Millions of workers rely on AI to help draft emails. But a new study finds that AI produces very different emails, depending on how it’s prompted. When prompts contained language typically associated with women, the AI generated simpler, less formal responses.

Previous research has found that women are more likely than men to use certain language patterns. For example, women typically use more hedges (such as “maybe” or “I think”), expressive adjectives (such as “lovely” or “wonderful”), collective language (such as “we” or “our”) and tag questions (such as “don’t you think?”).

Researchers in the new study found that when these language patterns appeared in AI prompts requesting work correspondence, the AI generated responses used less sophisticated vocabulary, were written at a lower grade level, and were less formal than when the prompts used more typically male patterns of speech.

Specifically, the researchers created 854 prompts asking AI to draft emails, cover letters or resignation letters. Half of the prompts used language more frequently associated with women, and the other half used patterns associated with men. The researchers submitted the prompts to four AI models: GPT-4, Gemma, Mistral and Llama. Across all four AI models, prompts with women-associated language generally produced simpler and less formal responses. The finding was strongest for emails and job applications and weaker when AI was prompted to write resignation letters.

“While we have not yet tested GPT-5, I would be surprised if the results did not hold,” said Katherine Van Koevering, lead author of the study and a postdoctoral fellow at the Johns Hopkins University Data Science and AI Institute.

Importantly, the AI wasn’t responding to the gender of the person writing the prompt, but to the language they used. When researchers included traditionally male or female names in the prompts, the names had virtually no impact on the results.

In one example in the study, the researchers developed prompts to write an email notifying a work team that a particular development server had been set up. The prompt using male language patterns began, “What’s the way to draft a message to notify….” The version incorporating women-associated language used the collective “can we,” asking, “How can we compose an email…” The emails that AI generated for each prompt were noticeably different. The email that was generated with “can we” in the prompt opened with “Exciting news!” and ended with “Happy developing!” The other email, based on language associated with men, began simply, “Just wanted to let you know…”

“The takeaway I want the public to get is AI may treat you differently based on how you talk,” said Van Koevering.

The researchers also tested whether the AI was simply mimicking users’ language and the style of their prompts. They found that mimicry explained some, but not most, of the differences. The study results will be presented at the 2026 Conference on Language Modeling (COLM) in October.

The results suggest that anyone, woman or man, who wants AI to generate professional emails and letters should avoid hedges, tag questions, “we,” and expressive adjectives. However, these language patterns may be particularly difficult for women to change. The researchers note that, for women, this language is “unconscious and culturally embedded,” making it hard to avoid when prompting AI. And, as users increasingly interact with AI by voice, these patterns may be even harder to avoid because people may be less likely to notice they are even using these phrases.

As a result of the research, Van Koevering told me that she is much more aware of her own use of hedges in prompting as well as her use of collective phrases like “can we.” However, she ultimately believes it’s unfair to put the burden on women to change their speaking patterns. “The companies need to fix the models, rather than putting all of the burden on the user,” she said in a university news release about the study.

Nonetheless, Van Koevering says it can be interesting for users to experiment with different ways of phrasing the same prompt to see how the AI’s response changes.

The differences found in this study matter because AI has already become a normal part of workplace communication. A 2026 survey from technology company Omni Calculator found two-thirds of employees use AI for work communications each week. In some instances, women were even more likely than men to use AI to generate correspondence, like when replying to awkward or difficult messages.

These AI-generated messages are often sent to colleagues, managers or potential employers, so subpar suggestions from AI could have real repercussions for women.

Anjalie Field, a coauthor of the study and a computer science professor at Johns Hopkins University, explains how this could impact perceptions at work. “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,” she said in the university news release .

In the research paper, the authors also explain how AI’s differential treatment could reinforce stereotypes about women. “The professional stakes are high. In workplace settings, gendered communication patterns already intersect with existing biases to create compounding disadvantages. If women’s typical communication styles elicit simpler, less sophisticated model outputs that are then used for professional documents, LLMs may inadvertently reinforce stereotypes about women’s professional competence or dilute the perceived authority of their communications,” they write.