Dialogue: Subtext and Voice Differentiation
Boundary note: this reference covers dialogue as a planning and
structure tool — what each character secretly wants in the scene, and a
checkable recipe for making voices distinguishable. Line-level prose quality
(literary dialogue rhythm, anti-generic phrasing) belongs to the
better-writing skill. Consult that skill before polishing dialogue prose.
Subtext as a planning input
Subtext is the gap between what is said and what each character actually wants. Before drafting any conversation, write down for each participant:
- Said want: what the character claims to want in this scene (their spoken goal).
- Hidden want: what they actually want — usually emotional, often something they would never say aloud (approval, safety, revenge, proof they matter).
- Why they can't say it: pride, fear, leverage, ignorance. The reason shapes every evasion.
Record these in the scene record's planning notes or the chapter outline:
dialogue-subtext:
- character: {character-kebab}
said-want: "{What they claim to want}"
hidden-want: "{What they actually want but won't say}"
- character: {character-kebab}
said-want: "..."
hidden-want: "..."When said-want and hidden-want are the same, the dialogue has no subtext — either the character is being honest (mark it as a deliberate choice) or the exchange needs a hidden layer added.
Dialogue as negotiation
Treat every exchange as a negotiation, not a Q&A. Each turn of dialogue should do one of:
- Probe: test the other person for information, weakness, or intention
- Parry: deflect a probe without revealing anything
- Advance: trade information for something, concede ground to gain ground
- Feint: appear to concede while actually maneuvering
Dialogue where characters take turns delivering information to the reader is an interrogation, not a conversation. If both characters' turns only pass facts, rewrite so at least one participant is trying to get something the other won't freely give.
Voice differentiation recipe
Build each character's dialogue voice from four checkable components
(recorded in their Voice & Speech Patterns profile section, with example
lines):
- Verbal tic: one recurring word, phrase, or hedge (not a caricature — one per character, used sparingly).
- Syntactic preference: short clipped sentences vs. long rolling ones; questions vs. declarations; formal vs. fragmentary.
- Topic steered toward under stress: when cornered, what does this character retreat to talking about?
- Topic refused: what will this character never discuss, and what do they do instead when it comes up (deflect, joke, go silent, leave)?
The tag-swap test
Strip the dialogue tags and speaker names from a finished exchange. Hand the bare lines to a cold read. If the voices are distinct, a reader familiar with the characters should be able to reassign most lines correctly. If not:
- The voices are not distinct enough — strengthen one character's syntactic preference and stress-topic, not by adding more tics.
- Do not "fix" it by giving every character a catchphrase. Tics are seasoning; syntax and motive are the meal.
Beat economy
- Keep ~3 spoken lines max between action beats or sensory anchors. A beat is a line of physical action, a noticed detail, or a silence with weight.
- Beats do double duty: they carry subtext (crossed arms while saying "I'm fine") and they break up speeches.
- For arguments: allow interruptions and cut-offs. Let one character finish another's sentence wrongly. People fighting don't wait their turn, and they answer the argument they heard, not the one that was spoken.