I know, I can already hear many of you “groaning” at the very title of this post but hear me out. To be up front, I do not believe AI can replace us as writers; screenwriting is too nuanced, too layered for a soulless stack of silicon synapses to comprehend. That being said, I decided to test the concept on one of my own screenplays and compare two versions of human coverage to that of ChatGPT 5.0.
To set it up, I uploaded a full PDF version of the screenplay for the AI to “read” and analyze, instructed it to suggest edits/changes that would bring it to a more contained, lower budget feature.
What I received:
-
For Premise, Structure, Character, Dialogue, Pacing, Theme, World-Building, and Market Path, it provided a STRONG, CONSIDER, or RECOMMEND rating.
- It was almost spot on with Human Coverage on all of these, including the notes it suggested for each area.
- It created a fairly accurate one-pager.
- The Logline it created was too long and wandering.
- It understood the Premise/Concept and came up with a decent commercial hook strategy.
-
On Structure, it accurately called out some segments that “slowed” the pace down or were repetitive and made suggestions as to what to replace those with or where to “lean into” other devices or areas that would maintain the pace.
- Both AI/Human noted some parallel plot scenes that were used to note time jumps, etc., and that they did nothing to advance the story.
-
On Character, it offered some interesting suggestions on the 4 main characters that could further define them or illustrate a clearer motivation/want, etc.
- It also noted that the protag lost some steam at a certain point in the script, became a bit too passive, exactly as the Human Coverage had noted.
- They both recognized one particular scene where a minor character’s actions/words did not make sense compared to his end game.
- On Dialogue, it called out specific phrases as examples of punchy/voicey instances that worked well and noted where I should trim repetition or dialogue that was too expository.
- On Pacing, it again recognized areas that could be compressed and more visual, how to better “show” things versus talking about them.
- World-Building made a good point of reducing some of the locations down to a few that would make the project more affordable.
- For Market Path, it outlined a budget range, a visual strategy, packaging for cast and suggested types, and comps with tone and scale as comparisons.
It then gave me a bullet list of what it felt was working, and key opportunities. Amazingly enough, these suggestions closely mirrored the Human Coverage notes. Finally, it provided a 7-point Rewrite Priority List for maximum Impact with little Effort.
For this writer, I found the AI feedback amazingly accurate, a bit scary, and topped with a dollop of surprise. It still lacked the “emotional” feedback I received from the Human Coverage notes, but for little to no cost, it made sense. I think you need to be judicial in what you accept from any form of Coverage, and not take any of it as concrete changes you have to make. The script is your work, your effort, and your story; where you take it is up to you.
Is this for everyone? Nah, likely not.
But… it provided a good roadmap to further tweak and hone the project with a quick polish. AI can be “garbage in, garbage out” for some things, but if you understand how to frame your requests, give it enough information and background on what you want, it does produce remarkable results.
In my experience (with scriptreader.ai and screenplayiq.ai) the AI coverage is very good at statistics like how much of the screenplay is dialogue vs. action and is also competent at analysing conflict, but falls flat on its ass when it comes to troubleshooting problems and creating more engaging scenes and dialogue. AI coverage, I find, is also incapable of understanding nuance and subtely in subtext unless the subtext is very on-the-nose. It also feels all characters are created equal and will "penalize" you if you don't have a ten-year back story for every character who has more than five lines. It has no balance of judgement in these regards.
For me personally, unless my audience is AI bots, I don't (fully) trust an AI bot to "analyse" my script. The human eye may not be as mathematical, but it's more, well, human.
I use Chatgpt for coverage - but I found you need to upload the final draft, 20 pages at a time, for an accurate response. When I tried uploading a PDF of the whole script it did not get into the details and often missed important beats. Also, I ask for Studio Level or Producers Eye coverage with no sugarcoating. It's usually a totally different review. I would not submit AI coverage results to a producer or my manager, but it gives me a good idea of what needs to be looked at. I find it helpful. I will still get human coverage in the end, but the script is in better shape.
