Get Quartz in your inbox

Free daily briefing on global business news.

10 GPT words Max Planck researchers say are spreading into spoken English

Delve, realm, and meticulous are among the ChatGPT-favored words a Max Planck preprint links to rising use in spoken English.

Listen to a podcast host or a conference speaker with fresh ears, and a phrase here or there might sound as though a chatbot drafted it. ChatGPT has worked its way into daily routines as a writing aid, a search engine, and a chatty companion. The shift has changed how people address one another and their devices, with results both good and bad. Vocabulary feels personal. A single phrase can mark a speaker as careful, casual, or eager to impress, and listeners pick up on those signals without pausing. A chatbot joining the circle of voices that guide how someone talks raises a fair question about whose patterns a speaker carries.

Much of the public argument so far has centered on text. Online debates have flared over whether certain cues, such as the em dash, expose material a machine produced. Earlier studies had found that ChatGPT altered how humans compose prose, which gave those suspicions some footing. Speech poses a harder puzzle. It moves faster than text, leaves no draft behind, and skips the editor. If chatbot phrasing has migrated from the page into spoken English, the familiar tells lose their value as evidence. A person might produce them out of reflex. Anyone trying to sort human expression from machine output would then face a murkier task.

A preprint from researchers at the Max Planck Institute for Human Development in Germany set out to measure whether the influence reaches people's mouths, and Gizmodo covered the work in a July 14, 2025, report by Gayoung Lee. The team identified GPT words, the terms ChatGPT tends to favor, and tallied how often they turned up in more than 360,000 YouTube videos and 771,000 podcast episodes from before and after ChatGPT's release. The researchers adjusted those counts for synonyms and scripted content. Their analysis drew on four models: GPT-4, GPT-3.5-turbo, GPT-4-turbo, and GPT-4o. The paper sits on the arXiv preprint server and had not gone through peer review at the time of the report. Levin Brinkmann, one of the co-authors, has discussed the work publicly. The 10 words below appear among the examples the report names, a sample of the wider lexicon the team tracks.

Delve represents the visible tip of an iceberg in spoken English, Max Planck co-author Levin Brinkmann told the Verge, with a broader shift lying beneath it. The researchers held up delve as one of three illustrations of GPT words, setting it beside realm and meticulous. Its place in that trio makes it the clearest doorway into the vocabulary the team set out to follow. Delve is also the term Brinkmann chose to signal how much more lies underneath.

Brinkmann's metaphor speaks to scale. An iceberg keeps most of its mass below the waterline, and his comparison suggests the chatbot's pull on speech reaches past the one term people have learned to notice. A listener who catches delve in a lecture or podcast episode may miss the quieter words traveling with it. The image redirects attention to those companions, terms that draw no comment because no one listens for them.

Delve sets a trap for anyone hunting signs of AI. Authors of a study on ChatGPT-generated text in medical publications conceded that the term does not belong to ChatGPT alone. They acknowledged that some manuscripts containing delve may have landed in the ChatGPT-generated group by mistake. The possible misreadings involved published research, evidence that human authors hold a claim on the word that a simple tally cannot detect. A speaker who says delve today could be echoing a chatbot or drawing on a human habit, and nothing in the word reveals which.

A word that people and ChatGPT both favor makes a shaky fingerprint by itself. The iceberg remark points toward a sturdier approach. One familiar term can mislead detectors and listeners alike, while a cluster of lesser-noticed words rising together tells a harder-to-dismiss story. Following the submerged vocabulary offers a steadier gauge of how deep ChatGPT's habits run in talk.

Realm belongs to the set of ChatGPT-favored terms that Max Planck researchers say grew more common in human speech across the 18 months after the chatbot's release. The report calls the rise noticeable, though it attaches no number to that description. A year and a half gave the chatbot's preferred phrasing time to travel from screens into conversation, and realm counts among the words the team ties to that drift.

The study's coverage names realm as one of three examples of GPT words, alongside delve and meticulous. Co-author Levin Brinkmann offered a longer roster of terms people use often, running from underscore to groundbreak, and realm is missing from it. Among the 10 words in this article, realm and delve alone sit outside Brinkmann's list. Nothing in the article explains the omission. The gap alone cannot show whether realm gained less ground than its peers. Realm's standing as a GPT word rests on its single appearance as an example.

Delve drew a direct comment from Brinkmann about its place in the wider shift. Realm receives no remark of its........

© Quartz