The Prolific survey platform [22] was used to recruit individuals assigned female at birth who lived in the United States for a study in which they were informed that they would be shown 2 different versions of online health content and asked which they preferred. In order to ensure respondents are genuine, potential participants must provide a photo taken with a state or federal ID when registering for the platform. The site validates IP addresses to prevent duplication, uses machine learning to screen for potential automation, and allows studies to withhold payment for suspicious responses, including those completed substantially faster than the average time for completion.

The Prolific quota function was used to recruit a population that was approximately half LGBTQ+-identified and equally divided among conservative, moderate, and liberal political orientations based on the site’s built-in demographic questionnaires. Individuals were also asked to self-identify within the survey, and those identifications were used in the analysis. Because of this, the actual sample demographics were not identical to the planned quota sample, reflecting the fact that individuals may have completed initial screeners for the Prolific platform up to several years prior to any given survey participation and some of their answers may have changed.

Individuals were asked to review 2 versions of 2 sets of patient-educational materials that were originally designed for adolescents assigned female at birth—the first on polycystic ovarian syndrome and the second on birth control. Within each set, 1 document was made inclusive, while the other used gender-specific language. Two study team members reviewed language differences in the documents to ensure consensus on language choice. The documents were otherwise identical, except that one (redundant) figure in the inclusive birth control document was removed due to the use of images of pregnant women in the figure (changes are listed in Table 1). Individuals were randomized using an allocation table generated in Excel, so that half received the inclusive document first in each set, while the other half received the explicitly gendered material first.

After reading each document, individuals were asked to rate the content on a 5-point Likert scale (−2 to +2) across the following domains: easy to understand, reliable, biased, meant for people like me, and relevant to people like me. These scores were then summed to provide an overall score for each document, with the bias question reverse scored. Individuals were also allowed to comment on each rating. After reviewing each pair of documents, individuals were asked if they had a version preference and to explain their reasoning. A separate comparison score was generated by subtracting individuals’ gender-inclusive score from the gender-specific score, so that higher comparison scores were associated with stronger implicit preferences for the gender-specific version.

Finally, after reviewing both sets of informational material, individuals completed a demographic survey that concluded with a question about whether sex education should be required as part of public K-12 education. They were then presented with a debrief that explained the purpose of the study, which included asking whether they had noticed that they were being presented with gender-inclusive material and if they preferred material to be gender-specific or gender-inclusive (explicit preference). Individuals were also given the opportunity to provide their closing thoughts. The questions are provided as a supplementary document. Individuals were not able to return to previous sections of the study or revise their answers to questions. All materials were accessed via Research Electronic Data Capture [23,24].

This study was approved by the Boston Children’s Hospital IRB (IRB-P00047950).

In total, 494 individuals participated in the study, 227 in the version that received the gender-inclusive documents first and 227 in the version that received the gender-specific documents first. There were 40 people who started the study but did not complete it, resulting in a response or completion rate of 92% (n=454). The study flow is outlined in Figures 1 and 2.

Participants were 41% (n=187) liberal, 24% (n=110) moderate, and 35% (n=157) conservative; 39% (n=176) were LGBTQ+, with 18% (n=32) identifying as transgender and/or nonbinary. These numbers do not match the intended quotas because individuals’ self-identification in terms of politics and/or sexual orientation and gender identity may have changed since they filled out the screeners on the survey platform. (Table 2; Figure 1)

After individuals were debriefed, and the purpose of the study was explained, they were asked, “When thinking about the health information that you read online or get from a health care provider, would you like it to” and then given the choices of “always be gender-specific (eg, women’s health, pregnant women, mother, etc)” (n=184, 40.5%), “sometimes be gender-specific” (n=59, 13%), “no preference either way” (n=131, 28.8%), “sometimes be degendered (eg, reproductive and sexual health, pregnant people, parent, etc)” (n=39, 8.6%), or “always be degendered” (n=41, 9%). We refer to this as their stated or explicit preference for the remainder of the analysis.

