Prompt Library

Dialogue Prompts That Start Mid-Argument

26 copy-paste prompts

26 opening lines and two-line exchanges built to drop you into a scene that's already in motion. Each one comes with the tension it creates and a craft note on making the conversation land.

In short: This page contains 26 copy-paste ready prompts, organized into 5 categories with a description and pro tip for each. The first 5 prompts are free instantly, no signup needed. Hand-curated and tested by the AI Academy team.

Louis Corneloup
By Louis Corneloup · Founder, Techpresso
Last updated ·Hand-curated & tested by the AI Academy team

Tension & Conflict Openers

5 prompts

The Discovery

1/26

✨ What it does

ChatGPT writes a scene that opens with You weren't supposed to find that and And yet here we are, revealing both the object and the relationship. Keep the draft where the found thing still shocks you, then write the next line yourself.

"You weren't supposed to find that." / "And yet here we are."

💡

Pro tip: Delay naming the thing as long as possible. The longer both characters talk around it, the more the reader leans in.

The Quiet Ultimatum

2/26

✨ What it does

ChatGPT writes a scene from I'm only going to ask you once, with a history of polite asks and a future of consequences. Play the other person as someone who will not flinch, then cut any speech that explains your threat.

"I'm only going to ask you once."

💡

Pro tip: Let the listener respond with an action beat instead of words — pouring coffee, checking a phone. Calm defiance reads louder than a comeback.

The Funeral Argument

3/26

✨ What it does

ChatGPT writes an argument that starts with This is not the place and It's the only place you can't walk out on me, under funeral decorum. Keep voices low enough for an audience next door, then raise only one line of yours.

"This is not the place." / "It's the only place you can't walk out on me."

💡

Pro tip: Use the surroundings as interruptions — a hand on a shoulder, a hymn starting — so the argument keeps getting compressed instead of released.

The Wrong Apology

4/26

✨ What it does

ChatGPT writes a scene from I said I was sorry and You apologized for the wrong thing, digging into two versions of the same event. Pick the draft where they still do not agree on the facts, then stop before you let them reconcile.

"I said I was sorry." / "You apologized for the wrong thing."

💡

Pro tip: Never let the wronged character explain it outright. Make the apologizer guess, and let each wrong guess deepen the damage.

The Borrowed Voice

5/26

✨ What it does

ChatGPT writes a scene that rejects borrowed words and demands the speaker say it in their own, stripping a coached speech mid-talk. Let them fumble once the script is gone, then keep the plainer version on your page.

"Don't use her words on me. Say it in yours."

💡

Pro tip: Show the coached lines as slightly too polished, then break the rhythm when the character finally speaks for themselves — shorter sentences, rougher grammar.

Romance & Banter

5 prompts

The Terrible Liar

6/26

✨ What it does

ChatGPT writes flirtatious banter from You're staring and I'm cataloguing your flaws, with both people pretending they do not know. Read it aloud with a partner, and keep the lie that still sounds like a compliment to you.

"You're staring." / "I'm cataloguing your flaws. It's taking a while."

💡

Pro tip: Banter works when both characters are good at it. Give the second speaker a return of serve, not just a blush.

The Almost-Confession

7/26

✨ What it does

ChatGPT writes a scene from Why did you really come back and Bad weather, where the two-word answer wants to be caught. Decide whether the other person pushes or lets it go, then write only that choice yourself.

"Why did you really come back?" / "Bad weather."

💡

Pro tip: Subtext thrives on the unsaid. Let the honest answer surface in a gesture — taking off a coat, sitting down — rather than in words.

The Old Joke

8/26

✨ What it does

ChatGPT writes a reunion from You still do that thing, What thing, and You know exactly what thing, using a shared joke neither can take back. Name the joke before you paste, then keep the draft that does not explain it.

"You still do that thing." / "What thing?" / "You know exactly what thing."

💡

Pro tip: Never explain the joke to the reader. Its meaning matters less than the fact that only these two people share it.

The Logistics of Love

9/26

✨ What it does

ChatGPT writes a scene from This is a terrible idea and a yes that still books eight o'clock, treating a reckless date like a calendar invite. Keep the brisk yes, then add only the one practical detail you would actually text.

"This is a terrible idea." / "Agreed. Pick you up at eight?"

💡

Pro tip: Play the scene completely straight. The less the characters acknowledge the romance, the more the reader feels it.

The Wrong Name

10/26

✨ What it does

ChatGPT writes a scene from You called me by his name and I called you by your name, flipping a mishear into closer attention than anyone expected. Play both the braced-for-betrayal beat and the reveal, then stop on the listen for you.

"You called me by his name." / "I called you by your name. You just weren't listening for it."

