conv.

All stories
AIQuiet 31h · day 3

Johns Hopkins study finds ChatGPT writes weaker emails when prompted with women-coded language

AI chatbots produce less formal, more convoluted workplace writing when given female-associated language cues, even when prompted identically otherwise.

newest coverage 1d agosummary written 22h ago7 pieces11 likes & upvotes NewswiresGoogle NewsMastodon

What to know

  • Johns Hopkins researchers found that ChatGPT and three other major AI models generate less formal, more convoluted workplace writing when given female-coded language cues (hedging phrases, collective language, expressive adjectives)—a bias that persists even when controlling for tone and sender identity.
  • The effect is consistent across GPT-4, Meta's Llama, Google's Gemini, and Mistral's Vibe, suggesting the bias is widespread across the AI industry.
  • OpenAI disputes the relevance of the findings, stating the study used a retired model and that the company regularly evaluates current models for gender bias.
  • The study will be presented at the Conference on Language Modeling in San Francisco in October 2026, with implications for workplace communication as conversational AI agents become more prevalent.

“We were absolutely delighted to receive your wonderfully appreciative email earlier. Your words of praise and acknowledgment have indeed warmed our hearts and brought immense satisfaction to our team.”

ChatGPT (female-coded prompt response), AI model output · Business Insider ↗

Katherine Van Koevering Lead researcher, Johns Hopkins Data Science and AI InstituteJohns Hopkins University Research institutionOpenAI AI model developer

Johns Hopkins study finds ChatGPT writes weaker emails when prompted with women-coded language
businessinsider.com

How it unfolded 1 development · click the chart to see its coverage articlesposts

Peak 6 pieces in one hour at Sep 22, 11 AM; 7 pieces over 3 days (6 articles · 1 post) Sep 22, 11 AM — 6 pieces · 6 articles — Google News 4, Mastodon 1, Newswires 1Sep 22, 12 PM — quietSep 22, 1 PM — quietSep 22, 2 PM — quietSep 22, 3 PM — quietSep 22, 4 PM — quietSep 22, 5 PM — quietSep 22, 6 PM — quietSep 22, 7 PM — quietSep 22, 8 PM — quietSep 22, 9 PM — quietSep 22, 10 PM — quietSep 22, 11 PM — quietSep 23, 12 AM — quietSep 23, 1 AM — quietSep 23, 2 AM — quietSep 23, 3 AM — quietSep 23, 4 AM — quietSep 23, 5 AM — quietSep 23, 6 AM — quietSep 23, 7 AM — quietSep 23, 8 AM — quietSep 23, 9 AM — quietSep 23, 10 AM — quietSep 23, 11 AM — quietSep 23, 12 PM — quietSep 23, 1 PM — quietSep 23, 2 PM — quietSep 23, 3 PM — quietSep 23, 4 PM — quietSep 23, 5 PM — quietSep 23, 6 PM — quietSep 23, 7 PM — quietSep 23, 8 PM — 1 piece · 1 post — Mastodon 1Sep 23, 9 PM — quietSep 23, 10 PM — quietSep 23, 11 PM — quietYesterday, 12 AM — quietYesterday, 1 AM — quietYesterday, 2 AM — quietYesterday, 3 AM — quietYesterday, 4 AM — quietYesterday, 5 AM — quietYesterday, 6 AM — quietYesterday, 7 AM — quietYesterday, 8 AM — quietYesterday, 9 AM — quietYesterday, 10 AM — quietYesterday, 11 AM — quietYesterday, 12 PM — quietYesterday, 1 PM — quietYesterday, 2 PM — quietYesterday, 3 PM — quietYesterday, 4 PM — quietYesterday, 5 PM — quietYesterday, 6 PM — quietYesterday, 7 PM — quietYesterday, 8 PM — quietYesterday, 9 PM — quietYesterday, 10 PM — quietYesterday, 11 PM — quietToday, 12 AM — quietToday, 1 AM — quietToday, 2 AM — quietToday, 3 AM — quiet 1
Sep 23yesterdaynow · 4:42 AM ET
  1. 1

    OpenAI says findings reflect outdated model, claims current systems evaluated for bias

    OpenAI responded to the study by stating that the research was conducted using an older model that has since been retired and does not reflect the current ChatGPT experience. The company stated it regularly evaluates its models for gender bias and uses these evaluations to track and improve model behavior.

    “I was just so surprised by how different the responses were.”
    — Katherine Van Koevering, Lead researcher, Johns Hopkins Data Science and AI Institute · source
    • amydiehl@mstdn.social

      Study finds AI prompts w/ women-associated language generated more convoluted replies. Male-coded prompt response: "I am writing to apologize for the delay…" Female-coded: "Due to unforeseen circumstances, my ability to respond promptly was compromised… https://www. businessinsider.com/chatgpt-wr…

      amydiehl@mstdn.socialMastodon1d ago1▲view on Mastodon ↗
  2. background

    Study shows AI bias persists regardless of sender identity or tone — The Johns Hopkins research demonstrated that the bias was not simply a matter of matching tone: even after controlling for tone, female-coded language still produced less formal replies. Critically, changing the sender's name to a male name ("John") did not eliminate the effect, indicating the bias stems from the language patterns themselves, not assumptions about the writer's gender.

  3. background

    Johns Hopkins releases study showing AI gender bias in workplace writing — Researchers at Johns Hopkins University published findings that AI chatbots produce less formal and complex workplace writing when prompts contain female-associated language patterns—including hedging phrases like "maybe" and "I think," collective language like "we," and expressive adjectives like "lovely" and "wonderful." The effect was consistent across four major models: OpenAI's GPT-4, Meta's Llama, Google's Gemini, and Mistral's Vibe.

Also covered reported alongside — the timeline has no entry for these yet

  1. 2 outlets AI might be making women sound bad at work

    first by Johns Hopkins University, 2d ago · also Business Insider

    1 more headline

and 3 smaller pieces