When People Ask AI About Sex: Generative Chatbots as the New Front Door to STI Care

1. Introduction

Sexually transmitted infections (STIs) remain a major global public health challenge. Despite advances in diagnostics, treatment, vaccination, HIV prevention, and digital health, access to timely and non-stigmatizing sexual healthcare remains profoundly unequal [1,2,3]. The World Health Organization’s definition of sexual health extends beyond the absence of infection or dysfunction, emphasizing physical, emotional, mental, and social well-being in relation to sexuality [4]. Yet, for many people, the first step after a potentially risky sexual encounter is not contacting a clinic, physician, nurse, pharmacist, or community organization. Increasingly, it is opening a generative artificial intelligence chatbot.

A person may type: “I have burning when I urinate after unprotected sex. Is it probably an STI?” Another may ask: “Can I get HIV from what happened last night?”, “Should I take the antibiotics I have at home?”, “Does this mean that my partner cheated?”, “Can I wait a few days before getting tested?”, or “Where can a trans woman be tested without being judged?” These are not merely requests for factual information. They frequently combine medical uncertainty, fear, shame, relationship distress, urgency, and privacy concerns.

When generative chatbots respond to such questions, they are no longer functioning solely as general information systems. They are occupying a position previously held by sexual health helplines, clinicians, peer educators, search engines, and trusted community organizations. In practical terms, generative artificial intelligence is becoming a new, informal front door to STI care.

The central question for venereology is therefore no longer whether patients will use generative AI to discuss sexual health. They already can. The more consequential question is whether this conversational doorway will lead people towards accurate information, appropriate testing, timely treatment, and person-centered care or towards false reassurance, unnecessary anxiety, self-medication, delayed treatment, and preventable transmission.

2. The Appeal of a Non-Judgmental Conversational Space

Sexual health is particularly suited to conversational technologies because it remains associated with embarrassment, stigma, fear of disclosure, and anticipation of negative judgment. Conventional services may be inaccessible because of geographical distance, cost, restricted opening hours, language barriers, concerns about confidentiality, disability, migration status, sexual orientation, gender identity, or previous experiences of discrimination.

Chatbots appear to address several of these barriers simultaneously. They are available continuously, respond immediately, can provide information in multiple languages, and permit users to formulate questions in their own words. Unlike a conventional search engine, a chatbot can respond to follow-up questions, simplify explanations, and adapt its language to the user’s apparent level of knowledge.

Research on conversational agents in healthcare has identified potential benefits relating to accessibility, patient education, self-management, and behavioral support [5,6,7,8]. Studies specifically addressing sexual and reproductive health have also reported considerable interest in chatbot-based services, particularly when users value anonymity, privacy, convenience, and freedom from interpersonal judgment [9,10]. Sexual and reproductive health professionals have similarly recognized the potential value of conversational agents, although they have expressed concerns about clinical accuracy, safeguarding, empathy, and accountability [11].

Purpose-built systems such as SnehAI in India and pleasure-oriented sexual health chatbots in Kenya illustrate how conversational technologies can deliver culturally adapted information, support engagement, and address topics that users may hesitate to discuss with professionals [12,13]. Clinic- and community-based deployment studies also suggest that carefully designed chatbots can facilitate access to information and, importantly, encourage some users to proceed to appointments or other forms of care [14]. Digital interventions may either reduce or reproduce sexual health stigma, however, depending on their language, assumptions, visual design, and underlying conceptualization of sexuality [15].

These findings are encouraging, but earlier sexual health chatbots were relatively restricted systems based on predefined content, decision trees, curated knowledge bases, or narrow conversational domains. Generative AI is qualitatively different. Large language models can produce novel, apparently personalized responses to almost any question. This open-ended conversational capacity increases usefulness, but it also expands the range of possible errors.

3. Fluency Is Not Clinical Safety

Generative chatbots often sound confident, empathic, and clinically sophisticated. This linguistic fluency may encourage users to attribute greater reliability to the response than is warranted. A plausible answer, however, is not necessarily an accurate answer, and an accurate general statement is not necessarily a safe recommendation for a particular person.

