line that does both jobs at once? If yes, the score goes up. If every detail is world-only — if the lyric reads like a set description rather than a person's interior — the score stays near 50, even if the language is pretty.

"I miss you" is where specificity begins, not ends

A common workshop note: don't say "I miss you." Describe the missing.

That's good advice, but it is only halfway there. Telling a writer not to say "I miss you" produces a lot of lyrics about rainy windows and unfinished coffee. Better. But still furniture. The full advice is: describe the missing in a way that could only belong to this narrator.

Take a concrete example. A narrator says "I miss you." The first revision: "I still can't drink my coffee black." Better — it's showing a specific thing. The second revision: "I still can't drink my coffee black because you always laughed when I tried to." Better again — the missing now has a reason and a scene. The third revision: "I still can't drink my coffee black, and she can't either, and we don't talk about you." Now the missing has a scene, a reason, and a whole world of relationship around it that no other narrator could have given you.

Every revision above sounds "specific" in the workshop-note sense. Only the last one sounds like it could only be this song.

What the evaluator looks for

When the Lyric Scoring Standard scores Sensory Specificity and Image System Coherence (metrics 5 and 6), it runs two questions in sequence:

  1. Can I point at any concrete image in this lyric? If no, the score stays below 55. If yes, proceed.
  2. Does each concrete image carry information about the narrator as well as the world? If no — if the images are decorations — the score caps around 65. If yes, the score can climb.

The second question is what makes the ceiling real. Lots of AI lyrics clear the first bar. Almost none of them clear the second on a first draft. The draft has streetlights and cigarettes and whiskey and empty beds. None of those things, as written, belong to any particular narrator. They are props. The evaluator sees the props, sees the absence of any reason the narrator would own these props, and holds the score.

When specificity works, the score climbs into the 80s, and you know it worked because you couldn't cut any image from the lyric without also cutting a piece of the narrator's character. That inseparability is the whole point.

Why this concept keeps losing

Specificity is difficult to teach and difficult to coach for the same reason: it sounds like the smallest edit possible. One swap: "streetlight" → "the streetlight on the corner where we used to meet." Looks cosmetic. Reads transformed.

Language models trained on lyric data have absorbed millions of generic-image lyrics. The prior is so strong that their first instinct on almost any prompt is to generate that kind of language. You can watch it happen in real time: ask any model for a verse about heartbreak and watch the "empty rooms, cold sheets, slow rain" vocabulary surface within two lines. That vocabulary is not bad. It is just unattached. It does not yet belong to a person.

The gauntlet pass in our forge is, in large part, a specificity intervention. When the eval panel returns a lyric with weak Sensory Specificity, the gauntlet prompt instructs the revision room to replace every unattached image with an image that carries a reason. The lift from that intervention, on average, is five to seven points on the composite. Sometimes more. It is the single highest-ROI revision we ever ask the room to do.

The rule, in one line

A lyric achieves specificity when each concrete image is so entangled with the narrator that removing it would change who the narrator is.

That is the bar. That is what the word means when we use it. And that is what a lyric has to clear if it wants the number on the score chip to mean anything at all.

When you are next revising a draft — whether you wrote it yourself or generated it — try this. Read every concrete noun. Ask: could any other narrator have given me this? If yes, you haven't reached specificity yet. Keep going. The line is not finished until the image is no longer transferable.

That is a high bar. It is also the bar that separates lyrics you forget tomorrow from lyrics someone else will still be quoting in ten years.