Screenwriting : AI and Script Analysis by Martin Reese

I don't trust AI to give me helpful feedback on a script, Martin Reese. Human feedback always.

Unfortunately I can see it permeating the industry Maurice Vaughan. Imagine if the Studios use AI to analyze scripts and will only read scripts that have certain score. For the Studios this will be seen a way to quickly weed-out scripts rather than having a script reader in the first-go-round. How long until AI is used to greenlight a script? I've already seen one that can generate casting suggestions.

Hopefully it doesn't come to that, Martin Reese. That kind of shortcut could be disastrous for creatives (incredible scripts could get rejected because AI scores them wrong) and studios/companies (AI will give them horrible feedback that'll lead to horrible movies and shows, which will affect profits).

Companies been using AI software well before 2020; Imagine Entertainment $30,000 fellowships use AI to weed thousands of submissions.

Some of the best script analysis I've seen has been AI generated.

That is really saying something CJ Walley

I'd never used AI coverage, but after reading a spec of mine a producer friend asked if he could run it through Greenlight (A.I.) I said, "What's that?" He thought the spec, a Feature Comedy, was strong, and believed it would score well. I said sure, let's see.

All scoring aside, I was incredibly impressed by the level of comprehensive coverage analysis. To be sure, it's not perfect, and you can see little flaws which indicate the lack of that human mind. It might be a character AI gets a little bit wrong, or a suggestion that's off because of the limitations when it comes to figuring out nuances, or a double-meaning. Basically, some areas of critical thinking, or reading between the lines, differentiating specifics, will have hiccups. But in my book, these are very minor issues. And I agree with CJ Walley. The coverage was as good as -- if not better than -- any coverage I've gotten from executives, or when I've paid for coverage myself.

As far as scoring goes, the script got an 8.5 + 99 percentile + Recommend. But if it scored lower, I would most likely feel exactly the same way about the quality of the coverage.

Now I was curious to see what another coverage might look like, so the producer and I decided to give Greenlight a different script of mine in a different genre. The producer had been heavily involved in developing this second script with me at one point, but was no longer attached to it. It's a Music Biopic.

Pushing the scoring to the side, the coverage itself came back super solid across the board in terms of the actual breakdown and analysis of the screenplay. And this is a very complex, complicated, layered story, adaptation, etc. Again, it was on par, or better than, most coverage I've ever seen.

Script #2 got an 8.2 + 98 percentile + Recommend.

So, "The Machine" liked my work x2 this time out, and the Producer paid for it. That's a win-win.

Would I pay for it myself? I have no idea of the cost, but, yes, I would. I will.

The feature writer below is a 2x Blacklist writer who posted his own 2022 Blacklist script ("HIMBO") and also his AI coverage of the script, which did not score well. He wrote an article about it, which is very informative + an interesting read on many levels...

https://nofilmschool.com/greenlight-script-coverage-results

First, make certain those offering advice that it works aren't on the advisory panel of an AI-based reading service. It's a conflict of interest.

Second, avoid it. There's no comparison of any kind to human reads. It's gobbledygook. Generic word salad.

@E Langley, I love your way with words!!

Martin Reese i agree with all of you !!! however this is my point! AI can break down patterns and highlight structure yes , but it can’t replace a seasoned industry pro. AI can't beat a human in my opinion , but i do think the human that uses it & is really talented will blow you out the water lol ... in any industry not just writing scripts !! . I think it’s a tool, not a replacement—especially for something as subjective as script analysis."

My mentor says AI is like having a best friend who’s always there, never judges, and does whatever you want —but this best friend is like Einstein aha . When used right, it can be a total game-changer. my buddy makes 45k a month using it selling wedding planners on etsy !!! 45k ... how can you look at that and go man i hate it , where are the photoshop guys at !!! ??? ummm? broke living with there moms

What everyone above has said. AI is inevitable as will be society doing their best to undo the damage done by releasing AI without appropriate guard rails and safety measures (such as kill switches) in place. Happy Hump Day :-)

In the final analysis, I would say, products and services are sold on hope and fear. AI is no different. The premise: Let's hope it won't last, if it does, we are all doomed. If one can obtain an ignoramus degree at University, this would certainly be a subject to major in.

Nothing can truly compete with the insight and creativity of a gifted human mind

I do Michael David, don't I. :) Thank you.