Recent evaluations of AI chatbots answering sexual health questions demonstrate both promise and significant variability. In a consensus study using real-world clinical queries, the accuracy of responses varied substantially across chatbot systems and prompting conditions. The base version of ChatGPT version 3.5 achieved an accuracy score of approximately 65%, whereas more specifically configured systems performed better, although errors persisted even in the strongest models [16]. Other investigations found that chatbot responses about STIs could be broadly reliable but remained limited by readability, incompleteness, lack of specificity, and the absence of individualized clinical assessment [17,18].

An evaluation of ChatGPT in HIV-prevention communication reported generally strong performance across several quality dimensions, including accuracy and inclusivity, but also highlighted the need for expert oversight and continued evaluation [19]. An earlier editorial in Venereology appropriately identified AI’s potential role in STI prediction, diagnosis, surveillance, and prevention [20]. The current challenge extends beyond these applications: general-purpose generative systems are already communicating directly with individuals before those individuals enter formal healthcare.

Evidence from broader medical contexts reinforces this dual interpretation. Large language models have demonstrated considerable capacity to encode clinical knowledge and generate responses rated favorably for informativeness and empathy [21,22,23]. Systematic evaluations nevertheless show heterogeneous performance, sensitivity to question wording, inconsistent citation practices, hallucinated content, and important variations between models, versions, and clinical specialties [24,25]. Consequently, impressive benchmark results cannot be equated with reliable performance in uncontrolled, real-world sexual health conversations.

The danger is not only that a chatbot may provide an entirely false statement. More subtle forms of failure may be clinically more consequential: omitting a critical qualification, failing to recognize urgency, providing guidance that is outdated in a particular jurisdiction, recommending an inappropriate testing window, or overlooking a symptom requiring immediate medical assessment.

4. Why Sexual Health Is a High-Stakes Use Case

Sexual health questions have characteristics that make them especially challenging for general-purpose generative models.

First, clinical risk often depends on details that users may not initially provide. Relevant information can include the anatomical sites involved, sexual practices, condom use, timing of exposure, symptoms, pregnancy possibility, vaccination status, recent antibiotic use, HIV-prevention strategies, previous infections, current medication, immunosuppression, and local epidemiology. A safe system must know which clarifying questions are necessary without subjecting the user to an intrusive or moralizing interrogation.

Second, some sexual health decisions are time-sensitive. A conversation about a recent potential HIV exposure may require urgent referral for post-exposure prophylaxis. Reports of severe pelvic or testicular pain, fever, neurological or ocular symptoms, sexual assault, pregnancy-related concerns, or rapidly worsening symptoms may require immediate clinical assessment. A chatbot that offers a lengthy explanation but fails to identify urgency has not provided safe care.

Third, testing recommendations depend on the infection, anatomical site, test technology, timing, symptoms, and local clinical guidance. Generic advice to “get an STI test” may be insufficient when extragenital testing, repeat testing, confirmatory testing, or examination is indicated.

Fourth, many users are not seeking biomedical information alone. They may be asking whether an infection proves infidelity, whether they should disclose a diagnosis to a partner, whether their sexual behavior is “normal”, or whether a clinician will judge them. A response can be technically accurate while reinforcing stigma, blame, heteronormativity, cisnormativity, monogamy assumptions, or misconceptions about particular communities.

Fifth, the interaction may include disclosures of coercion, exploitation, sexual violence, self-harm, child sexual abuse, or non-consensual image sharing. These situations require safeguarding pathways that cannot be reduced to probabilistic text generation.

5. Seven Failure Modes That Venereology Must Anticipate

5.1. Confident Misinformation

Generative models are designed to generate probable linguistic sequences rather than independently verify clinical truth. They may produce incorrect information, fabricated citations, nonexistent services, or unsupported probabilities while maintaining an authoritative tone. Hallucination is particularly dangerous when the user has limited health literacy or perceives the chatbot as a medical authority [25,26].

5.2. False Reassurance and Delayed Care

A chatbot may normalize symptoms that require examination or suggest that infection is unlikely without obtaining sufficient information. Reassurance is psychologically attractive and may therefore be followed more readily than precautionary advice. Delayed testing and treatment can affect both individual outcomes and onward transmission.