💡

Pro tip: The beat after this exchange matters more than the exchange. Hold the silence for a full sentence of description before anyone speaks again.

Mystery & Secrets

5 prompts

The Unlocked Door

11/26

✨ What it does

ChatGPT writes a domestic threat scene about a door the speaker swears they locked, answered by Then someone else has a key. Decide who has the extra key before you run it, then keep the colder how-long question.

"I locked it. I know I locked it." / "Then someone else has a key."

💡

Pro tip: Keep the dialogue clipped and practical — people in fear talk logistics. Save the emotion for the body language.

The Witness

12/26

✨ What it does

ChatGPT writes a scene from I saw you there that night and No, you saw someone wearing my coat, where the denial is worse than a confession. Choose whether the coat is a lie, then write the next accusation yourself.

"I saw you there that night." / "No. You saw someone wearing my coat."

💡

Pro tip: A good denial should open a door, not close one. Let the second character's explanation raise a question the accuser never thought to ask.

The Inherited Lie

13/26

✨ What it does

ChatGPT writes a reunion from Mom told me you were dead and She told me the same about you, stacking grief and anger in one exchange. Let both sit with the shared lie, then do not rush the hug on your page.

"Mom told me you were dead." / "She told me the same about you."

💡

Pro tip: Resist the urge to vilify the absent liar immediately. The scene is richer if both characters keep instinctively defending her.

The Professional

14/26

✨ What it does

ChatGPT writes a detective exchange from How did you know it was me and You're the only one who never asked questions, where caution is the tell. Keep the restraint as the proof, then cut any speech that gloats for you.

"How did you know it was me?" / "You're the only one who never asked questions."

💡

Pro tip: Work backward from the reveal. Plant two or three scenes earlier where the character conspicuously doesn't ask the obvious question.

The Recording

15/26

✨ What it does

ChatGPT writes a confrontation that starts after the tape is already heard, so the accused must guess how much you know. Negotiate in real time, and stop when they offer the first real trade you would take.

"Before you say anything — I've already heard the tape."

💡

Pro tip: Let the accused probe with partial admissions — "So you know about the money" — and watch the other character's face for whether the tape covered that.

Like these prompts? There are full tutorials behind them.

Learn the workflows, not just the prompts. 300+ easy-to-follow tutorials inside AI Academy — and growing every week.

Try AI Academy Free

Sci-Fi & Fantasy

5 prompts

The Translation Problem

16/26

✨ What it does

ChatGPT writes a bargain with something inhuman from Your language has no word for what I am offering and Then I'm definitely not signing. Let wit be the human's only armor, then refuse any term you cannot name.

"Your language has no word for what I am offering." / "Then I'm definitely not signing."

💡

Pro tip: Give the inhuman speaker a consistent verbal quirk — no contractions, literal idioms — so its voice stays alien without phonetic gimmicks.

The Last Broadcast

17/26

✨ What it does

ChatGPT writes a radio scene from Station Twelve, do you copy and You shouldn't still be transmitting from there, where relief turns to dread. Decide what happened to the station before you generate, then keep the line that knows too much.

"Station Twelve, do you copy? Anyone?" / "...You shouldn't still be transmitting from there."

💡

Pro tip: Radio dialogue forces voice-only characterization. Use protocol language breaking down — call signs dropped, procedure abandoned — to show fear.

The Apprentice's Question

18/26

✨ What it does

ChatGPT writes a mentor scene from You taught me everything and No, I taught you everything I wanted you to know. Let the apprentice audit years of trust mid-talk, then cut the mentor's smirk if it reads as a lecture to you.

"You taught me everything." / "No. I taught you everything I wanted you to know."

💡

Pro tip: The apprentice's next line shouldn't be anger — make it a specific question about one lesson. Specificity makes betrayal concrete.

The Time Debt

19/26

✨ What it does

ChatGPT writes a return scene from How long was I gone and Depends, how long do you think you were gone, with the answer stalling. Pick cryosleep, a portal, or a curse yourself, then keep the hedge that is worse than any guess.

"How long was I gone?" / "Depends. How long do you think you were gone?"

💡

Pro tip: Let the returned character name a number first. The gap between their guess and the truth is the scene's entire emotional payload.

The Oath Loophole

20/26

✨ What it does

ChatGPT writes a fantasy betrayal from You swore you would never hurt me and the loophole of standing aside instead. Play the technically honest oath, then let the colder standing-aside land last on your page.

"You swore you would never hurt me." / "I swore I would never hurt you. I said nothing about standing aside."

💡

Pro tip: Establish the exact wording of the oath in an earlier scene so the reader can replay it and realize the loophole was always there.