I've been playing around with some of the chatbots and I am impressed in most regards. I think that in another two, maybe three years, they'll provide the same level of feedback as your average paid reader. Currently, they are much better at analysis than creativity. Task them to be creative and they inevitably come back with derivative ideas. They'll often ask if I want them to rewrite a scene incorporating their suggested changes and when I say go ahead what comes back is inevitably bad, if not atrocious, dialog. They will hallucinate sometimes. They will get facts wrong. But If your goal is to write to formula they are good at keeping you on task. They are easy graders I suspect because my scripts have never been judged worse than "consider with revisions." Oh, I take that back. One time I did get a "pass." I think it was from Grok. A quick anecdote: I recently purchased script coverage from Stage 32 from an agent and former development exec who made one big mistake. She over estimated the number of pages in the script by about thirty, which ChatGPT is notorious for doing also. I'm not suggesting she submitted pure AI to me. She did not, and her notes were helpful and fairly spot on. But I think she got some help!
I'd personally rather have a person looking at my work than a machine.
Happy New Year!
Circling back on this thread as I have been looking at Greenlight Coverage, https://glcoverage.com/, as they advertise as being specifically created for providing coverage and rewrite options for screenplays. It seems to be more accurate than just using ChatGPT or other generic AI bots. You can read two of the sample coverage reports, which are decent, but I wanted to know if anyone else has experience with this one, and would the coverage supplied be something you might use when shopping the project.
"Try Greenlight Coverage for free" and then, after you share your email address, they reveal that they'll only share a portion of the coverage from your script. That's what I call a red flag.
Hey Eric,
Thanks for the reply on this one. Yes, they do allow you to submit a PDF of one of your projects, then provide limited feedback so you can decide if it is relevant to your story, etc. I don't believe there is any "red flag" in what they offer, but it makes sense that they would not give you full coverage for free. If they did, then why would anyone then subscribe? I also feel that they provide the sample coverage reports from two well known features, so you can get a solid idea of how it works, how accurate it may or may not be, etc. I was hoping someone within our SR community had given it a whirl and could give some insight on what it was like, and if they thought it was worth the $$$. For $65, you get coverage on two scripts, so that's a pretty good bang for your buck.
As I said, this is not likely for everyone, but if you are working on multiple projects, it seems it would be an inexpensive way to get feedback on those so you can judge whether things are working in plot, structure, and so on.
Not buying it, Clark. That's a scammy way to advertise. Buyer beware, as they say.
Hi all,
Hopefully, someone who has actually tried using the service can provide some insight (not conjecture) on the experience. If not, that’s cool, as I still believe The Gauntlet is the best bang for your buck.
There's quite a bit I take issue with on that Greenlight Coverage site. No explanation of who runs it, along with being "industry vetted", offering "unparalleled coverage quality", and the somewhat ambiguous "Our clients' movies have been greenlit by..." statement.
The samples seem okay. The scoring is worrisome. The coverage is more analytical than it is insightful.
Not directed at this service specifically, I want to throw a few thoughts into the mix. I use AI every day for multiple tasks and have developed my own GPT. I've also been on advisory boards for AI-based startups.
Hi CJ,
As always, your observations and experience are insightful and objective. I do agree that AI does not bring the nuance to much, yet... and the "coverage reports" it can generate are very analytical. In my attempts to utilize ChatGPT as a general tool to vet ideas, etc., if you are not careful, they can lead you astray and completely steer you off the whole meaning or theme of your script. For new writers, that could be a dangerous and depressing avenue. yet being the curious type and not afraid to try something unproven, I took the plunge and opted in for one month on Greenlight to see how it's coverage compared to scripts I had already received human coverage on, from the likes of Script Reader Pro.
I will say that although analytical and at times, providing very perplexing comments, the AI versions mimicked some feedback correctly, but most of it was not much use. Based on what it provided, I would not recommend using it. Again, as I have said before, if you have a strong project, then run it through the Gauntlet. At least you get industry pros providing honest feedback and actionable notes. In the end, I dropped $65 bucks to find the answer, and if our conversation in this forum helps other writers avoid the pitfall, it was well worth it.
As you implied in previous comments, there's really no substitute for trying these things out.
I just don't see how generative AI can provide meaningful coverage for writers, full stop. All it can do is find patterns between script elements and what sounds like feedback. Worse still, what it thinks sounds like feedback will be coming from the likes of Reddit and other open online forums, along with previous coverage documents that have zero context with the writer's specific needs. Again, I don't see where a source of craft can come from that isn't going to be either online low-tier garbage or a library of books that contradict one another. It's just too subjective and complex to replicate without creating some kind of aggregated soup of consensus, which is the very antithesis of analysing art.
