THE STORY VAULT BLOG
AI is so controversial. And I don't understand why.
AI can be abused. Horribly abused. When fakes are created, either of celebrities or average people, AI can cause emotional harm that is simply heinous. There's no controversy here. We cannot permit this. Safeguards must be instituted. Punishments for those who circumvent those safeguards must be harsh. This is not controversial.
People who use AI to write stories for them are not writers. This is not controversial.
AI is really bad at writing fiction. Those who fear that AI is a better writer than they are will never be professional writers. They will never be more than hobbyists. I don't think this is controversial.
AI has incredible potential to create benefits for people and society. AI is neither good nor evil. We humans are in control. If the end result is negative it will be only because we allowed our flaws to supersede our better natures.
How Should Writers Use AI?
Hell if I know. I can only tell you how I use it and why.
When I complete a draft I provide the following prompt to ChatGTP, Mistral Chat, and Gemini -
Analyze and rate this screenplay on a scale of 1 - 10, in increments of .1, using the following criteria:
Concept & Premise
Story, Structure & Pacing
Tone, Genre & Originality
Characters & Arc
Dialogue
Action & Visuals
Themes & Resolution
Emotional Impact
Originality & Voice
Overall Cohesion & Rewatch Value
Quality of Writing
Overall Score
The AI returns an evaluation of the strengths and weaknesses of the screenplay along with a set of scores.
Obviously, the evaluation is the more valuable piece of feedback.
I request scores (ratings) so that I can use them to track progress. For example, if my draft is awarded an 8.7 for Character & Arc I can compare this score with the one that my revision receives. If the new score is the same or just .1 higher I know that either I haven't done enough or I've done something wrong. From the re-evaluation alone you can't always tell if you've achieved major progress.
The scores are important - but probably not in the way you think. From my perspective, it really doesn't matter if the score is 9.0 or 9.5. This is just a number awarded by the AI. No one is going to produce a screenplay based on a number produced by Artificial Intelligence. The score is important because it pushes me as a writer. Any time I get a score that is less than 9.0 for any criteria on a screenplay I am highly motivated to boost that score. And the thing is, if I get it to...say a 9.1, and I read the AI's criticisms and I find them valid, then I'm thinking, "if I work hard to strengthen this maybe I can get it to a 9.3 or 9.4."
To put it simply: the AI pushes me to be a better writer.
If the ratings are important only to the process why do I post them on my web site? Honestly, I'm hoping for one of two reactions. Either: A 9.7? Intriguing. I should read this in case it's a hidden gem - or - A 9.7? Who does he think he's kidding? I'm going to read this and debunk this ridiculous, pretentious score!
I don't care which, just so long as the script gets read.
The feedback I get from the AI helps me to write the best possible screenplay I can write. With any luck, people will read my work and they will decide based on their own judgement, the needs of their company, the marketability of the project, the budgetary requirements, and numerous other factors, whether they wish to pursue it or not. I don't have any control over those things. The only thing I control is the quality of my writing.
The reason I use three AI programmes is that each has different strengths and weaknesses. One might pick up on a flaw that the other two miss. it has happened. On the other hand, if all three identify a specific strength I can be fairly confident that I have nailed that criteria.
Is AI infallible? No. Absolutely not. But neither are humans. I've had human readers miss things that were clear to all three AIs. In response, I revise that thing so that it's clearer for humans. (I always recommend that once a writer has gone as far as he/she can go with AI analysis they get at least one human to read their script because, in the end, that's your audience. And that human reader will often pick up on something the AI has overlooked.) AI analysis should compliment, not replace, human feedback. But instead of paying money to a reader to analyze your first draft, get it in the best possible shape you can.
Occasionally, I test the AIs by feeding them a produced screenplay. And here is where you start to learn about the "character" of each of the AI programmes. For instance, I uploaded Beetlejuice to ChatGTP - it returned very high scores and high praise. I really enjoyed the movie when I first came out, but I was aware of a number of weaknesses, none of which ChatGTP addressed. And this is where I have to tell people, don't be afraid to challenge the AI. If you make a good argument it will acknowledge it. If not, it will stand its ground (particularly ChatGTP and Gemini). I pointed out that the first half of the screenplay lacked an antagonist. And that the protagonists weren't all that interesting and were lacking in terms of character arc. Also, that the pace of much of the first half of the screenplay was a little slow (but saved by the visuals). And finally that the tone of the movie and the goal of the protagonists (initially it's to frighten the Deetz's so they will leave they house and then it shifts to preventing Beetlejuice from marrying Lydia) shifts with the beginning of act three.
