Can AI Write a Good Screenplay? An Honest Test
Not can it produce pages. It can produce endless pages. The real test is whether a stranger in a dark room feels anything, and that is a harder bar than the hype admits.
Can AI write a good screenplay? People ask it hoping for a clean yes or a comforting no. The honest answer sits in the gap between them, and to see it clearly you have to first fix the question. The machine can obviously produce a screenplay. It can produce a hundred before your chai goes cold. The real question is whether it can produce a good one, and to answer that you have to agree on what good means. Not correct. Good. Those are not the same word, and the whole confusion lives in mistaking one for the other.
So let me run the test the way I would run it at my own desk, honestly, giving the machine full credit where it has earned it.
What it passes without breaking a sweat
Hand a modern model a premise and it will give you something that looks unmistakably like a screenplay. Correct format. Scene headings in caps. Three acts that turn roughly where they should. A midpoint, a low moment, a climax. Dialogue that moves information from one character to another without falling over. It has read more scripts than any human who has ever lived, so it knows the shape of the thing cold. Even careful writeups on the field, like RIT's piece on screenwriters and the machine, grant that the structural competence is genuine and improving. Pretending the pages are gibberish is denial, and denial loses arguments.
If your bar for good is coherent and correctly built, the machine clears it. That should not reassure you. It should tell you the bar was in the wrong place.
Where the test actually happens
Here is my real test, the only one that has ever mattered in a dark room. Does anyone feel anything. Not admire the structure. Feel it, in the chest, the way you feel a wicket fall in the last over when you did not think your team could win. A screenplay is not a document. It is a machine for making strangers care about people who do not exist. Correctness is the chassis. Feeling is the engine. The model builds a beautiful chassis with no engine inside, and then revs the empty bay for you, and the sound is convincing right up until you notice the car has not moved.
Read enough machine pages and the pattern shows. The scenes are about the right things. The emotion is named correctly and never quite present. A character says the sad line at the sad moment, and it lands like a waiter reciting the specials. Everything is where it should be and nothing is alive, because the writer never lived. I made the fuller version of this case in the piece on whether AI replaces screenwriters, and it comes down to one wall the machine cannot climb.
A screenplay that is correct but not felt is a body without a pulse. It has every part and it is still not alive.Wr. Sarkhedi
The two things it cannot fake
When I look at where the machine collapses, it is always the same two places. First, it cannot want anything. Every good film runs on a character who wants something so badly we forget we are watching. The machine can label a want. It cannot burn with one, because it has never wanted a thing it was ashamed to want, and that shame is where the best scenes are born. Second, it cannot behave. The finest writing shows feeling through action rather than announcement, and a machine that has never folded a dead parent's laundry will always reach for the announcement. That is exactly the discipline of show, don't tell, and it is the discipline the machine fails most quietly and most often.
A test I ran at my own desk
Let me walk through an invented version of the test I actually run, scene by scene, so this stays concrete rather than theoretical. Say the premise is a small one: a retired schoolteacher in a Gujarat town discovers, going through her late husband's papers, that he sent money every month for thirty years to a former student she has never heard of. A machine given this premise will, competently, build you a serviceable spine. The teacher finds the ledger. She investigates. She meets the former student, now grown, and learns the truth behind the money. There is a scene of confrontation, a scene of revelation, a scene of quiet acceptance. The scene headings will be correct. The act turns will land roughly where a three-act structure expects them to land. If you read only the outline, you would call it sound.
Now sit inside the scene where she actually confronts the former student. A machine version tends to have the teacher say something like, "I do not understand why my husband kept this from me for thirty years," followed by an answer that explains the mystery cleanly. It is coherent. It is also the emotional equivalent of a form being filled out. A human writer who has ever kept a secret, or discovered one kept from her, knows that the real line in that scene is not a question. It is something sideways, something that reveals more by what it avoids saying. Maybe she does not ask about the money at all in that first exchange. Maybe she asks what her husband's laugh sounded like, because she has just realized this stranger heard it in rooms she never did. That choice, the sideways question instead of the direct one, comes from having lived inside jealousy and grief long enough to know they rarely announce themselves cleanly. The machine can be prompted toward that choice. It will not arrive there on its own, because it has no memory of what jealousy actually felt like at two in the morning.