I have received many sets of notes over the past 3 years and presumably some of them were AI generated, but the one truly bad set of notes must have been. It was for an action script with assassins having animal codenames and the feedback said: "interesting use of creatures for fight scenes" or something and I just face-palmed.

But upwards of 85% of the notes I got, AI or not, has been useful (even if frustrating at first).

I recently tried Greenlight Coverage just to see what the buzz was about. I submitted two scripts and the AI feedback was surprisingly detailed and it even provided a few coherent suggestions on how to improve troubled areas. That said... it missed some obvious problems, it confused a gay character as being straight (recommending I build the intimate relationship between the gay character and the straight protagonist) and a few other issues I knew about going in. AI is cool but not serviceable in my opinion. Maybe in a few more years. But, to be fair, I've gotten feedback from human readers where I can tell it is either a junior reader or they just skimmed the script, offering up obvious or even ridiculous suggestions and comments.

Michael Fitzer What I don't like about 90% of human Readers is their inability to allow for innovation. They were so trained on rules, rules and more rules so when they come across something unorthodox that of course can work great, their radar says no, no this and that should happen on page 4 and the protag must this and that and the antag that and the revelation must be here, etc. So, you are paying a Reader to attempt to downgrade your script to the same drivel that fills the worlds trash cans (or recycle bins). They are the Guardians against anything new, innovative, risky or interesting happening in a script.

Martin, I started using WriterDuet's Screenplay IQ program (one of Greenlight Coverage's competitors)...and I gave it a shot back in July after reading about another Stage 32 member's experience with Screenplay IQ.

I've run four of my scripts through IQ, and I like how it uncovers themes I never really envisioned when I wrote those screenplays.

Now...the bad news: Screenplay IQ missed describing a major character (Henry) in "Got Any More Bullets, Sister?" The program left out the final scene in "Pixie Dust," where that 1929 Ford Model A got auctioned off and Myriad became an official sorority at the U of Minnesota-Twin Cities. When IQ combed through "Kitten on the Keys," it described two people Kitten mentioned in the dialog (Henry and Callie)...despite the fact that the twosome didn't have any scenes, let alone dialog of their own. And in IQ's assessment of "Tin Mine," it had two of Sue Ellen's piano students (Judy and June) setting foot at Maxwell Music...but the two students never did. (They trained in Sue Ellen's apartment.)

I'm curious to see what IQ does with "Jingle Belles" and other efforts of mine...but one thing's for sure: There's nothing like a pair of human eyes and a human brain to break down a script.

Someone should run ai on the Chinatown script. I bet it has many ideas for improvement and it doesn't get the Title at all, etc. It will probably give it a "pass". lol

This is fun. So the vast majority of script readers are out of work writers (a not very writer, because all of the good writers with good attitudes are working right?) or they are not writers at all, like AI.

Seriously, look at the comps coverage offers, that is real value. If you get a comp of something that made a profit in the last five years you might be more likely to get a recommend.

AI is smart. I can't say that about all the people who do script analysis.

What Dan Guardino said.

I hear during the gold rush, the ppl who made money sold shovels.

It's the same with script service businesses.

I have a new script I want feedback on: MR. CRANKY AND THE MISANTHROPE.

Sounds like my wife's nickname for me.

What a lucky gal.

Give the AI "The Computer Wore Tennis Shoes", maybe it will be a "recommend" and we will know the damn thing is sentient.

E Langley , who is the protagonist? It sounds like they are playing "Bad Cop/Worse Cop'. lol

Maybe it's not really a script. LOL

I have run script analysis using ChatGPT on one of my scripts and it gave me very similar notes to coverage I have received from actual people. In some cases even clearer notes. How I would use AI is to give me feedback to refine my script before sending it off to get read by an actual person because each read adds up in cost and some readers give some way off feedback. I don't follow all the advice AI gives, because I don't agree with everything it recommends or finds a problem with.

I still sent it to an actual person after simply because AI isn't going to buy my script and subjectivity still matters in the decision-making process. But I find AI a handy way to get some preliminary, rough notes.

Sure not as well as a veteran, but can it get your work further along than someone without experience just viewing it for you? If you have something completely original it will likely be pretty bad But if you are following best points, maybe like other options it might get you 80 percent of the way.

