Now that LLM-generated prose is everywhere, human beings are eager for ways to sniff it out. While early tells like em-dashes and “delve” are long gone, researchers say there are still plenty of telltale habits that AI models fall back on when writing prose.

A new study from the marketing firm Graphite looked at the writing habits of frontier models, sussing out each model’s favorite words and phrases. While old tells like em-dash use have been stamped out, models still fall back on contrast-heavy constructions, with each model version showing its own unique quirks. The biggest surprise is how broad the scope of tells turns out to be. Graphite found 13,000 phrases that were at least twice as common in the AI content as human content — their definition of a “tell.” 

“It turns out that Claude models are actually getting closer to the human word distribution over time,” Graphite’s chief AI officer Greg Druck told TechCrunch. “And for the GPT models, it’s getting further away.”

Studying AI-generated writing at scale required a careful study design. Graphite started with a corpus of 10,000 articles published before the release of ChatGPT, serving as the human-generated control group. Then researchers had different AI models rewrite the articles from summaries, hoping to eliminate as much source bias as possible. With matching samples from both humans and each model, they could compare how often certain words and phrases appeared in AI writing, as well as broader patterns in sentence construction.

According to Graphite’s results, Claude Opus 5.5’s biggest tell is the word “dependable,” which pops up 23 times more often than in human samples. While Opus 5.5 now avoids the “it’s not X, it’s Y” sentence construction, it still tends to say something “is more than an X, it’s a Y.”

Above all, Opus loves to tell you why things matter, using the phrase “this matters” 116 times more often than human writing, while “why X matters” occurs 92 times more often.

OpenAI’s Astra has a different set of tip-offs. This model loves to describe “another dimension” of whatever it’s talking about, and tends to hedge claims by saying an action “may provide” or “can provide” a particular benefit. Its biggest tell is what Graphite calls the “corrective framing,” where a topic is defined as “not simply X” or offered as an alternative, “rather than relying on X.” According to graphite’s research, those constructions were more than 100 times more common in Astra-generated prose than in human writing.

Notably, all the frontier labs seem to have responded to the idea that models overuse em-dashes. In Graphite’s samples, Opus 5.5 used the punctuation mark 99% less often than Opus 5. Astra now uses it 88 percent less than human samples, whereas Gemini 3.1 Pro has almost completely eliminated the em-dash from its writing.

But while individual tells change, Graphite says the overall number is mostly holding steady. “It’s not like the tells are decreasing,” Druck told TechCrunch. “They are managing to remove the most well-known tells, but other ones pop up. And every model version has its own.”

It’s surprising that tells are so persistent, given the labs’ focus on human-like writing styles. In the Opus 5.5 release, Anthropic boasted that the model “communicates more naturally than prior models,” saying early users “found its writing clearer and easier to follow.”

OpenAI made similar claims when releasing the GPT-6 versions of Sol and Luna, saying users could “expect to see more clarity, less jargon, [and] fewer odd turns of phrase.”

But Druck is skeptical about how much the labs can do to completely eliminate telltale construction or phrases.

“A general hypothesis I have is that the labs are less able to control some of these things than you might expect,” Druck says. “These are giant models with billions of parameters. They have some finite number of tests they can run, and things slip through.”

When Igor Dobrianskyi looks at a car dealership lot, he sees a wasted opportunity.

“Millions and millions of used cars are sitting on parking lots, depreciating and losing value,” Dobrianskyi said in a recent interview, adding that there are 76,000 dealerships in the United States. “At the same time, there are people who need a car for a few months, but the options are actually very limited and expensive.”

Dobrianskyi’s new startup, MyMonthlyCar, aims to connect both sides of that equation through an online platform that offers flexible month-to-month rentals from local dealerships. MyMonthlyCar was selected for the 2026 Startup Battlefield 200, a cohort of promising early-stage startups that have earned a spot to exhibit at this year’s TechCrunch Disrupt. This year’s TechCrunch Disrupt will be held October 13 to 15 in San Francisco.

MyMonthlyCar, which is registered in Delaware and based in Florida, was co-founded by Dobrianskyi; Kostiantyn Gitko, who is chief product officer; and CTO Vadym Zotov. All three are from Ukraine, said Dobrianskyi, who moved to the U.S. with his wife and young daughter after the Russia-Ukraine war began.

