ChatGPT Influence on Spoken Vocabulary

Key Facts

  • ChatGPT reached 100  million100\,\text{ million} users within 22 months of its late 20222022 release (fastest-growing consumer app).
  • Study analyzed >700 000700\,000 hours of YouTube & podcast speech to detect vocabulary change.
  • Researchers identified “GPT words” (e.g., delve, realm, meticulous) that ChatGPT frequently adds when asked to polish text.

Methodology

  • Step 1: Used ChatGPT to edit millions of written pages; extracted repeatedly inserted words ⇒ labeled as GPT words.
  • Step 2: Collected speech corpora
    • YouTube: 360 000360\,000 videos
    • Podcasts: 771 000771\,000 episodes
    • Samples taken from periods before and after ChatGPT release.
  • Step 3: Compared GPT-word frequency to “synthetic controls” (weighted synonyms not common in ChatGPT output).
  • Observation window: 18  months18\,\text{ months} post-launch.

Findings

  • Sharp surge in GPT-word usage across both scripted and spontaneous speech during the 1818-month window.
  • Indicates AI-to-human “cultural feedback loop”: patterns learned by LLMs return to influence human language.

Implications & Concerns

  • Rapid, large-scale diffusion of AI-generated phrasing may narrow linguistic diversity.
  • People imitate sources perceived as knowledgeable; growing reliance on AI could amplify its authority over other cultural influences.
  • Future monitoring should expand beyond individual words to sentence structure, idea framing, and broader discourse patterns.

Takeaway

  • AI’s effect on spoken language is already measurable within 2.52.5 years; key question shifts from if to how profound its cultural reshaping will be.