Like in other for a of art I'm sure that it'll get it the point where it can provide you with enough to get your script to as mediocre as anything else. It can't understand the spark of genius that is going to make you stand out from the rest.

As I mentioned earlier, having never before used any AI-based coverage on my material, I had no pre-conceived anything. I came to it from an outside suggestion -- a producer's suggestion -- and so it truly was, "Sure, let's see." I've had plenty of great notes and reads from creative executives, and I've had my fair share of terrible notes and reads from creative executives. Notes that might make you wonder how long they've been doing it (probably not long), or why they're in that position to begin with. Just because a reader or a creative exec is "human" doesn't guarantee anything. It's like anything else -- you have individuals with different levels of talents and abilities. Strong, insightful, helpful, notes only happen when the individual has the ability and talent to convey those concepts to the writer in a clear and constructive manner. In my experiences, I can tell you, some don't do the job very well.

My brand new 2-script experience with AI-generated coverage notes, was, without question, on par with the best notes I've gotten from creative execs.

I don't have anything to do with AI. Zilch. I'm just a feature writer who wants to write the best material I possibly can. Great notes are great notes. Wherever they come from, I'll take them and use them.

AI can't truly experience emotion, so it will always have a blindspot. I don't think anyone as a writer can truly say they want their most impactful scenes reduced down to a series 0s and 1s. There's just also stuff the human mind picks up on that a system eternally operating in the past tense can't understand - new trends, new patterns, new phrases or lingo. Cultural nuance is often completely lost by AI, and let's not even get started on the inherent biases of machines programmed on historical data. I personally see zero to negative zero value in using AI for script analysis, but to each their own

I agree with the blindspots and such that you mention, Pat Alexander. I said as much in my first post. I will say, however, that the most impactful scenes of my music biopic were not reduced down to numbers at all. In fact, the coverage actually deftly articulated very gray and somewhat ambiguous areas of the story, and the central character's most impactful actions, which ultimately led to his death that was officially deemed "accidental."

These areas of the screenplay were not written with expository dialogue. Quite the contrary. Dialogue was pulled way back to a minimum, and in some instances, the minimal dialogue utilized was there to only convey the underlying sentiments, which were, at times, the opposite of the actual words being spoken. In a story where there's a lingering question about what "actually" happened, and why it happened, you want to try and present a narrative that doesn't focus on drawing one definitive conclusion -- your own -- but to showcase a narrative that allows for the audience to experience the story and come to their own thoughts and feelings about it. In the case of my screenplay, I found that the AI coverage was fully able to read between those tricky lines and articulate upon those gray areas, and capture the complexities of a biopic story which is complicated and non-linear.

And that did surprise me.

CJ Walley if you'd care to expand on your experiences with AI coverage, I'd love to hear your thoughts...

DT Houston I was asked in to look at a AI driven script analysis service that was in development. Like yourself, I was on the fence but aware of the progress in generative AI via the likes of ChatGPT and Midjourney.

I found the results staggering, like, have to get up and pace around the room staggering. There was stuff in there that just blew my mind, such as the AI identifying my own work as not just similar to Tarantino but old Tarantino. There was a lot there that I hadn't realised about my own material too, observations that were very useful from a marketing perspective as it described it from outside the box, so to speak.

Overall, the appeal of the system was limited to me. I'm not looking for feedback. I don't particularly care what any person or cpu thinks about my spec work as I'm happy with it. What it did do is actually help reinforce a lot of positive feelings I already had and build my confidence in the material further.

The released service felt very cut back from what I saw in development and I was disappointed by that. It also went down paths I didn't see as beneficial. It didn't have many of the elements I liked the most. I didn't see a great deal of use in it, but then, as mentioned, the whole feedback/coverage thing is kinda lost on me anyway.

There are many robotic elements to script analysis that AI can replicate. The issue is when you get outside of the data available to train AI on. Therefore, I see any argument for or against pretty silly unless you define the scope of analysis taking place. This is a huge limiting issue for any discussion about AI as people often throw the baby out with the bathwater.

There's been a lot of talk about AI now and I've generally found that those denouncing it have next to no experience of it. They just don't like it. It's an emotional standpoint for them. A lot consider themselves consultants too, so obviously see it as competition.

