Human adoption of LLM vocabulary in speech and writing
Parent: Human adoption of LLM vocabulary and its effect on tic baselines · Published reference · snapshot 2026-09-23
↓ Facts as markdownall context files
Source-anchored research for Human adoption of LLM vocabulary in speech and writing. Run status: COMPLETED.
These notes link each claim to its source. Some citations identify archived research reports that are no longer hosted on this site. A published reference means the content is available; it does not certify independent review or accuracy.Read the editorial policy and follow the sources before relying on a claim.
How it works
- A preregistered experiment (496 participants) found brief chatbot interaction makes participants adopt specific words, persisting through distractor tasks, supporting exposure-to-active-vocabulary as the mechanism. (low confidence) [source]
Comparisons and alternatives
- Florida State University (Juzek et al.) analysed 22.1 million words of science/tech conversational podcasts; nearly three-quarters of target AI buzzwords rose after ChatGPT, some more than doubling. 'Underscore' rose while its synonym 'accentuate' did not, which points to AI-specific rather than generic drift. (medium confidence) (verified as of 2025-08-26) [source]
Facts and statements
- Yakura et al. (Max Planck) analysed about 737,083 hours of unscripted conversation from 824,634 podcast episodes and report that ChatGPT-favoured words (delve, showcase, boast, meticulous) rose abruptly in spontaneous speech after ChatGPT's release; they use synthetic-control analysis to argue for a causal link. (high confidence) (verified as of 2026-07-16) [source]
- The 'GPT words' list was derived by having ChatGPT edit millions of emails, essays and articles and extracting words it repeatedly added; usage was then tracked in about 360,000 YouTube academic talks and 771,000 podcast episodes, with a surge within roughly 18 months of release. (medium confidence) [source]
- The measured effect is lexical only; the Max Planck paper does not test semantic or grammatical influence and frames vocabulary as an early indicator of wider cultural influence. (medium confidence) [source]
Children
- No children recorded.