Individuals who stated they preferred inclusive language tended to be younger (χ²₁₆=28.9; P=.02) and more liberal (χ²¹₁₆=154.3; P<.001) and also considered religion less important (χ²₁₆=96.8; P<.001) than those who preferred gender-specific language. They were also more likely to be LGBT (χ²₈=27.7; P<.001) and specifically transgender or nonbinary (χ²₈=28.3; P<.001). Most strikingly, the differences in preference were strongly associated with differences in perspectives on sex education in public schools, which was used as a proxy for people’s overall beliefs on reproductive health education (χ² ₁₆=62.4; P<.001). Every person who preferred inclusive language supported comprehensive sex education in public schools, while no one who thought sex education should not be allowed in public schools had an explicit preference for inclusive language.

Of the people who saw the inclusive versions first, 56.9% (n=132) stated they did not notice a difference between the documents in the first pair, and 41.4% (n=94) stated they did not notice a difference between the documents in the second pair. These numbers were 61.6% (n=141) and 42.2% (n=96) for those who saw the gender-specific versions first. After the debrief, when asked explicitly about when they noticed that some of the documents were inclusive, 57.7% (n=262) stated that they did not notice until the debrief, 21.1% (n=96) stated that they noticed after the first pair, and 21.1% (n=96) stated that they noticed after the second pair. This did not differ by version (χ²₂=0.6; P=.75). Individuals with the strongest preferences (ie, always) tended to notice sooner than those with weaker (ie, sometimes) or no preference (χ²₈=56.3; P<.001; Table 3).

Looking just at individuals who had a stated preference for gender-specific or gender-inclusive documents (ie, excluding those who chose “no preference”), the majority of individuals who had a stated preference for gender-specific health education documents did not choose the gender-specific document as their preferred version for either the first or second pair of documents, although the number of individuals with a concordant preference did increase for the second document pair. The individuals who preferred content to always be inclusive were substantially more likely to choose the concordant version of their document than those from other stated preference groups (n=20, 48.8% for the first pair; n=24, 58.5% for the second pair). This pattern was also evident in the preference scores generated by questions asked after each individual document, where the only group with large differences in rating scores between the paired versions was those who wanted content to always be inclusive (Table 3, Figure 3).

We hypothesized during study design and data collection that most participants who preferred gender-specific language would not notice inclusive language, even when they wanted the language to always be gender-specific. While this was true in the quantitative responses, analysis revealed that some people felt strongly about minor differences between the documents. This manifested qualitatively in the free-text responses, which were categorized using inductive content analysis (Multimedia Appendix 1).

When reviewing the comments on why individuals preferred their chosen versions, the category with the greatest number of comments was “perception versus reality,” dominated by those who appeared to perceive substantially more differences than were actually present between document versions (n=82), which they used to justify their preferences (ie, citing things such as different layouts, flow, and comprehensiveness). That category also included individuals commenting during the debrief, usually with surprise, that they had not noticed the difference between the pairs of documents (n=21). The next most common category of comment was those related to discourse including comments mentioning inclusiveness (n=33), bioessentialism (n=26), and the use of specific language from media tropes (n=11). Finally, there were groups of comments evaluating specific types of language (gender-inclusive, n=27; gender-specific, n=14; problems with inclusive, n=12; pros and cons of both, n=8), personal preferences for gender-specific (n=14) versus gender-inclusive (n=13) content, and comments mentioning identity or writing quality (Multimedia Appendix 2).

Confirming our primary hypothesis, the majority of study participants who were opposed to the use of inclusive language either did not notice a difference between gender-specific and gender-inclusive materials or preferred the materials with inclusive language. Those individuals who had a strong preference for inclusive information were significantly more likely to both notice the differences and choose the concordant document than those with other preferences. Nonetheless, it is important to acknowledge that half of that group still did not notice a difference between the materials when the first pair was presented to them.

The study results also confirmed our secondary hypothesis. Individuals had stronger explicit or stated preferences about gender-inclusive language than were represented by their actual ratings and choices of different types of content. In other words, it appeared that people had opinions about this issue even when the presence of inclusive language did not affect them in practice. This was demonstrated by the fact that what people stated as their explicit preference for gender-specific information did not actually predict the content they said they preferred, but it did predict their political views.

