Verified against ChatGPT · 2026-08-10
Draft a two-character confrontation scene in proper screenplay format, driven by subtext not statements
Writes a formatted screenplay scene where two characters want incompatible things and neither says so directly, using action lines and dialogue to carry what's actually being negotiated underneath the surface conversation.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Draft one screenplay scene in standard format (slugline, action lines, character names in caps, dialogue) between exactly two characters who each want something the other one won't give them directly. SLUGLINE AND CONTEXT INT. KITCHEN - NIGHT. Two former business partners meet for the first time since one bought the other out of the company. CHARACTER A WANTS An apology, or at least an acknowledgment that the buyout was a betrayal, not just business. CHARACTER B WANTS Forgiveness, without having to admit the buyout was personal as well as strategic. WHAT NEITHER WILL SAY OUT LOUD They were also ending a romantic relationship at the same time as the business split, and both are pretending this meeting is only about the company. PAGE LENGTH 2-3 pages RULES Neither character should state their actual want directly at any point in the dialogue — the scene's job is to dramatize the gap between what's said and what's meant, so every line of dialogue should be doing something other than announcing intent: deflecting, testing, probing, changing the subject, or making a smaller ask that stands in for the real one. Use action lines only for what a camera could actually capture — a physical action, a look held a beat too long, an object picked up or set down — never for a character's internal thought or motive; if the scene needs the reader to know a motive, it has to be inferable from behavior and dialogue, not stated in an action line. Give the scene a clear turn: a moment partway through where the balance of power or information shifts, so the scene isn't just two people repeating their positions until the page runs out. End on the beat where the unresolved tension is at its sharpest, not on a line that resolves or explains it — a strong two-hander often ends on a exit, a refusal, or a line that reframes everything just said, not a summary. WHAT NOT TO DO Do not have either character deliver an on-the-nose line stating They were also ending a romantic relationship at the same time as the business split, and both are pretending this meeting is only about the company. even once, even as a slip — the entire point is that it stays beneath the surface and the audience infers it. Do not pad the scene with small talk that doesn't advance the underlying negotiation between the two wants; every line should be doing double duty as surface conversation and subtextual maneuvering. OUTPUT FORMAT 1. The scene in standard screenplay format. 2. A short subtext key below it: for each major exchange, one line stating what was actually said versus what was actually meant, so I can check the subtext is landing as intended.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Two-hander confrontation scenes live or die on subtext, and the single most common failure mode when a model is asked to write conflict dialogue is to have characters state their wants and grievances plainly, because explicit statement is the most direct path to conveying information and models default toward directness unless explicitly redirected. Naming both wants and the unspoken stakes for the writer, while explicitly banning either character from voicing the unspoken stakes even once, forces GPT-5.1 to route the actual content of the scene through indirect dialogue moves — deflection, a smaller stand-in ask, a pointed silence — which is mechanically closer to how real confrontations under emotional stakes actually play out, since people rarely announce what they most want precisely when it matters most. Restricting action lines to only camera-capturable behavior addresses a specific screenplay-format failure: models frequently slip into prose-style interiority in action lines ("she feels betrayed but hides it"), which is unfilmable and is also a category error in the format itself — a script is a blueprint for what can be shot, not a description of interior states, and a director or actor has nothing to do with an unfilmable line. Requiring a clear power-shift turn partway through prevents the flatter default of two characters repeating their positions at each other until the page count is met, which is what happens when a model treats "conflict" as static opposition rather than a dynamic exchange with a hinge. Ending on the sharpest unresolved beat rather than a resolution matters because scenes that resolve too cleanly undercut the tension the whole page built, and a model given no explicit instruction about the ending will often default to a soft, conclusive final line out of a general pull toward narrative closure.
What you get back
A stands at the counter, turning a coffee mug slowly in her hands without drinking from it. B watches the mug, not her face. B: You still take it black. A: (a beat) You remembered that and not the part where you had Legal call me on a Friday.
Verified against
ChatGPT GPT-5.1 · 2026-08-10
Changelog
- 2026-08-10 — 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
