The Screenplay AI Gets Wrong — and What a Trained Script Skill Fixes

You paste in a concept — two characters, a setting, a rough idea of what needs to happen — and Claude gives you something that looks like a screenplay. The INT./EXT. headers are there. The action lines are present tense and reasonably lean. The dialogue is spaced correctly. You read through it and feel, briefly, like you have something.

Then you try to work with it. The character who was established as withdrawn in scene one is suddenly cracking jokes in scene three with no bridge between. The act break lands in the wrong place — the tension peaks early and then flatters out before the climax. The dialogue is grammatically fine but sounds like no one you've ever met. You can't shoot this. You can barely revise it, because fixing one scene pulls threads that run through four others.

This is the specific failure mode of generic AI in screenplay writing: it produces structure without story. It knows what a script looks like but doesn't understand what a script does. What changes things is bringing actual craft knowledge into the generation process — knowing what's working in the genre right now, what structure the story's concept demands, and how the specific characters you've described actually speak.

Why Script Format Fools You Into Thinking You Have a Draft

Screenplay format is highly standardised. One-inch margins, Courier 12pt, sluglines in a specific style, action blocks written in present tense, character names centred above dialogue. Generic AI learned all of this from the vast amount of correctly-formatted script content on the internet. It formats impeccably.

That precision creates a problem. A page of well-formatted screenplay text reads, at first glance, like professional work. The visual structure of the medium — the white space, the clean blocks, the familiar layout — carries an authority that a page of raw prose doesn't. You're much more likely to read through a badly-formatted short story and notice it's bad than to read through a correctly-formatted but dramatically inert screenplay and notice the same thing.

The problems in a generic AI script are structural and below the surface. Scene transitions exist but don't build. Characters speak in ways that reveal information to the audience rather than in ways those particular people would actually speak. The climactic moment arrives, technically, in the right place — but there's no weight behind it because the preceding scenes didn't earn it. These aren't typos you can fix. They're architectural.

What Live Research Changes for a Script Skill

A screenplay isn't an eternal form that exists outside of time and market. Every genre has conventions that shift — what's reading as fresh versus tired, what structural approaches are getting traction with development executives, which dialogue styles feel contemporary versus which ones mark the writer as working from a 2015 mental model of what good scenes look like. A skill built to research before it writes goes and finds this out. Not from training data. From what's current.

That research shapes the output in ways that are specific rather than decorative. A thriller concept in 2026 gets treated differently from the same concept in a skill that learned from scripts written ten years ago — because the pacing expectations, the character complexity readers want, and the types of twists that still land have all shifted. Generic AI doesn't know this. It averages across everything it was trained on and produces something that could have been written in any year.

The calibration also extends to production context. A short film script for a micro-budget shoot has different requirements than a feature spec that needs to function as a writing sample. A corporate video script needs to deliver a specific message inside a specific runtime. A pilot needs to establish a world and leave room for the series — which is a completely different structural problem than a standalone screenplay. The Screenplay / Script skill asks where you are before it writes — and the output reflects that, specifically.

What the Screenplay / Script Skill Actually Does

The skill opens with three calibrating questions that define the output before a single line is written. Genre, production context, and character intent — these aren't decorative. They're the variables that determine the entire structural approach. Here's how it works in practice:

Generic Output vs NovaKit Output

The difference becomes concrete in a single scene. Here's the same brief — a tense confrontation between two estranged siblings at their father's house the day after his death — handled two ways:

The difference isn't polish. It's craft. The generic version tells you there is tension. The calibrated version creates it — through what isn't said, through action that carries subtext, through dialogue that reveals two people who share a shorthand built from years of conflict. The left side could have been written by anyone. The right side could only have been written with an understanding of how scenes actually work.

Who Gets the Most from Screenplay / Script

The primary audience is writers who already understand screenplay craft well enough to know when something is wrong — but who are using AI to accelerate the first-draft process, explore structural alternatives, or unblock scenes they've been circling. For this person, the failure mode of generic AI isn't just frustrating. It's professionally useless. A script that looks right but reads flat doesn't help you pitch, sell, or produce anything. What they need is a starting point that's actually workable — not a formatting exercise.

The skill also does significant work for non-writers who need professional script output: marketing teams producing brand video scripts, founders building investor pitch videos, educators creating structured lesson content in script format. For these users, the calibration to format and message rather than dramatic arc is the relevant differentiator. The skill knows which mode you're in because you told it.

The Output You Walk Away With

The skill produces a complete draft structured to your specified format — whether that's a short film of 10–15 pages, the opening act of a feature, a pilot cold open, or a 3-minute corporate video script. What comes back isn't a fragment or a scene list. It's a usable draft with character voices established, scene structure built for the genre you specified, and action lines that are shootable rather than descriptive.

The path from there is revision rather than rebuilding. You have something with bones — a structure that can absorb edits, a voice that can be refined, scenes where the dramatic logic is present even if individual lines need work. That's the difference between AI output that's a starting point and AI output that's a liability. The Screenplay / Script skill gives you the former.

A script that looks like a screenplay but doesn't work like one is worse than a blank page, because it tricks you into thinking the problem is solved. The formatting is irrelevant if the story isn't there. The skill exists to make sure both arrive together.