Story continuity

How to Keep Characters Consistent When Writing With AI: Voice and Knowledge

Keep dialogue and viewpoint distinct with voice rules, knowledge boundaries, a blind dialogue test, and targeted revision prompts.

When writing with AI, keep characters consistent by defining how each person notices, speaks, and interprets information. A biography alone is not enough. Two characters can have different histories yet sound identical if the drafting prompt gives them no distinct conversational goals or limits on what they know.

This guide focuses on voice and viewpoint. For the broader question of motive and behavior, use the character contract guide.

Define voice as choices, not a catchphrase

Record sentence habits, attention, evasion, and emotional pressure. Mara might name physical details, ask practical questions, and avoid promises. Ivo might answer indirectly, notice routes and distances, and become unusually precise when afraid.

Avoid making every line demonstrate the same trait. A character who always speaks in metaphors becomes a device rather than a person. The voice rules should guide selection, not force a verbal signature into every sentence.

Give dialogue a private goal

Before a conversation, write what each speaker wants from the other. Mara wants Ivo to reveal what is missing from the map. Ivo wants her to keep moving before the tide rises. Their dialogue can remain distinct because their strategies differ, even without obvious stylistic quirks.

Illustrative exchange:

MARA: That corner was there this morning.
IVO: The lower stair will be underwater in six minutes.
MARA: Then you have five to explain it.

Mara points to evidence; Ivo redirects to urgency. The third line advances both the conflict and the scene's time pressure. This invented exchange demonstrates a method, not a model benchmark.

Track who can know what

Attach a knowledge boundary to the viewpoint request. Separate direct observation, information received from someone else, suspicion, and objective truth. A character may suspect deception without knowing the hidden motive.

Ask the assistant to flag any sentence that states a fact outside the viewpoint character's knowledge. Review those flags yourself: deliberate omniscient narration is different from an accidental slip inside a limited viewpoint.

Run a blind dialogue check

Remove speaker labels from a short passage and read it aloud. Can you infer the speakers from their goals and responses, or only from repeated names and catchphrases? If every line could belong to either person, revise the conversational strategies before adding decorative vocabulary.

This is an editorial exercise, not a rule that every line must uniquely identify its speaker. Natural conversation includes ordinary shared language. Look for a pattern across the exchange.

Revise without changing the facts

Revise this exchange to distinguish the speakers' strategies.
Mara: seeks concrete evidence; avoids promises.
Ivo: redirects toward practical urgency; conceals the map's missing part.
Preserve the outcome, known facts, and viewpoint limits.
Do not add a confession, new backstory, or repeated catchphrase.
Explain the changed choices after the dialogue.

Compare the revision against the original. An assistant may improve surface style while making Ivo reveal information too early. Keep the accepted knowledge state in your chapter ledger and update it only when the scene actually discloses something.

Try the check on one conversation from each major relationship. Voice becomes more convincing when it responds to who is listening and what is at stake, while remaining grounded in the same underlying character.

FROM THE PAGE TO THE FRAME

Have a story in mind?

Shape a concept in the local Ficlora workbench. Save it in your browser or export it as text.

Explore the workbench