Artificial intelligence platforms increasingly shape public understanding of preventive healthcare. Patients frequently seek automated guidance before discussing immunization with their clinicians. However, systematic evaluations reveal significant demographic variation in AI communication patterns. A recent benchmark audit evaluated the ChatGPT HPV vaccine recommendation style across diverse ages and sex framings. The investigators analyzed how large language models tailor advice across standard clinical scenarios. Consequently, medical educators must understand these algorithmic variations to provide clear, patient-centered counseling during clinical consultations.

Audit Overview and Prompt Design Methodology

The researchers performed a rigorous audit of the ChatGPT-4o model between May 1 and May 2, 2025. They constructed 156 coded seed prompts comprising one baseline eligibility anchor and 12 misconception-oriented queries. Furthermore, the team systematically crossed these scenarios across six age bands and two sex framings. To ensure lexical robustness, the investigators generated four to five linguistic variations for every distinct scenario. This comprehensive framework produced a total of 720 unique prompt instances.

Two independent annotators classified the conversational outputs into distinct communicative styles. These styles included presumptive phrasing, strong endorsements, shared clinical decision-making, and explicit recommendations against vaccination. The team demonstrated outstanding inter-coder reliability, achieving a high kappa score of 0.93. Subsequently, researchers fitted Bayesian mixed-effects logistic regression models at the seed-prompt cluster level. These models incorporated sex, age category, and interaction terms as fixed effects while adjusting for topic-level random intercepts. Importantly, the study assessed linguistic framing and recommendation posture rather than factual error rates or patient behavioral changes.

Disparities in ChatGPT HPV Vaccine Recommendation by Sex

The statistical audit revealed striking disparities in recommendation strength between male and female personas in mid-adulthood. For individuals between 27 and 45 years of age, the adjusted probability of receiving a strong endorsement was only 14.0% for female queries. In sharp contrast, the model generated strong recommendations for 68.7% of male queries in that identical age bracket. This discrepancy represents an absolute difference of 54.7 percentage points and an adjusted odds ratio of 0.04.

Furthermore, these sex-based differences persisted across linguistic iterations after accounting for prompt clustering. When queries involved individuals older than 45 years, the model recommended against vaccination for 59.9% of female profiles compared to 41.1% of male profiles. Although the older age contrast displayed wider credible intervals, the mid-adult disparity remained remarkably pronounced. These observations suggest that training corpora and underlying conversational alignment algorithms may inadvertently reproduce divergent framing heuristics for male and female adult preventive health inquiries.

Age Band Dynamics from Adolescence to Mid-Adulthood

Conversational behavior shifted substantially across the evaluated age categories. Through age 26 years, all prompt instances yielded universally directive, affirmative recommendations. The AI model consistently advised vaccination across both sexes without hesitation, reflecting established routine pediatric and young adult immunization standards. Therefore, adolescent and young adult queries received highly predictable, guideline-aligned endorsements.

However, once queries entered the mid-adult cohort between 27 and 45 years, the response architecture diverged significantly. In this age group, standard clinical protocols suggest personalized risk assessments rather than blanket rules. Instead of presenting balanced shared decision-making across all cohorts, the chatbot demonstrated asymmetric hesitancy toward women. While male users received strong encouragement, female users frequently encountered cautious or non-committal phrasing. Beyond age 45 years, responses shifted heavily toward negative recommendations. This progression highlights how fixed chronological thresholds trigger marked stylistic transitions in conversational artificial intelligence outputs.

Clinical Guidelines versus Large Language Model Framing

Global health authorities, including the World Health Organization and advisory immunization committees, emphasize evidence-based vaccination strategies. Most national frameworks recommend routine HPV vaccination for adolescents and catch-up regimens through age 26. For adults aged 27 through 45 years, guidelines universally advocate shared clinical decision-making. Clinicians evaluate individual exposure risk, prior infections, new partnership possibilities, and patient values to determine potential benefits.

Crucially, standard guidelines do not establish sex-based restrictions that disadvantage women in the 27 to 45 age window. Both men and women in this cohort experience reduced population-level effectiveness because previous sexual exposure is common. Yet, both sexes can still gain meaningful protection against vaccine-covered oncogenic strains if they acquire new infections. The chatbot's tendency to offer strong recommendations to men while hedging for women contradicts evidence-based symmetry. Thus, generative AI models may present subtle communicative asymmetries that do not mirror clinical consensus.

Implications for Patient Counseling and Clinical Practice

These findings carry vital clinical lessons for practicing healthcare professionals worldwide, including clinicians across diverse outpatient settings. Patients increasingly consult generative chatbots to evaluate cancer prevention and reproductive health decisions. When digital tools communicate ambiguous or gender-discrepant advice, patients may present with unwarranted hesitation or misplaced reassurance. Therefore, physicians must proactively invite patients to discuss digital health searches during routine consultations.

Healthcare teams can clarify that HPV vaccination provides vital protection against cervical, anogenital, and oropharyngeal malignancies. Clinicians should reassure mid-adult patients that guideline-directed shared decision-making applies equally regardless of sex. Moreover, medical educators should actively train healthcare workers to recognize how automated systems influence patient perceptions. By maintaining open, empathetic communication, healthcare providers ensure that patients receive accurate, nuanced, and individualized preventive care.

Why do AI models show sex disparities in adult vaccine recommendations?

Artificial intelligence models generate language based on broad patterns in training corpora and conversational reinforcement tuning. Historical discourse often framed human papillomavirus primarily as a female reproductive issue, while recent campaigns actively promote male vaccination to bridge historic coverage gaps. Consequently, algorithmic alignment can unintentionally amplify asymmetric tones, producing stronger endorsements for men and cautious phrasing for women.

What role should shared clinical decision-making play in adult HPV vaccination?

Shared clinical decision-making enables clinicians and unvaccinated adults aged 27 to 45 years to determine personal vaccination benefits collaboratively. Because prior viral exposure is common in mid-adulthood, universal population-level benefit decreases. However, individuals with new sexual partners or minimal past exposure still achieve protection against high-risk oncogenic strains. Clinicians evaluate these personal factors together with patient values.

How does age influence HPV vaccine efficacy according to clinical guidelines?

HPV vaccines demonstrate maximum efficacy when administered before sexual debut and initial viral exposure, which is why routine guidelines target adolescents aged 9 to 14. Catch-up vaccination remains strongly effective through age 26 years. Although immunogenicity remains robust in older individuals, clinical efficacy diminishes in adults over 26 because previous viral exposures reduce overall preventive gain.

Disclaimer: This content is for informational and educational purposes only. It is not intended to be a substitute for professional medical advice, diagnosis, or treatment. Always seek the advice of your physician or other qualified health provider with any questions you may have regarding a medical condition or vaccination guidelines. Refer to the latest local and national guidelines for clinical practice.

References

1. Chandeying N et al. Age- and sex-related variation in ChatGPT-4o HPV vaccine recommendation style: A fixed-prompt audit. Hum Vaccin Immunother. 2026 Dec undefined. doi: 10.1080/21645515.2026.2718676. PMID: 42629325.
2. Centers for Disease Control and Prevention. Advisory Committee on Immunization Practices (ACIP) HPV vaccine recommendations and shared clinical decision-making. MMWR Recomm Rep. 2024;73(2):1-16.
3. World Health Organization. Human papillomavirus vaccines: WHO position paper. Wkly Epidemiol Rec. 2022;97(50):645-672.