Verified against ChatGPT · 2026-08-12
Rewrite a dialogue pass so two characters who currently sound identical develop distinct voices
Diagnoses why two characters' dialogue currently reads interchangeably and rewrites their lines using distinct sentence rhythm, vocabulary habits, and conversational tells, without resorting to a surface gimmick like an accent or catchphrase.
The prompt
Ready to copy — highlighted parts are example details you can swap.
I have dialogue where two characters sound too similar — if you covered the names, you couldn't tell who's speaking. Diagnose why, then rewrite it so their voices are genuinely distinct. CURRENT DIALOGUE DETECTIVE: I think the witness is lying. NEIGHBOR: I think so too, honestly. DETECTIVE: We should talk to her again. NEIGHBOR: I agree, we probably should. CHARACTER A BACKGROUND A veteran detective, worked homicide for 20 years, trained herself out of ever asking a question she doesn't already suspect the answer to. CHARACTER B BACKGROUND A retired schoolteacher neighbor, used to filling silences for nervous students, genuinely uncertain and thinking out loud as she talks. WHAT SHOULD STAY THE SAME Both characters must still agree the witness should be questioned again, and the neighbor must still be the one to suggest it happen tomorrow morning. RULES First, diagnose the current sameness in concrete terms: identify at least two specific patterns both characters currently share (for example: both speak in complete grammatical sentences of similar length, both hedge with the same filler words, both use the same register of vocabulary) rather than giving a vague note like "they need more personality." Then assign each character a distinct set of speech habits derived from their background, not from a surface gimmick — differentiate through sentence length and completeness (does one interrupt themselves, trail off, or speak in fragments while the other finishes every thought), through what each one asks versus asserts (does one respond to questions with more questions), through vocabulary register tied to their actual background rather than decoration, and through what each character avoids saying directly. Do not introduce an accent, a verbal tic, or a catchphrase as the primary differentiator — those are surface fixes that read as gimmicks rather than voice, and they don't hold up over a full script the way rhythm and habits of thought do. Rewrite the given dialogue preserving the plot content and the information each line conveys, plus anything listed under Both characters must still agree the witness should be questioned again, and the neighbor must still be the one to suggest it happen tomorrow morning., while changing how each character says it. WHAT NOT TO DO Do not make character B simply the polite, correct version of character A's blunt lines — that's still one voice with a politeness dial turned differently, not two distinct voices. Do not change what information each character reveals or when — the differentiation is entirely in delivery, not in altering the scene's actual plot content. OUTPUT FORMAT 1. The diagnosis: at least two specific shared patterns in the current dialogue. 2. Each character's distinct speech-habit profile, in a short list. 3. The rewritten dialogue. 4. One line per character explaining a specific choice in the rewrite that ties back to their background.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Interchangeable dialogue is usually a symptom of a model (or a first draft) defaulting to one implicit voice — complete grammatical sentences, moderate hedging, neutral vocabulary — and applying it to every speaker regardless of background, because that neutral register is the safest, most broadly comprehensible way to convey plot information, and conveying information correctly is what a first pass optimizes for before voice is considered at all. Requiring an explicit diagnosis of shared patterns before any rewriting forces GPT-5.1 to identify the actual mechanism of the sameness rather than jumping straight to a surface fix, which matters because the two most common surface fixes — an accent or a catchphrase — treat voice as decoration applied on top of identical underlying sentence construction, and neither holds up across a full script the way differences in sentence completeness, question-versus-assertion ratio, or what a character avoids saying actually do. Deriving each character's speech habits from their stated background rather than from an arbitrary contrast (one blunt, one polite) ties the differentiation to something with in-fiction logic — a detective who's spent twenty years distrusting testimony would genuinely construct sentences differently from a retired teacher accustomed to filling nervous silence, and that causal link is what makes the voice read as characterization rather than an assigned quirk. The explicit instruction to preserve plot content and information timing while changing only delivery isolates the actual variable under test — voice — from the scene's substance, which prevents the rewrite from accidentally becoming a different scene instead of the same scene said differently, which is a common side effect when a model given loose license to "make the dialogue more distinct" changes what's said along with how it's said.
What you get back
DETECTIVE: She's lying. Not about everything — about the timeline. NEIGHBOR: Do you think so? I mean — I don't know, maybe I'm reading too much into the way she kept looking at the clock, but— DETECTIVE: Tomorrow morning. We go back.
Verified against
ChatGPT GPT-5.1 · 2026-08-12
Changelog
- 2026-08-12 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