Dialogue Craft Exercises

6 prompts

No Tags Allowed

21/26

✨ What it does

ChatGPT writes the full scene after We need to talk about what you did with the money, with no he said or she whispered, so each voice is only vocabulary and rhythm. If you lose who is speaking, generate again until the voices stay distinct.

"We need to talk about what you did with the money." — Write the full scene that follows without a single dialogue tag. No "he said," no "she whispered."

💡

Pro tip: Give each character one verbal habit before you start: one interrupts, one answers questions with questions. Distinct habits replace tags.

The Subtext Drill

22/26

✨ What it does

ChatGPT writes a two-page scene where Are you staying for dinner is the only question that matters, and the real subject is whether the relationship survives the year. Neither character may mention the relationship; keep the draft that still feels like dinner to you.

"Are you staying for dinner?" — Write a two-page scene where this is the only question that matters, but the real subject is whether the relationship survives the year. Neither character may mention the relationship.

💡

Pro tip: Pick a concrete object — a set table, an unopened bottle — and let characters handle it whenever they can't say the true thing.

The Interruption Engine

23/26

✨ What it does

ChatGPT continues Before I tell you, promise me you won't and I'm not promising anything, with no one ever finishing a sentence that holds the secret. Let interruptions do the rhythm, then stop before the secret is said on your page.

"Before I tell you, promise me you won't—" / "I'm not promising anything." — Continue the scene, but no character is ever allowed to finish a sentence that contains the secret.

💡

Pro tip: Interruptions need a beat to land on — a door, a phone, a third person entering. Vary the interrupters so the device doesn't show.

One Side Only

24/26

✨ What it does

ChatGPT writes a whole phone call from one side only, starting from twenty years being a long time to wait to tell someone. Reconstruct the unheard half from your pauses and repeats, then leave that half off the page.

"No. No, I understand. I just think twenty years is a long time to wait to tell someone." — Write an entire phone call, but the reader only hears this character's side.

💡

Pro tip: Write both sides first, then delete one. The remaining lines should bend slightly toward what was cut, like a wall that remembers a door.

The Beat Rewrite

25/26

✨ What it does

ChatGPT writes the exchange that leads to taking the house and all of it twice: once with only said and asked, once with only action beats. Compare the two side by side, and keep the pace you want for the scene.

"Fine. Take the house. Take all of it." — Write the exchange that leads to this line twice: once using only dialogue tags (said, asked), once using only action beats (she folded the letter, he didn't sit down).

💡

Pro tip: Compare your two versions line by line. Wherever the beat version is stronger, the action is doing emotional work — keep those. Cut beats that just choreograph.

The Eavesdropper

26/26

✨ What it does

ChatGPT writes She can never find out it was us from a third person who overhears and catches only every other line. Assemble the truth from the fragments, then decide what the eavesdropper will do with it yourself.

"She can never find out it was us." — Write the scene from the point of view of a third character overhearing it, who only catches every other line.

💡

Pro tip: Choose carefully which lines get lost. The missing halves should be the ones the eavesdropper most needs, so each heard line reframes their guess.

Free tool

Prompt Optimizer

Turn a rough idea into a structured, professional AI prompt.

Try it free →

Frequently Asked Questions

Dialogue prompts are opening lines or short exchanges — usually one to three lines of speech — that you build a scene around. Instead of starting from a premise or a character sheet, you start from words already being spoken, which drops you into the middle of a conflict and forces you to work out who these people are and what they want from each other.
Start by deciding what each character wants from the conversation and what they're unwilling to say out loud. Good dialogue runs on subtext: the words are a negotiation, and the real subject sits underneath them. Then control pacing with beats — small physical actions between lines — and keep tags simple. "Said" is nearly invisible; ornate tags like "expostulated" pull the reader out of the scene.
Either works. A single line ("I'm only going to ask you once.") gives you total freedom over who answers and how. A two-line exchange locks in a dynamic — accusation and deflection, offer and refusal — which is useful when you want to practice a specific kind of tension rather than invent one from scratch.
Aim for 500 to 1,500 words. Long enough that the conversation has to turn at least once — someone gains or loses ground — but short enough that you can finish in a single sitting. If the scene wants to keep going past that, you've probably found a story rather than an exercise, and you should follow it.
Yes, and the most useful approach is adversarial: write your scene first, then ask the AI to play one character while you play the other, or to flag lines where both characters sound the same. AI is good at generating variations and stress-testing voice consistency. It's weakest at subtext, which is exactly the part worth practicing yourself.

Prompts are the starting line. Tutorials are the finish.

A growing library of 300+ hands-on tutorials on ChatGPT, Claude, Midjourney, and 50+ AI tools. New tutorials added every week.

7-day free trial. Cancel anytime.