Morris Chapdelaine always has a daunting pile of scripts on his desk. As an indie producer, he reads about three books a week, gives the rest to interns and film students, and sends back detailed reporting. But he has a hard time getting over them all.
At a film festival, some friends suggested I study artificial intelligence to reduce my workload. “I was able to branch out into anything AI-related,” he says. “Some of it scares me.”
But Chapdelaine did some research and ended up signing up with Greenlight Coverage. Greenlight Coverage uses a large-scale language model to summarize your screenplay and rate elements such as plot, character arc, pacing, and dialogue on a scale of 1 to 10. Additionally, a verdict of Pass, Consider, or Recommend is given.
He found the AI to be more honest than human feedback (even his own), but it also doubled his reading pace.
“It’s a huge time saver,” he says. “And it keeps getting better and better.”
If there’s one thing AI does well, it’s summarizing documents. Therefore, of all the jobs in your development pipeline, the first job may be the most vulnerable: the script reader. The industry’s first gatekeeper may one day be a software program.
In fact, machines are already playing a role. At WME, agents and assistants use ScriptSense, another AI platform, to categorize submissions and track client work. Aspiring screenwriters are also turning to AI tools like ScreenplayIQ and Greenlight to provide (sometimes overly flattering) feedback on their drafts.
At major studios, human story analysts are still digging through piles of submissions just as they were 100 years ago. But as AI invades everyone’s workflows, they’re starting to worry about their jobs.
Jason Hallock, a story analyst at Paramount, recalls his first disturbing experiment with ChatGPT, the bot that sparked the current AI frenzy. “How often do you change?” he wondered. “Six weeks? Or six months?”
Working with the Editors Guild, which represents about 100 unionized story analysts, he decided to investigate. Earlier this year, he began experimenting. He asked the AI tool to cover several scripts and stacked up the reports against human-generated coverage. It was a test to see if he and his colleagues could compete.
Since the dawn of Hollywood, story analysts have been the thresher, sorting the wheat from the chaff. AI proponents argue that algorithms make the process more efficient, objective, and fairer, allowing new voices to be heard rather than relying on readers who bring their own subjective preferences to the work.
But something can also be lost. Human readers are the first to sense whether a script has potential, whether the characters are interesting, whether the story captivates them and has something new to say. Can AI do that?
“The most important thing I’m looking for is, ‘Do I care?'” says Holly Sklar, a longtime story analyst at Warner Bros. “I can’t be bothered with an LLM.”
Still, AI seems to be coming regardless. So instead of ignoring it, some people are trying to understand it.
“No one wants to lose their job,” said Alegre Rodriguez, an analyst at the Editors Guild who participated in Hallock’s study. “We’re not sticking our heads in the ground and pretending it doesn’t exist, we’re not hunkering down and waiting for someone to give us a pink slip. I think people are dusting themselves off and thinking, ‘How do I stay in this game?’
Kartik Hosanagar is a business professor at Wharton and an Internet marketing entrepreneur. He is also a movie buff and has several scripts in his drawer. It’s a drama about a start-up and a thriller about a murdered Indian diplomat. As an outsider to Hollywood, he had trouble selling his scripts. Therefore, he developed an algorithm that levels the playing field by objectively evaluating talent. This adventure didn’t work out, but the next one did. Hosanagar developed ScriptSense, one of the most talked-about AI scripting platforms today. The pitch: “Rate the script 100x.”
“There is a huge unread pile,” Hosanagar says. “This is a great way to clear the pile and understand where to focus your attention.”
In March, Hosanagar sold the company to Cinelytic, a service provider that integrates ScriptSense into a suite of management tools. “It’s about saving time,” says Tobias Kwiser, the company’s CEO. “Opportunities are left out because there isn’t enough capacity to look at all the work. Unknown writers don’t have a chance because top agencies aren’t submitting scripts.”
ScriptSense provides an overview, character breakdown, structure, and casting suggestions. The tone is relatively neutral. It does not offer praise or criticism.
