AI Script Analysis
Coverage tells you how the script feels. Analysis tells you what the script is — scene by scene, page by page, in numbers that don't care about your feelings. Here's the quantitative read, and the one rule that keeps it from flattening your movie.
Every writer carries a mental map of their script, and every writer's map is wrong. You think your first act is lean because you remember it as lean. You think your lead is on every page because she's on every page of your imagination. The draft itself — the actual pages — tells a different story, and until AI came along, seeing that story meant a weekend with index cards and a highlighter. Now you can paste a script into ChatGPT or Claude and get the honest map in minutes. That's AI screenplay analysis: not opinions, measurements.
This is a different tool than the coverage from last chapter, and it's worth keeping them separate in your head. Coverage is a reader's judgment. Analysis is a survey crew — it walks the property and reports the terrain. Analysis tells you where. Craft tells you what to do about it. Confuse the two and you end up taking rewrite orders from a measuring tape.
The script by the numbers
A clean export helps here — Fountain files are ideal because they're plain text, and a Final Draft export to text works fine too. Then ask for the quantitative read. These are the measurements that have actually changed my drafts:
- Scene lengths — every scene, longest to shortest. Three five-page scenes in a row is a pacing fact, whatever the writing quality.
- Dialogue-to-action ratio — overall and scene by scene. Watch for the stretch where the movie quietly becomes a radio play.
- Character page-presence — first appearance, last appearance, and the longest gap for each principal. Vanishing acts hide here.
- Actual act breaks vs. intended ones — ask where the story turns on the page, then compare to where you meant it to turn.
- Scene-purpose tagging — have it label each scene: advances plot, reveals character, both, or neither. The "neither" list is your cut list's first draft.
- Location and time-of-day spread — a bonus for your producer brain: repetition here is a budget note as much as a story note.
The single most revealing item on that list is the act-break comparison. When the AI reads your act two as starting eight pages after you think it starts, that's not a machine error — that's eight pages where your protagonist has decided something in your head but not on the page. I've never run that comparison on a draft, mine or anyone's, without learning something uncomfortable.
Finding the sag — and leaving the good weirdness alone
Where do the numbers point? Act two. It's always act two. The sag shows up in the data before you'd ever admit it in a read: scene lengths creep up, the dialogue ratio climbs, the antagonist's page-presence drops off a cliff, and somewhere around the middle you find a run of scenes tagged "reveals character" with nothing tagged "advances plot" for ten pages. That cluster of symptoms is the sag, located with an accuracy no gut-read can match. What the numbers can't tell you is the cure — whether those ten pages need a complication, a compression, or a scene where somebody finally does the thing they've been avoiding. That decision is the job. It's your job. The analysis just tells you which pages to have the argument about.
And one warning, because this is where quantitative thinking turns toxic: do not sand off intentional irregularity. Some of the best scripts ever written have a seven-page dialogue scene sitting dead center, or a first act that's all texture, or a lead who disappears for twenty pages so their absence can do the work. If an "anomaly" in your numbers is a choice — if you can say out loud what it's doing for the story — keep it and defend it. Analysis serves the story, not the average. A script that measures perfectly normal is usually a script nobody remembers.
A director I was cutting for swore his thriller's midsection was tight — he'd rewritten it four times and could recite it. We ran the page-presence numbers and his villain was absent for a huge stretch of the second act. Not underused. Absent. Four human passes had missed it because every individual scene read fine; the pattern only existed across thirty pages. One new villain scene, placed where the gap was widest, and the whole back half started ticking. The numbers found in five minutes what four rewrites couldn't.
You now have judgment (coverage) and measurement (analysis). Next chapter we bolt everything together into the engine that actually produces finished screenplays: the rewrite loop — draft to notes to passes and around again, on a stack that costs nothing.
Everything Module 3 teaches, built into one tool — Scene Analysis reads your script and returns beat-by-beat structure, character, and pacing notes. Coverage-grade feedback on your draft, in minutes.