5.3. Unnecessary Alarm

The opposite error is also harmful. Overinclusive lists of severe diseases can intensify health anxiety, promote repeated testing, or encourage inappropriate emergency service use. Sexual health responses should communicate uncertainty proportionately rather than framing every symptom as either harmless or catastrophic.

5.4. Self-Diagnosis and Self-Medication

Users may request exact treatment regimens, ask whether they can use medication prescribed to another person, or seek confirmation that leftover antibiotics are appropriate. Chatbots that provide decontextualized prescribing information may inadvertently facilitate incorrect dosing, contraindicated medication use, incomplete treatment, masking of symptoms, and antimicrobial resistance.

5.5. Decontextualized or Geographically Inappropriate Advice

STI guidelines, testing pathways, service availability, age-of-consent provisions, confidentiality rules, partner-notification procedures, and access to HIV prevention vary across jurisdictions. A general-purpose model may blend recommendations from different countries without disclosing the source or applicability of the advice.

5.6. Reproduction of Stigma and Inequity

Language models learn from social and digital data that contain racism, sexism, homophobia, transphobia, ableism, stigma towards sex workers, and assumptions about sexual behavior. Algorithmic systems can reproduce inequities even without explicit discriminatory rules [27,28,29]. In sexual health, biased assumptions may lead a system to over-associate particular infections with specific populations, ignore the needs of transgender and gender-diverse users, or frame consensual non-monogamy and sex work as inherently pathological.

5.7. Extraction of Intimate Data Without Meaningful Consent

Sexual health conversations may reveal sexual orientation, gender identity, HIV status, reproductive intentions, histories of violence, partner information, location, and other highly sensitive data. Users may not know whether these disclosures are retained, used for model training, reviewed by humans, linked to other data, or shared with third parties. The apparent privacy of a one-to-one conversation must not be confused with actual confidentiality [30,31,32].

6. From General Information to Safety-Oriented Conversational Triage

A generative chatbot cannot conduct a physical examination, collect diagnostic specimens, verify a patient’s identity, assess all contextual variables, or assume professional responsibility. It should therefore not present itself as a substitute for a clinician. This does not mean that such systems have no legitimate role in STI care.

A safer role is conversational orientation and triage: helping users understand possible explanations, identify urgency, prepare for a consultation, locate appropriate services, reduce stigma, and take evidence-based next steps. The objective should not be to simulate diagnostic certainty, but to move the user towards the correct level of care.

This distinction must be explicit. Statements such as “I cannot diagnose an STI from this conversation” are useful but insufficient when placed at the end of an otherwise highly diagnostic response. Uncertainty and scope limitations should shape the entire interaction.

Specialized sexual health chatbots may offer greater safety than unconstrained general-purpose systems when they use curated content, retrieval from authoritative guidelines, defined clinical boundaries, structured triage pathways, and direct links to human services. Systematic reviews of contraceptive and sexual health chatbots suggest that usefulness depends less on conversational novelty than on evidence quality, co-design, privacy, cultural adaptation, and integration into care pathways [33,34,35].

7. Minimum Requirements for Generative AI in STI Care

Any public-facing generative AI system that provides sexual health guidance should meet a minimum safety standard. These requirements derive from emerging AI governance frameworks, healthcare ethics guidance, and the distinctive clinical characteristics of STI care [26,27,28,29,30,31,32,36,37,38,39,40] (see

Table 1. Minimum Requirements for Generative AI in STI Care.