Our data suggest that individuals may hold beliefs about the concept of inclusion that differ from the reality of how they experience it. For some people, this may reflect their having been influenced by media and other forms of social polarization to be upset about the concept of inclusion rather than the reality of it. It is also possible that outside of the echo chamber of highly polarizing virtual communities, the importance or performance of highly gendered and explicitly anti-inclusive preferences is reduced. The act of engaging with potentially applicable health content about the real self may support different attention to that content than the virtual self who may be more engaged with anti-inclusion rhetoric [27]. That said, there were a minority of participants who did appear to be primed for identifying inclusive content, some of whom reproduced common, inaccurate comments about gender and sex in their open-ended responses.

The fact that most people did not appear to perceive a difference between gender-inclusive and gender-specific content when they had not been primed to look for it suggests that discourse in the United States about the potential harms of inclusive language, often amplified by self-described ultra-conservatives or right-wing groups, does not reflect the lived realities of most Americans. Hyper-politicization of diversity, equity, and inclusion (DEI) can lead to discussions of these concepts in the context of health care inducing stronger emotional reactions than their actual implementation would [21,28,29]. Our research suggests that, in the coming months and years, it may be important to universally adopt inclusive language as a change in nomenclature that promotes accuracy and does not have significant bearing on the way in which critical health information is perceived by the majority of patients, rather than as an explicitly DEI approach.

In the future, we hope to analyze the use of inclusive language in real-world situations and in populations of younger patients and their caregivers. Our intention is to decouple the perception of inclusive language from DEI efforts and instead focus on it as a way of accurately and nonjudgmentally describing the bodies of individuals receiving care—similar to the use of accurate anatomic terminology. We also intend to test whether inclusive language is received differently when used in combination with gender-specific language versus on its own (eg, women and other people with ovaries vs people with ovaries). We hypothesize that combination language will have broader acceptability but possibly also lead to stronger negative reactions in a small minority of people at both ends of the political spectrum—who could become primed to notice and respond to the presence of inclusive language. Our next step in doing so will involve focus groups with a diverse range of people—exploring their reactions to these research results to understand their interpretation and the implications for dissemination. We would also like to use this method to examine people’s responses to changing sex and gender language in information unrelated to health—for example, information on the benefits of sports participation for people of all sexes versus all genders. We hypothesize that subject matter, salience, and valence may all play a role in an individual’s response.

The strengths of this study include a robust study design and a large, diverse respondent sample. The sample also clearly engaged with the research. This could be seen in the extensive comments across open-ended fields.

The main limitation of this study is that it is impossible to rule out rendering differences that may have contributed to the perception of some users that there were substantial differences between the 2 sets of virtually identical documents. The open-ended responses both confirmed and disconfirmed this. Some participants reported that they had believed the 2 documents in each set to be largely identical. Others seemed to believe that the content was completely different, describing different formatting, different lengths, 1 version being much easier to understand, and so forth.

Because the perception of difference among some users was so substantial, care was taken to replicate the study on multiple machines, varying device types (eg, phone, computer), operating systems, and browsers. Across systems, the only identifiable differences between the materials were the minor differences detailed in Table 1.

The study was also, potentially, limited by the sociopolitical environment in which it took place. Participants completed the study in late December 2024, after the election of Donald Trump but before his inauguration. This may have contributed to experiences of polarization among participants. The study was also specifically designed to capture a politically broad range of the US population and cannot be extrapolated to populations outside the United States.

Because of the sampling choice to purposively recruit for political and LGBTQ+ diversity over other forms of diversity, the sample is not an accurate representation of the full diversity of the United States. For example, looking at the racial and ethnic makeup of the United States, the proportions of the White (67% vs 59%), Black (20% vs 13%), and multiracial (4% vs 2%) populations were higher in this sample than reported in 2022 Census data, while the Hispanic (4% vs 19%) and Asian (2% vs 6%) populations were lower, and that of American Indian or Alaskan Native population (0.8% vs 0.7%) were similar [30].

Despite substantial conservative discourse about the ways in which gender-inclusive sexual and reproductive information is offensive and harmful to women, the majority of individuals assigned female at birth do not appear to notice when such content has been made inclusive—even when they state an explicit preference for gender-specific content. The exception is those individuals for whom inclusive content is explicitly relevant or preferred, who both notice and choose content that is concordant with their preferences at much higher rates than those who prefer content to be gender-specific. This suggests that, by and large, inclusive health education is broadly useful and acceptable.