AGI would be a potentially different story, but again, it would have to be trained in a way that ignores a lot of bad craft think and brave enough to pick a lane when it comes to taste.
The Gauntlet represents the reality of what it takes to get a reasonable roundup of what the actual gatekeepers into the big studios think about material. The response to it by many communities has really showed the situation we're in culturally.
I like your "technology". This is how I draw up a technical plan for a script.
You have to be careful with uploading full scripts as they will use to train the AI. You might even find an example from your script used as a reference.
AI is not intelligent. Script algorithms like any other is programmed by humans. The human mind cannot retain decades of information save for a few savants. So all the successful/blockbuster movie scripts since the 1960's are programmed into a computer with many specific parameters set to "cover" your script. Maybe quantum computers, if they are real, will eventually be able to "recognize" subjective nuance and complexities.
PS To use an example taken from a very good British series called, The Humans. All limited programing, Human Androids with 3000 pages of code were sent 15,000 pages of human behavior code and became self-aware.
Indeed, Generative Artificial Intelligence is not actually intelligent. It's just really good at mimicking intelligence.
All an LLM like ChatGPT is doing is establishing context and responding with a guess based on correlated training data. It's literally a prediction engine.
That's incredibly powerful with well-documented objective tasks.
However, it can still fall apart if the data is bad. I once asked ChatGPT what led bulbs I needed for my Jeep interior lights, and it got it wrong three times in a row. It did that because the forums it pulls that data from are full of muddled answers. Someone asks for a vanity light bulb spec for a 97-01 Jeep Cherokee, and one person mentions what worked in their 84' while someone else responds with a link to a kit they used on their 02'.
You throw in the subjective, like asking for coverage on a script, and these things are just responding with averaged-out replies that best align with the context. It isn't thinking. It's just looking for patterns between context and typical responses to that context.
Going back to the conversation of using AI as a reviewing tool versus one that you employ to write dialogue, plot, etc., I tried out the "First Pass" (https://thefirstpass.app/) evaluation program. It reads your finished script and provides a readiness score on whether the script is "ready" to pitch, what is working, what needs improvement, etc. Now, most of you who have read my other posts know I am not afraid to test these programs, as I am curious on how accurate and meaningful they can be, then share that experience here. I've tried Greenlight, ChatGPT of course, and a few others over the last year, but none of those were helpful, and in some cases, completely inaccurate and utter gobbledygook.
However, I forked out $7.99 USD, to give the full product a shot. I was able to upload (5) completed scripts, and it then generated a full report on each one. You get a readiness score (X points out of 100), a Story Snapshot with a logline, comps, projected audience demographic, and genre(s). It also creates a very short synopsis, then a Rubric Breakdown covering Concept, Structure, pacing, character, dialogue, formatting, marketability, originality and clarity; each with the (x out 10) score. It then lists what it sees as working and what needs work, a verdict of Pass, Needs Polish, etc., Finally, it sums up the 5-6 things it thinks you need to do as Next Steps.
All (5) of these scripts have received purchased human coverage (Script Reader Pro or similar), so I used those notes as my measure of comparison; Human versus Bot. I was shocked at how similar they were, and in most cases, the AI Bot called out the same points as the Human Reader had noted. The other thing I was impressed by is the actual report; simple, pretty darn accurate, and to the point. In those other AI Bot responses, you could easily tell it was more of an amalgam of Thesaurus inspired words and phrases that most humans would never say or write.
For me, this seems to be a decent tool (and cheap) to generate a guide that allows you to go back and run a polish on the project. That being said, I would not use it on a script that was a first draft, but one I felt was closer to being ready to send out to contests or to pitch. I know many of you will scoff at this, but, in my experience, this one was actually helpful and provided good, actionable points I could use. I am happy to send anyone a PDF of the actual report if you like, just shoot me a PM here on SCR. Finally, I have no connection to the product and nothing to gain, but when I find something I feel is a realistically useful tool (not something that writes my script), then it makes sense to share it with other writers.