“Our design philosophy was that we were not going to make decisions for the user,” Hosanagar says. “You’ll never see the words ‘Great!'” or “Reject.” ”
Platforms for screenwriters have a different philosophy. Greenlight founder Jack Chan believes in the power of AI to make critical decisions. “What AI is really good at is averaging things out,” he says. “In terms of feedback, you’re trying to reach a wide audience. You want the public to like your work. That’s where AI really shines.”
ScreenplayIQ provides a qualitative assessment, but not a numerical score. Programs that summarize the plot and evaluate the “growth” and “depth” of characters help writers see their work from an outside perspective. “Our aim is to support writers if they feel they are struggling and need support,” says developer Guy Goldstein. “It’s holding a mirror up to your script. You wrote it with intent. It’s looking at whether that intent came through.”
To test the AI platform, Hallock needed a script. Screenwriters can be sensitive about feeding material to an AI model because they expect the material to be used for training. But a close friend provided me with an old script for that purpose. One was an unproduced script for the Syfy Channel about killer insects. The other was touted as the “dark heart of the universe.” The author didn’t care if the AI trained it.
“He said he wanted AI to become dumber,” Hallock said.
He collected a few others and passed them all on to human analysts. He then compared their coverage to loglines, synopses, and notes produced by six AI platforms. The results were both encouraging and disturbing.
The AI-generated loglines were indistinguishable from human loglines. Maybe it could have been a little better than that. Differences began to appear in the AI-generated summaries. “They tend to have the quality of 11th grade writing,” Hallock says. “The same kind of constructions are used, such as ‘Our story begins with…'”
The more complex the script, the more likely the AI will get things wrong. That means you’re more likely to misattribute one character’s actions to another’s or hallucinate plot points.
When it came to notes that required actual analysis rather than mere distillation, humans were a clear winner. The AI program was “almost a complete failure across the board,” Hallock said.
The script for “The Heart of Darkness” was a hot topic in Hollywood 20 years ago, but even though it did not sell, it received a “recommendation”. That was a consistent problem. According to Rodriguez, the models were “biased against the writer” rather than offering honest criticism.
“They definitely tell me all the things that are positive and going well,” she says. “But when they had to address a problem, they weren’t always able to identify the problem.”
In some cases, AI programs were not evaluated. They were cheerleading.
“He has a puppy-like quality,” Hallock said. “It wants to please you.”
A romantic comedy was praised by AI as “a compelling, well-written coming-of-age story that balances humor, heartbreak, and the bittersweet reality of navigating your 30s. Strong character development makes it a standout.”
Human readers, on the other hand, were blown away by saying, “It’s a typical Las Vegas girl template. Especially with Sidney Sweeney, it has potential for light streaming content. The language is vulgar, but the jokes don’t stick. It lacks the feel of ‘Girls’ Trip’ or ‘Bridesmaids.'”
Chan defended Greenlight’s preference, saying only 5% of scripts submitted to the platform receive “recommendations.” “That’s a very small number,” he says. “I wouldn’t say we’re seeing significant inflation.”
Hosanagar said one reason ScriptSense doesn’t make recommendations is because the AI can be too sycophantic. “Can AI reach a level of real significance?” he asks. “I think we can get there. We’re not there yet.”
Rodriguez said many analysts were encouraged by the study. AI may be faster, but it can’t pull something original and brilliant out of the pile.
“Humans will still need to look at these reports and review the material,” she says. “It doesn’t save as much time as they think.”
And those who rely too much on it may miss out on something great. However, the study is not entirely reassuring, concluding that “studios may be tempted to sacrifice quality and accuracy in favor of cheap and fast.”
AI model creators say these concerns are misplaced. “It’s not about taking away jobs,” Kwisar said. “We think of it as human enhancement.”
Avail CEO Chris Giliberti says story analysts are already using the company’s AI platform to perform routine tasks, which frees up time to tackle more difficult analytical tasks. “It’s unstoppable,” he says. “The cat is out of the bag. This is making people’s jobs and lives easier.”
But Sklar is concerned about where this is going. Today’s managers value people’s opinions. However, there may be a younger generation that is more comfortable with AI summaries. She worries that some people in Hollywood, “cost-cutting people who have no idea what we’re doing,” will come to see her role as superfluous.
“It keeps me up at night,” she says.