Domain	Minimum Operational Requirement
Scope and identity	Clearly identify the system as AI-generated support; distinguish information, orientation, and triage from diagnosis or treatment; disclose relevant limitations.
Clinical safety triage	Detect urgent time windows and red-flag symptoms; prioritize immediate referral when HIV post-exposure prophylaxis, emergency assessment, safeguarding, or sexual-assault care may be indicated.
Clarifying questions	Ask only clinically necessary, behavior- and anatomy-based questions; avoid assumptions about gender, orientation, relationship structure, or sexual practices.
Evidence grounding	Retrieve information from current, authoritative guidelines; identify the jurisdiction and date of the guidance; avoid unsupported estimates and fabricated references.
Actionable referral	Provide concrete next steps, including how and where to access testing, treatment, vaccination, HIV prevention, mental health support, or emergency services.
Inclusive communication	Use non-judgmental, sex-positive, trauma-informed, and gender-inclusive language; accommodate different levels of health and digital literacy.
Privacy protection	Minimize the collection of identifiable or unnecessary sexual data; explain retention, secondary use, human review, and model-training practices in accessible language.
Human escalation	Offer a rapid route to a trained professional when the question exceeds the system’s competence, the user remains distressed, or risk cannot be assessed reliably.
Continuous validation	Evaluate performance across languages, cultures, literacy levels, genders, sexual orientations, disabilities, and racial or ethnic groups; repeat testing after every material model update.
Accountability and monitoring	Document model versions, content sources, safety events, corrective actions, and governance responsibility; establish mechanisms through which users can report harmful answers.

).

These safeguards should not be optional features added after deployment. They should determine whether an AI system is sufficiently safe to enter the sexual healthcare environment.

8. Inclusive Language Is a Clinical Requirement

Inclusivity is sometimes treated as a communication preference rather than a safety issue. In STI care, this distinction is untenable. Risk assessment based on identity labels alone is frequently inaccurate. A man who identifies as heterosexual may have sex with men; a transgender man may require cervical screening; a transgender woman may require site-specific testing based on anatomy and sexual practices; and people in monogamous relationships may still present with an STI acquired before the relationship or through non-sexual transmission routes.

Chatbots should therefore use anatomy- and behavior-based questions where clinically appropriate. They should not infer sexual practices from gender identity, pronouns, marital status, or orientation. Neither should they frame certain identities or consensual practices as intrinsically risky.

A sex-positive approach does not minimize infection risk. Rather, it communicates prevention without shame and recognizes pleasure, autonomy, intimacy, and relationship diversity as legitimate components of sexual health. This may improve trust, disclosure, and engagement more effectively than moralistic messaging.

9. Privacy Must Extend Beyond a Disclaimer

The sensitivity of sexual health information requires privacy protections substantially stronger than a generic statement that users should not share personal data. A person asking about an exposure may inadvertently disclose names, locations, dates, medical histories, or identifiable partner information. Some users may assume that deleting a visible conversation deletes the underlying data.

Developers and healthcare organizations should implement data minimization by design. Systems should avoid requesting names, exact addresses, photographs, or other identifiers unless essential to a clearly defined service. Users should be informed whether conversations are stored, reviewed by humans, used to improve the system, transferred internationally, or connected to advertising and analytics infrastructures.

Privacy considerations are particularly urgent for users living in settings where same-sex activity, sex work, HIV exposure, abortion, gender diversity, or adolescent sexuality may be criminalized or heavily stigmatized. In such contexts, inadequate data governance can create social and legal harms beyond the immediate clinical encounter.

WHO guidance on AI for health and large multimodal models emphasizes transparency, human oversight, protection of autonomy, accountability, and rigorous evaluation [30,31]. Similar principles appear in UNESCO’s ethical framework, the NIST AI Risk Management Framework, and the European Union’s regulatory approach [38,39,40]. Sexual health services should translate these broad principles into operational requirements appropriate to intimate, stigmatized, and potentially legally sensitive information.

10. Evaluation Must Move Beyond Answer Accuracy

Current evaluations commonly assess whether chatbot answers are factually correct, complete, readable, or similar to expert responses. These indicators are necessary but insufficient.

A clinically meaningful evaluation should ask whether chatbot use leads to: (1) appropriate and timely STI testing; (2) correct anatomical-site testing; (3) timely access to HIV post-exposure prophylaxis; (4) reduced inappropriate antibiotic use; (5) successful linkage to treatment; (6) appropriate partner notification; (7) improved vaccination or prevention uptake; (8) reduced sexual health stigma and shame; (9) improved understanding without excessive anxiety; (10) equitable outcomes across demographic and social groups; (11) fewer delays caused by false reassurance; and (12) recognition and escalation of safeguarding concerns.