The two things it cannot fake, examined closer
When I look at where the machine collapses, it is always the same two places, and they are worth taking apart individually rather than as a slogan.
First, it cannot want anything. Every good film runs on a character who wants something so badly we forget we are watching. The machine can label a want, can write "she wants closure" in a beat sheet, but it cannot burn with one, because it has never wanted a thing it was ashamed to want. That shame, that gap between the want and the self-image, is where the best scenes are born. Think of a character who wants his brother's failure more than he wants his own success, and hates himself for it. A machine can describe that dynamic if told to. It will not stumble into the specific, embarrassing detail on its own, the detail that makes the want feel like it belongs to an actual person rather than a case study.
Second, it cannot behave. The finest writing shows feeling through action rather than announcement, and a machine that has never folded a dead parent's laundry, never had to decide whether to keep a voicemail it cannot bear to delete, will always reach for the announcement instead of the gesture. That is exactly the discipline of show, don't tell, and it is the discipline the machine fails most quietly and most often. It knows the rule exists. It does not know, in its bones, why a folded shirt says more than a sentence about grief, because it has no bones.
A screenplay that is correct but not felt is a body without a pulse. It has every part and it is still not alive.Wr. Sarkhedi
What that means for a real screenplay
So can it write a good screenplay? It can write a passable one, fast, and passable is about to be worthless because it will be everywhere. It cannot write a screenplay that a stranger carries home and cannot put down. Here is the useful way to hold this, in practice:
- For the frame, the machine is a fair assistant. It will rough a structure and catch a hole, the same way a good outline in an outlining pass catches a hole before it becomes forty wasted pages.
- For the pulse, it is useless, because the pulse is lived and it has no life to draw on. This is the same wall discussed in observation and empathy as the writer's real instruments, tools no model actually possesses.
- For the whole, a screenplay is judged by its pulse, not its frame, which is why the answer to "can AI write a good screenplay" is no, even though the answer to "can AI write a screenplay" is technically yes.
Correct is what the machine gives away for free. Alive is what it has never once been.
The honest conclusion, and the good news inside it
The honest test returns an uncomfortable result for the hype and a hopeful one for you. The machine has raised the floor. Competent, correct, structurally sound pages are now free and infinite, which means competent is no longer a living. But raising the floor also raises the value of everything above it. A script that actually makes someone feel something is now rarer and worth more, not less, because it is the one thing the flood cannot produce.
I watched this exact sorting happen once already, when cheap content drowned the internet and everyone declared writing dead. What died was filler. What survived, and grew scarce, and grew valuable, was truth. The screen is next. So do not try to beat the machine at correctness, a game you will lose. Learn how to actually break in, then spend your life on the pulse it cannot manufacture, and keep your own conduct honest while you do it, which is the whole subject of the ethics of using these tools. The test the machine keeps failing is not a bug it will patch next year. It is the whole reason a human is still worth paying to write the thing.
Can AI write a full screenplay by itself?
It can produce something formatted correctly, with a plausible structure, acts that turn where they should, and dialogue that reads clean. Whether that qualifies as a good screenplay depends on whether a stranger in a dark room feels anything, which is a much harder bar than structural correctness.
What can AI actually do well for screenwriting?
It is a reasonable structural assistant. It can rough out a scene order, catch a plot hole, or suggest an alternate line reading fast. It is genuinely useful as a mirror and a first-pass tool.
Why can't AI write a truly good screenplay yet?
Two specific failures show up again and again: it cannot want something the way a person who has been ashamed of a want can, and it defaults to announcing emotion instead of showing it through behavior, because it has never actually behaved under grief, fear, or desire.
Does AI make human screenwriters less valuable?
It makes competent, correct pages cheap and abundant, which actually raises the value of writing that is genuinely felt, because felt work becomes the rarer, harder-to-fake thing in a market flooded with correct-but-empty pages.