MyMonthlyCar does what its name suggests, with one twist. The startup only rents used cars on a month-to-month basis; no short-term options here. But it does give dealerships the chance to offer customers a rent-to-own option.

“So we don’t bring them only customers to rent on a monthly basis, we basically bring them the clients who potentially can buy this car as well,” he said.

MyMonthlyCar doesn’t charge dealerships to list cars on its website. Instead, MyMonthlyCar charges dealers 10% of each transaction. It also charges the customer a separate 10% fee.

The idea for MyMonthlyCar stems from Dobrianskyi’s previous experience in the industry. The founder owned a car rental company in Ukraine, but the lack of financing options there limited his ability to scale. In 2016, he launched a peer-to-peer car marketplace called SizeCar, where owners rent their cars to other drivers, much like Turo does today. SizeCar eventually spread to 40 European cities.

The startup has signed on seven dealerships to test the service and is working with an insurance broker to finalize its own insurance program, which will let customers choose among different types of coverage.

“Insurance is the key for this business, and you need to have your own insurance as a platform,” he said, noting that the No. 1 question from dealers was about insurance and liability coverage.

Despite its early-stage status, the founders have bullish projections for the startup. They plan to sign on 100 dealerships with 2,000 monthly rentals and $300,000 in revenue in the company’s first year of operation, which will kick off later this year once the insurance program launches. Over the next five years, the goal is to generate $42 million in revenue, Dobrianskyi said.

The startup has yet to raise venture capital and is currently bootstrapped. But Dobrianskyi said the plan is to raise a seed round, with the funds helping the company hire more developers to build out the platform, including an AI tool to help dealers identify which cars are best to rent out at any given time.

To check out MyMonthlyCar the other startups that are part of TechCrunch’s Battlefield competition (as well as to network with the folks funding them), join us next month at Disrupt.

Amazon Web Services released an open-source decision model inspired by TypeSafe’s Jev, with AI developers increasingly seeking intelligence that is more suited to computer automation than frontier LLMs.

Amazon’s Strands Decider 2B, released the same week OpenAI announced a similar offering, is a high-speed, low-cost way to sort between pre-decided options and deliver a measure of how confident it is in its choice. The model is fully open-sourced, available now, and small enough to run locally.

Amazon distinguished engineer Marc Brooker came up with the project after seeing Jev and trying to build his own take on such a model. The homebrew project was successful enough—it briefly reached the top spot on the Jevbench ranking for models of its size—that Amazon engineers cleaned it up and released it as an offering from their Strands Labs, an organization developing new tools and protocols for deploying AI agents.

Brooker says the need for a tool like this emerged in conversations with AWS customers, whose agentic workflows didn’t always require the capability or cost of a fully-featured LLM all the time.

“What originally piqued my interest in this class of models was that they make a perfect decider for a workflow step— ‘what is the next thing for me to do here, based on where I am?’” Brooker told TechCrunch. He said it offers customers “a workflow step that can be structured in a way that is more reliable, thanks to the confidence scores, thanks to the closed domain of answers, [and is] lower latency, potentially lower cost.”

Like other decision models, Strands Decider is built on the “torso” of an LLM, in this case Qen3.5-2B, but instead of generating text, it delivers calibrated choices. TypeSafe named their model Jev after the economist William Stanley Jevons, with hopes of invoking his theory that the falling cost of something—like computer intelligence—can, in fact, increase its demand.

The fact that dozens of similar models have been produced by researchers since TypeSafe debuted its idea shows the wide interest, but also raises the question of how valuable they can be. Brooker suggests that the challenge will be in optimizing the model’s speedy decision-making without compromising its intelligence.

“There is a very careful balance to be found where you want to push its performance on accuracy and calibration on these kinds of tasks, without degrading its performance on understanding different languages, on having the kind of knowledge it has, which is what makes it general purpose and interesting and useful,” he told TechCrunch.

Still, he doesn’t necessarily expect the frontier labs to dominate the space, especially since, with smaller markets, the cost to build something interesting is in the hundreds or thousands of dollars.

For their part, TypeSafe executives say they are keeping their heads down and improving future models.

“I get that people think it’s a gold rush, but they might be underestimating the difficulty of making the models actually smart,” CEO and founder Diogo Almeida told TechCrunch, saying that for now, he didn’t see real competition for his company emerging yet.

“The current batch seems more like ML people wanting to implement a cool architecture than a team deeply dedicated to making intelligence useful.”