Evaluations must also consider model instability. A chatbot’s performance on a fixed set of questions at one point in time cannot guarantee that the same answers will be produced after a model update, safety-policy modification, retrieval-system change, or alteration in the user prompt. Continuous post-deployment surveillance is therefore essential.

Adversarial and intersectional testing should include ambiguous symptoms, slang, misspellings, mixed languages, low-literacy questions, indirect disclosures of assault, transgender-specific scenarios, disability-related accessibility needs, sex work, chemsex, non-monogamy, migration-related barriers, and questions from adolescents. Systems that perform well only when presented with carefully worded textbook questions are not ready for real-world sexual healthcare.

11. Clinicians and Sexual Health Services Should Occupy the Digital Front Door

Health services may be tempted to view generative chatbots as external technologies beyond their responsibility. That position is increasingly difficult to defend. When patients arrive at consultations after receiving AI-generated advice, chatbot interactions become part of the clinical pathway regardless of whether the healthcare system formally endorses them.

Clinicians should ask patients, without judgment, whether they have consulted a chatbot and what advice they received. This can identify misconceptions, understand the patient’s concerns, and prevent embarrassment about disclosing AI use. Digital health literacy should include guidance on how to question AI-generated information, verify sources, protect privacy, and recognize situations requiring professional assessment.

Sexual health organizations should also consider developing or endorsing specialized conversational tools linked to authoritative information and local services. Leaving the digital front door entirely to commercial, general-purpose platforms creates a vacuum in which clinical governance, public health priorities, and community participation may be secondary to engagement, data collection, or product growth.

Community involvement is indispensable. LGBTQIA+ communities, people living with HIV, sex workers, adolescents, migrants, people with disabilities, and other populations disproportionately affected by barriers to sexual healthcare should participate in system design, testing, governance, and evaluation. Co-design should include actual authority over decisions, not merely consultation after key technological choices have already been made.

12. A Research Agenda for Venereology

Venereology should treat generative conversational AI as a substantive research field rather than a peripheral digital-health curiosity. Priority questions include:

• How frequently do people use general-purpose chatbots before or instead of accessing STI services?
• Which types of questions are most likely to produce unsafe or inequitable responses?
• Can chatbot-assisted triage improve the timeliness of testing, HIV prevention, and treatment?
• What is the psychological effect of AI-generated reassurance, uncertainty, and risk communication?
• How should systems respond to questions involving infidelity, disclosure, coercion, sexual assault, or partner violence?
• What governance structures are required when chatbots are integrated into public sexual health services?
• How can model performance be monitored without creating repositories of highly sensitive sexual data?
• Which forms of human escalation are most acceptable and effective?
• How do language, culture, health literacy, gender identity, and sexual orientation affect both chatbot performance and user trust?
• What are the unintended consequences of normalizing AI as an intermediary in intimate health decisions?

Answering these questions will require collaboration between venereologists, infectious disease specialists, psychologists, sexologists, nurses, epidemiologists, computer scientists, ethicists, public health professionals, legal experts, and affected communities.

13. Conclusions

Generative chatbots may lower the threshold for asking difficult sexual health questions. They may offer a private starting point for people who are frightened, geographically isolated, stigmatized, or uncertain about how to access care. They may translate complex information, normalize testing, reduce shame, and guide users towards appropriate services.

The same systems can also provide convincing misinformation, miss time-critical interventions, reproduce stigma, encourage self-medication, and collect some of the most intimate data a person can disclose. Their conversational fluency should not be mistaken for clinical competence or accountability.

The appropriate response is neither unconditional enthusiasm nor categorical rejection. Venereology should define the conditions under which generative conversational systems can contribute safely to STI prevention and care. These conditions must include evidence grounding, structured triage, privacy protection, inclusive communication, human escalation, continuous evaluation, and clear institutional accountability.

The ethical objective is not to build a chatbot that merely sounds like a sexual health clinician. It is to build a system that recognizes the limits of conversation, protects the person behind the question, and knows when digital dialogue must become human care.