The 87 Words We Banned (And Why Your Lyrics Are Better Without Them)
SongForgeAI scans every generated lyric for 87 specific words and phrases. They are not offensive. They are worse — they are boring.
Somewhere in the SongForgeAI codebase there is a file called banned-terms.ts. It contains 87 words and phrases that get flagged and cleaned from every generated lyric. None of them are profanity. All of them are worse: they are the words AI reaches for when it has nothing specific to say.
What is on the list
We are not going to print the full list — partly because we update it, partly because knowing the exact words would let people game the system. But the categories are instructive.
Metaphor crutches: Words that gesture at depth without delivering it. The kind of word that makes a line sound poetic until you ask what it actually means.
Emotional shorthand: Words that name a feeling instead of showing it. If a line needs the word "heartbreak" to communicate heartbreak, the imagery has already failed.
AI comfort words: These are the words language models reach for most frequently when generating creative text. They appear in roughly 40% of unfiltered AI lyrics. Once you start noticing them, you cannot stop — they are the tell that separates AI-generated text from something a human would write.
Why banning works better than prompting
We tried prompt-level instructions first: "avoid cliches," "do not use overused metaphors," "write with specificity." It helped. But the banned list catches what prompts miss — the words that slip through because the model considers them natural, not cliched.
Post-generation scanning is the safety net. The writing room is steered toward originality by the prompt. The banned-terms scan catches whatever slipped through. And the Imagery Originality metric in the scoring system penalizes the patterns the scanner might miss.
What replaces the banned words
This is the important part. Removing a word is only useful if something better takes its place. When the scanner flags a line, the revision pass does not just delete the offending word — it rewrites the line to deliver the same emotional payload through specific, concrete imagery.
A line that relied on a banned emotional shorthand gets rebuilt around a visible detail. A line that used a metaphor crutch gets grounded in a real object or action. The result is a lyric that communicates the same feeling through better craft — and that scores higher on every metric that matters.
Listen to any song on the Examples page and try to find a line that sounds like it could have been written by any AI chatbot. If you find one, the system missed it. Forge your own and see how the pipeline handles specificity from the start.