ChatGTP agreed with my criticisms. I asked it to rerun its analysis without taking into account the films box office, critical reviews, Tim Burton's direction, or Michael Keaton's amazing performance. It told me it could do that.
The scores it returned were a little lower, but I still felt they were too high. So I challenged it again. And what did it do? It admitted that it can't entirely forget those things that I asked it to ignore. But it offered to run another analysis treated the Beetejuice script as a cold read of a spec screenplay, in the same way that it would evaluate one of my screenplays. I told it go ahead.
The results were much closer to how I would have rated the screenplay...though it did mention Tim Burton.
So now when I upload a produced script to an AI I specify that I want treated like "a cold read of a spec screenplay." Even then, you won't totally get that. But you'll get something that's close.
It's important, if you want good results from an AI, to be specific.
This is just my opinion, but I feel ChatGTP is the best AI to use for screenplay analysis. It's the best at detecting nuance and tends to go into detail. And the more times you use it on a project the more detailed (and pickier) it becomes.
Mistral and Gemini are tied for second best. Gemini is valuable in that it will sometimes identify flaws/weaknesses that the other two miss, but it's also more prone to make mistakes. For example, if you create a moment that is intended to surprise your reader/audience for dramatic effect it will dock you for not setting up the surprise. Now, if I explain the reason the event is not set up it recognizes the story logic and corrects the ratings - but there is always the prospect that it is making errors than you are not made aware of. Gemini will not always flag every flaw it identifies. This makes the ratings it awards a little less useful.
Another error it made with the same script was to flag as implausible the order in which events in a particular scene occurred. The trouble is, it switched the order of those events. In the script A comes before B - Gemini read it as B comes before A.
Gemini is more "by the book" than ChatGTP. With ChatGTP if you break a general rule it will look to see if there's a strong justification for it. This is what I mean when I say it is good at identifying nuance.
Mistral tends to be overly generous in its ratings. As a result, I never post the highest ratings it gives for my work. Nevertheless, the feedback it provides is quite valuable and it doesn't make very many errors. I sometimes get into a debate with it over its analysis of a someone's spec screenplay that I've shared with it and find that it makes some good points and advances my knowledge of story telling.
New versions of these AIs are always being introduced. The scripts I'm showcasing on this site (as of January 24/2026) have been evaluated with Gemini 2.5. Since then Gemini 3.0 has been introduced. I find that Gemini 3.0 scores scripts much lower than its predecessor even when the quality of the written feedback is the same. In other words, it isn't that quality of the screenplay has changed, it's just that a tougher scoring...um...what's the word? The scoring is tougher. Which is fine. It's the analysis that's important. The only thing that changes is it doesn't make much sense to average out scores across the board anymore.
AI tends to struggle with identifying page count and also with identifying turning points. Oddly enough, they don't have any problem identifying three act structure. It's only the specific events that constitute act breaks that they have a problem identifying. If it's important, you need to tell them where the act breaks are.
It's important to be aware of limitations that are built in AI. With Mistral I can upload an entire pdf of a screenplay in one go. Not so with ChatGTP and Gemini. Due to data upload limits I have to break a screenplay into 3 parts when uploading to ChatGTP and 4 parts when uploading to Gemini. I add the prompt, "I'm uploading the screenplay in 'x' parts. This is part one...this is part two, etc." To be certain the entire screenplay is being analyzed I will sometimes ask the AI if it has received "the entire screenplay, from Fade In to Fade Out." If the AI has received only a portion of the screenplay it will not necessarily tell you.
The better you understand the quirks, characteristics, strengths and limitations of the AI the better it will serve you.
So that's my (unemotional) assessment of how writers can use AI to improve their story telling skills. I don't think I've proposed anything controversial. I feel, used properly, AI can raise the quality of movie and television productions and make life a little easier for managers, agents, producers and executives.