So many people are thinking in black and white instead of shades of grey. It's like all these references to "industry pros" which is a huge generalisation based on little but speculation. AI gets called nonsense yet the vast majority of people I see offering feedback and coverage have zero credentials themselves, and that's if they aren't operating anonymously.

"Overall, the appeal of the system was limited to me. I'm not looking for feedback. I don't particularly care what any person or cpu thinks about my spec work as I'm happy with it. What it did do is actually help reinforce a lot of positive feelings I already had and build my confidence in the material further."

Thank you for sharing your insights CJ Walley. Really interesting. And I echo your sentiments above.

After reading your discussion, I decided to test Greeenlight for free. Waste of time. Some members here have read my screenplay, and given me better advice.

A problem with AI is that it assumes that the most common solution is also the best solution, So it doesn't like new ideas, even though the audience's taste changes year by year.

So I plan to continue with my strategy to contact other members. "You do me a favor and I do you a favor." Both the person who writes the comments and the one who reads the comments learn by doing so.

But alright, if your story is a mess, than it is too early to ask other humans to read. In such cases I can imagine that you ask AI for help. To clean up before you ask humans with remaining problems.

I have myself noticed tha AI can be useful for checking grammar and spelling.

I've read some AI produced script analysis and it gave an ok overview of the events and inter-character story but it lacked in subtext and the significance of the events to characters and the story. Still no substitute for an experienced reader.

Hey Ewan Dunbar, That's what I couldn't articulate. I don't believe the AI can get subtext.

Martin Reese thought I'd add an update, as I put a third new spec of mine through Greenlight Coverage the other day. It's probably the most challenging feature I've ever tackled. It's a complex family drama with mental health elements, psychosis, and a young lead character who believes certain things are happening to him and his body, when in reality opposite things are happening. There's also highly- fantastical elements related to the psychosis, and very dark and tragic consequences which ultimately transpire. It's not linear. There's substantial flashbacks. And for the first 2/3 of the story, it's written in a way that the audience is experiencing the story from the lead character's POV and mind -- but the audience doesn't quite know that yet. The audience simply thinks the story being told is the lead character's reality. A strange reality. This was a really tricky script to write and pull off. Not a lot of room for error, and if you mess it up, it's toast.

Once again, I found that Greenlight was able to pinpoint and understand and articulate just about everything going on within the story. And, yes, that includes SUBTEXT. It was able to make critical subtext, story and plot distinctions, and read between the lines where actual dialogue was minimal or even non-existent. To say I was impressed would be an understatement. Of the three scripts I've now given Greenlight, this was, by far, the most complicated and difficult piece. My final 99 pages had often felt like 2x that amount because conceptually it just took a ton of work to figure out and get right.

I found a handful of specific suggestions to be extremely solid and worth considering. As CJ Walley had mentioned earlier in his post, I too have found that the AI coverage is on par with the best coverage I've seen. For the record, the scores on this spec were: 8.5, 98 percentile, Recommend. If I had scored lower, I would still be highly impressed.

I wouldn't say AI is on par with the absolute best coverage I've seen. I'm fortunate enough to have some experience at the higher end of things with readers who work for sales agents, studios, and major agencies. Those guys have a level of insight the AI models cannot be trained on. They are the true definition of "pro readers", and one of the reasons I get frustrated with that term being abused.

When I say AI has generated some of the best coverage I've seen is with the remit typically available to the amateur world. A lot of that is quite robotic and easy to learn online. AI doesn't really know about production logistics and commercial aspects, but then most people calling themselves pro-readers don't either.

I hear you CJ Walley. I'm not factoring in those type of production logistics/commercial aspects into my evaluation. Just the analysis and breakdowns of the writing, story, structure. I've seen/read very good to excellent coverage on feature writing -- not just on my own work, but on other writers' work, too -- from readers at CAA, WME, Disney, Gersh, WB, New Line, a few others. The difference between strong, insightful readers and other readers who don't quite have that level of skill quickly becomes very apparent. I'm not saying AI coverage is perfect. It's not. However, I do think, from my recent experiences with using Greenlight AI, that it offers a very cost effective and time efficient method of getting a pretty solid barometer. A temperature reading. On the work. And like any coverage, you look for things that can be useful to you and the material as you move it all forward...