Worried your writing sounds like AI? These tools can help

Programmer Simon Willison released the LLM cliché highlighter on August 28, a tool that flags 38 signs of AI writing, while George Strakhov, founder of AI automation consultancy move38, created the AI style linter Slop Tells to help writers avoid inadvertent AI tells. These tools come as LinkedIn and Substack introduce moderation efforts and detection tools like Pangram and Turnitin are used by universities, reflecting growing concern among human writers about sounding like AI.

Today, AI https://www.fastcompany.com/section/artificial-intelligence -generated writing is not just found in conversations with chatbots, but in a major chunk of all LinkedIn posts, Substack articles, and tweets https://www.fastcompany.com/91584892/linkedins-new-anti-slop-button-is-coming-to-every-platform ; in medical papers https://pmc.ncbi.nlm.nih.gov/articles/PMC12504183/ ; and even in published literature https://www.nytimes.com/2026/03/19/books/ai-fiction-shy-girl.html . You’re not imagining it—as we collectively consume more of this content, some of its quirks start to feel glaringly obvious. And honestly? In this paragraph alone, I’ve already used three idioms that might set off your AI alarm bells. Now, multiple designers are building their own tools to help other writers avoid the same mistakes. Recently, companies like LinkedIn and Substack have introduced their own moderation efforts https://www.fastcompany.com/91584892/linkedins-new-anti-slop-button-is-coming-to-every-platform to cut back on fully AI-generated text. Meanwhile, AI detection tools like Pangram and Turnitin are helping universities and organizations https://gradpilot.com/news/pangram-labs-default-ai-detector-college-admissions-2025 suss out AI-generated content in application essays and work documents. For the writers themselves, though, AI detection isn’t the problem: Increasingly, writers who use no AI at all are beginning to worry about their work containing an inadvertent AI tell. “What’s interesting is that these models do speak in a specific way, and you develop a sense of it—you develop the smell of the writing,” says George Strakhov, founder of the AI automation consultancy move38 and creator of the AI style linter Slop Tells https://sloptells.com/ . To put human writing through the AI smell test, designers like Strakhov are starting to make their own interfaces to catalog and highlight key bits of phrasing, formatting styles, and specific words that are making your prose sound like it came from a robot. Here are two that you should know about. The LLM cliché highlighter https://tools.simonwillison.net/llm-cliche-highlighter , designed by programmer Simon Willison and published to his X https://x.com/simonw/status/2093277255438860358 on August 28, is a strong first step for writers looking to identify AI jargon in their work. The tool tracks 38 different signs of AI writing, including rhetorical tics like “The punchline is,” “Here’s the twist,” and “Don’t take my word for it.” Users input the text they’d like to test into a window or copy the link to an entire article and the tool will highlight any flagged sentences in light yellow and direct AI tell matches in a darker yellow. It will also present a counter at the top of the page displaying the total number of flagged sentences, matches, and chain items patterns like “no X, no Y” that showed up throughout the text. Given its clear visual approach, the cliché highlighter is a solid choice to get an initial sense of how AI-saturated your text is reading. However, the tool’s scope is fairly limited, and it’s not designed to evolve with the changing LLM landscape. According to Strakhov, AI tells are constantly changing as chatbots evolve and people evolve with them—which is why he designed his tool, Slop Tells, to constantly reassess which phrases, words, and constructions are no longer passing the AI smell test. Back in 2024, the word “delve” was practically unavoidable when talking with ChatGPT, despite the fact that it’s a fairly uncommon word for most American English speakers. At the time, experts theorized https://www.theguardian.com/technology/2024/apr/16/techscape-ai-gadgest-humane-ai-pin-chatgpt that this pattern was due to the outsourcing of reinforcement learning for ChatGPT to human workers in Nigeria, who use “delve” much more frequently in common speech. In any case, this tendency became so glaring and offputting to users that all the major AI companies have phased it out—to the point that, Strakhov says, it’s “absent from every current model” in 2026. It’s just one example of how AI tells behave like a living, breathing organism. “This is the interesting loop: you teach your tools what they should do, and then the tools start doing that all the time, and then you develop almost an antibody response,” Strakhov says. Users don’t like that they see particular syntax patterns “over and over again,” he adds. Clearly, AI companies are aware of this cycle and are actively updating their models accordingly. So, Strakhov designed Slop Tells to keep track of the changing landscape in real time. Slop Tells, which Strakhov vibe-coded using a variety of AI agents, uses what he refers to as three different “piles” of information to inform its model. The first pile is human writing from before 2022 across different registers, including forum comments, essays, LinkedIn-style posts, and answers to cooking questions. To generate the second pile, an AI agent takes the first pile of human-written content and reverse-engineers the prompts that might’ve produced each post. Then, those prompts are fed through today’s AI models to see how a chatbot would write about the same topic. For example, Slop Tells might take a pre-2022 cookie recipe, generate the prompt “write me a recipe for cookies,” and then give that prompt to ChatGPT. The differences between that first pile of human writing and the second pile of commensurate AI writing are what generates the most up-to-date AI tells. Strakhov’s model then feeds the results through a third filter of well-regarded pre-AI authors’ work—like speeches and published literature—to ensure that tool isn’t erroneously flagging qualities of good writing, like the use of em dashes. This whole process automatically restarts every week, so that Slop Tells can suss out any changes to the major AI models. The results are striking: some oft-cited AI tells, like the em dash and the phrase “Great question ” are almost entirely out of use, while new tics, like the word “genuinely” and the phrase “what actually matters” have taken their place. Like the LLM cliché highlighter, Slop Tells will highlight AI jargon in any pasted section of text, but it will also give users a sense of whether that jargon is currently active in models or faded, as well as linking out to a rich compendium of information on each example. This tool is better for a more granular and informed look at the AI tells in your writing. “What you’re exposed to is what forms your taste,” Strakhov says. “Broadly speaking, if somebody grows up reading a whole lot of AI generated text rather than Harry Grey or whoever, they develop that as a taste. As people, our taste is very malleable, and that is the thing that I’ve been worried about for a while. Now it’s happening, and it requires a lot of conscious effort on behalf of the humans who use these tools both to write and to read to decide where they’re willing to change their own taste accordingly, or where they’re willing to defend it.”