Scientists have noticed that AI chatbots, much like some human writers, tend to repeat certain words too often. Now, they’re using this habit to spot when researchers secretly use AI to write academic papers.

For example, if a paper uses words like “garnered” or “burgeoning” a lot, it might raise a red flag that AI was involved, kind of like how a student might overuse fancy words to sound smarter on an essay.

According to a New York Times report, a study from Germany’s University of Tübingen found that AI chatbots often use 454 specific words, such as “encompassing” or “intricate.”

By analyzing these, researchers estimated that 13.5 to 40 percent of abstracts in biomedical journals—short summaries of research papers—might have been written entirely or partly by AI.

With about 1.5 million papers published yearly on PubMed, a database for medical research, this suggests at least 200,000 papers could involve AI, and that number might be low if authors edit the AI’s work to hide its use.

Some researchers don’t bother hiding their AI use. For instance, Arizona State University’s Subbarao Kambhampati shared an example on X where a low-quality radiology journal included a note admitting the paper was written by a chatbot.

The paper even said, “I’m very sorry, but I don’t have access to real time-information or patient-specific data as I am an AI language model.”

Other cases are harder to spot. Unless you know the phrase “regenerate response”—a ChatGPT feature that lets users ask for a rewritten answer—you might not notice it slipping into respected journals, as the blog Retraction Watch pointed out in 2013.

Another example involved a paper about millipedes that included fake references made up by AI, like a kid inventing book titles for a bibliography. The paper was pulled from one database but later popped up on another with the same fake sources.

Then there’s the case of a journal retracting a paper after readers noticed it was nonsense, including an AI-generated image of a rat with comically oversized genitals. It’s as if someone submitted a science project with a ridiculous, computer-made picture and thought no one would notice.

The issue gets trickier because some academics are now changing how they write to avoid suspicion, avoiding AI-favored words like “delve,” according to Kambhampati in the NYT.

The University of Tübingen researchers, in their Science Advances study, suggest that if these findings hold true, AI’s role in academic writing could have an “unprecedented impact on scientific writing in biomedical research, surpassing the effect of major world events such as the COVID pandemic.”

Dmitry Kobak, a coauthor of the study, finds this trend surprising. “I would think for something as important as writing an abstract of your paper,” he told the NYT, “you would not do that.” It’s like trusting a calculator to write your wedding vows—you’d expect more care for something so important.