AI Prompts for Writing Dialogue: A Playwright's Case Study

Make this dialogue more emotional and powerful. [PASTE SCENE]

Picture this: you're a community-theatre playwright with a staged reading in three weeks, and one scene is dying. Two characters — estranged sisters — meet after their father's funeral. On the page they explain their feelings to each other in clear, complete sentences. In your head it's devastating. Out loud, at the table read, the actors trail off, and one of them says the thing every writer dreads: "Nobody really talks like this."

That's the scene we're going to fix. It's a perfect test case for ai prompts for writing dialogue, because dialogue is where AI's default instincts are most wrong and most fixable.

The Problem the Playwright Faced

The sisters' scene had everything on the surface. "I've always felt like Dad loved you more," one said. "That's not true, I was just better at pretending," said the other. Clear. Motivated. Completely dead.

The problem is a classic one: the characters were saying exactly what they meant. Real people almost never do — especially not grieving people, especially not estranged sisters who've spent twenty years not saying things. The scene had no subtext, no evasion, no weapons disguised as small talk. It read like a therapy transcript written by someone who'd never been in therapy.

The Wrong Approach

Her first instinct was to ask AI to punch it up:

Make this dialogue more emotional and powerful. [PASTE SCENE]

The result was worse. The model turned the volume up — more tears, bigger declarations, a monologue about childhood. Now the sisters were saying what they meant louder. "More emotional" to a model means more emotional words, which is the opposite of what a buried, bitter scene needs.

⚠️ Common mistake: Asking AI to make dialogue "more emotional" or "more powerful." Those words push the model toward characters announcing their feelings, which is exactly what kills real dialogue. Powerful scenes usually have characters hiding feelings that leak out sideways. Ask for the mechanism — hiding, deflecting, attacking indirectly — not the mood.

The Correct Prompt

The fix started by giving each character something to protect:

Two estranged sisters meet after their father's funeral. Rewrite is NOT what I want — instead, help me set up the scene: - What is each sister trying NOT to say out loud? - What is each one secretly hoping the OTHER will say first? Give me your read, then I'll write the dialogue myself.

What this does: It builds the pressure system underneath the scene before a single line is written. Once you know what each character is hiding and hoping for, the dialogue almost writes itself — as evasion, not explanation.

Then, after she drafted a new version:

Here's my rewrite. For each line, tell me if the character is saying what they mean (mark ON-NOSE) or hiding/deflecting (mark INDIRECT). I want mostly INDIRECT with one ON-NOSE line at the emotional peak. Don't rewrite — just mark.

What this does: It gives you an X-ray of your own scene's directness. The goal isn't zero on-the-nose lines — it's earning the one direct line by surrounding it with evasion.

⚡ Pro tip: The single on-the-nose line, placed at the peak, is where a scene breaks open. If everything is indirect, the audience gets lost; if everything is direct, there's no build. One honest line, late, after a scene of dodging, is the pattern that makes people cry.

Results and What Changed

The rewritten scene barely mentioned the father. The sisters argued about who was taking the good casserole dish home. They discussed logistics — the house, the paperwork, who'd call the lawyer. And underneath every mundane line, the twenty years of hurt pressed up. Then, near the end, one sister said quietly, "I didn't know how to call you," and the whole thing landed — because it was the first honest thing either had said.

At the next reading, the actors stopped trailing off. One of them teared up. Nobody said "people don't talk like this," because now they did.

⚡ Pro tip: Great dialogue is usually about the wrong thing on purpose. Characters fight about the dishes because they can't fight about the grief. Ask the model: "What mundane thing could these characters argue about that stands in for what they really can't discuss?" That displacement is the engine of subtext.

How to Apply This to Your Situation

The method transfers to any dialogue-heavy work. A novelist can run the ON-NOSE/INDIRECT marker on any conversation that feels flat. A UX writer crafting chatbot personas can use the "what is this character hiding" prompt to give a bot texture instead of chirpy helpfulness. A video game writer building companion characters can set up protection systems so party banter has an undertow.

The core move is always the same: build the pressure underneath before you write the words on top. Dialogue is the visible tip of an invisible system of wants and fears. Get the system right and the lines follow.

⚡ Pro tip: When you're stuck, ask each character out loud, "What do you want from the other person in this scene, and why can't you just ask for it?" The answer to the second half is your subtext. If a character can just ask, the scene has no tension — give them a reason to hold back.

Testing Dialogue Against the Actor

The truest test of dialogue is whether a person could actually say it out loud, in one breath, meaning it. AI can simulate that read:

Read each line of this dialogue as if you're an actor about to perform it. Flag any line that would be hard to say naturally — too long for one breath, too written, tongue-twisting, or emotionally unmotivated. Don't rewrite; just flag. [SCENE]

What this does: Written dialogue and spoken dialogue diverge in specific ways — real speech is shorter, messier, full of interruption. This catches the lines that only work on the page.

⚡ Pro tip: Real people interrupt, trail off, and talk over each other. Ask the model to mark where your scene is "too polite" — where characters wait their turn like they're in a meeting. Overlap and interruption are how you make a scene sound alive rather than transcribed.

⚡ Pro tip: Give each character a distinct verbal fingerprint — one uses long winding sentences, another clips everything short. Ask the model: "Could I tell who's speaking with the dialogue tags removed?" If not, your characters share one voice, and the scene flattens.

Next Steps

Take your flattest scene and run the two prompts above in order — set up the hidden wants, then mark your draft ON-NOSE versus INDIRECT. Aim for mostly indirect with one earned honest line. You'll feel the scene change under your hands.

The setup prompt and the directness-marker are the two tools you'll reuse on every dialogue scene you ever write, so don't leave them buried in a chat log. Save them in PromptABCD as a dialogue pair — "pressure setup" and "on-nose audit" — and your subtext toolkit is ready the next time a scene reads like a therapy transcript.