Researchers at Johns Hopkins University discovered that popular AI models, including GPT-4, produce noticeably less professional and less formal responses when prompted with language patterns commonly associated with women. The finding suggests that millions of workers relying on AI to draft workplace emails may unknowingly receive subpar suggestions based on how they phrase their requests.

What Did the Study Find About AI and Gender Language?

A team of researchers created 854 prompts asking AI models to draft emails, cover letters, and resignation letters. Half of the prompts used language more frequently associated with women, while the other half used patterns associated with men. The researchers tested four AI models: GPT-4, Gemma, Mistral, and Llama.

Across all four models, prompts containing women-associated language patterns produced responses with simpler vocabulary, lower grade levels, and less formal tone. The differences were strongest for emails and job applications, though weaker for resignation letters.

The language patterns that triggered less sophisticated AI responses included hedges like "maybe" or "I think," expressive adjectives such as "lovely" or "wonderful," collective language like "we" or "our," and tag questions such as "don't you think?".

In one concrete example from the study, researchers created two prompts to write an email notifying a work team about a development server. The prompt using male-associated language began with "What's the way to draft a message to notify..." while the version with women-associated language used "How can we compose an email..." The resulting emails were strikingly different. The AI generated "Exciting news!" and "Happy developing!" for the female-pattern prompt, but simply "Just wanted to let you know..." for the male-pattern version.

Why Does This Matter for Women in the Workplace?

The stakes are significant because AI has already become routine in workplace communication. A 2026 survey from technology company Omni Calculator found that two-thirds of employees use AI for work communications each week, with women sometimes even more likely than men to use AI when replying to difficult or awkward messages.

These AI-generated messages are often sent to colleagues, managers, or potential employers, meaning that lower-quality suggestions could have real professional consequences. The concern extends beyond individual emails to broader workplace perceptions and advancement opportunities.

"If you prompt a model to write an email you're going to send to someone else at your company, and you're using language features that women more commonly use, you'll get back a response that's less complex, at a lower grade level, and less formal. That's going to reflect on how the recipient of that document perceives you," explained Anjalie Field, a coauthor of the study and a computer science professor at Johns Hopkins University.

Anjalie Field, Computer Science Professor at Johns Hopkins University

Researchers also noted that AI's differential treatment could reinforce existing stereotypes about women's professional competence. If women's typical communication styles consistently produce simpler, less sophisticated model outputs that then get used in professional documents, large language models (LLMs), which are AI systems trained on vast amounts of text data, may inadvertently reinforce biases about women's authority and expertise.

How to Adjust Your AI Prompts for More Professional Results

While the researchers emphasize that companies should fix their AI models rather than placing the burden on users, they also found that experimenting with different prompt phrasings can yield noticeably different results. Here are practical adjustments anyone can make when using AI for professional writing:

• Avoid hedging language: Replace phrases like "maybe," "I think," or "possibly" with more direct statements that convey confidence and certainty.
• Eliminate tag questions: Remove conversational additions like "don't you think?" or "wouldn't you agree?" that soften the tone of your request.
• Use individual rather than collective framing: Replace "we" or "our" with "I" or "the" when appropriate to shift the AI's response toward more formal, individual accountability.
• Remove expressive adjectives: Cut back on descriptive words like "lovely," "wonderful," or "exciting" when drafting professional correspondence.

However, researchers acknowledge that these language patterns are often unconscious and culturally embedded, particularly for women, making them difficult to change consistently. As voice-based AI interactions become more common, avoiding these patterns may become even harder since people are less likely to notice their own speech patterns in real time.

"The takeaway I want the public to get is AI may treat you differently based on how you talk," said Katherine Van Koevering, lead author of the study and a postdoctoral fellow at the Johns Hopkins University Data Science and AI Institute.

Katherine Van Koevering, Postdoctoral Fellow at Johns Hopkins University Data Science and AI Institute

What About GPT-5 and Future Models?

The study tested GPT-4 and three other models, but did not yet evaluate GPT-5, OpenAI's next-generation model. However, Van Koevering expressed skepticism that the problem would disappear in newer versions. "While we have not yet tested GPT-5, I would be surprised if the results did not hold," she noted.

Van Koevering

The researchers also tested whether AI was simply mimicking the style and tone of users' prompts. They found that mimicry explained some, but not most, of the observed differences, suggesting the issue runs deeper than surface-level pattern matching.

The full study results will be presented at the 2026 Conference on Language Modeling (COLM) in October. For now, the research underscores a growing concern in AI development: that widely used models may inadvertently encode and amplify existing gender biases, even when no explicit gender information is provided to the system.