What BioShield AI can and cannot do.

What BioShield AI is built to do well

BioShield AI is a structured risk-framing assistant. It is narrow on purpose, and within that scope there are a few things it does consistently:

• Frame next steps. Given symptoms, exposure, household setup, and risk factors, it returns whether the situation reads as monitor-at-home, telehealth, urgent care, or ER. Example: "three-day low fever, no red flags, healthy adult" reads as monitor.
• Surface red flags. If anything in your description matches an emergency pattern, it flags it directly and recommends faster action. Example: "sudden severe chest pain radiating to the jaw" surfaces a 911 prompt and stops the rest of the intake.
• Account for vulnerability. Older adults, infants, pregnancy, immunocompromise, and major chronic conditions get explicitly weighed. Example: a 39°C fever in a healthy 30-year-old is framed differently than the same temperature in an 80-year-old on chemotherapy.
• Translate symptoms into watch criteria. Rather than telling you what is happening, it tells you what to watch for over the next 24 to 72 hours. Example: "fever climbing past 103, new shortness of breath, or fluids stop staying down."
• Help you prepare for a clinician visit. It can compose a structured summary, in plain language, with timeline, measurements, medications, and a specific question. Example output: a five-line note you can read straight into a telehealth window.
• Stay calm and calibrated. Most symptom checkers fire red flags at everything. BioShield AI tries to be matter-of-fact about routine viral illness while being unambiguous about the patterns that need same-day attention.
• Hold context within a session. Within a conversation, it tracks what you have already said about household, age, and prior symptoms, so you do not have to repeat yourself. Example: once you mention you are 28 weeks pregnant, that fact carries through every later question.

What it will explicitly not do

• It will not diagnose. No "you have flu," no "this is appendicitis." Many illnesses overlap in their early symptoms; only a clinician with examination, labs, and history can responsibly assign a diagnosis.
• It will not prescribe. Medication choice and dosing depend on your full medical record. Generic over-the-counter categories are within scope; specific prescription advice is not.
• It will not replace a clinician. When a situation calls for hands-on examination, imaging, or labs, the AI says so and routes you toward in-person care.
• It will not guarantee accuracy. Like any large-language-model system, it can occasionally misread input or miss important context. It is one structured input into a decision, never the decision itself.
• It will not promote products. Risk framings, escalation criteria, and content are not influenced by sponsors or advertisers. There is no affiliate layer underneath the recommendations.

Specific situations where it should not be used

• Active emergencies. If a symptom is severe or rapidly worsening, call 911 or your local emergency number first. Talk to BioShield AI later, if you want, when you and your family are safe.
• Lab-result interpretation. Bloodwork, imaging reports, and pathology results need a clinician with your full chart, not a chat window. The AI will not try to read a lab panel for you.
• Managing complex chronic disease alone. Heart failure, diabetes management, transplant care, oncology treatment, and complex psychiatric conditions need ongoing professional management. The AI can help you frame a question for your team; it cannot replace the team.
• Replacing prenatal or well-child care. Routine pediatric visits and prenatal monitoring catch issues that no chat tool can detect. Use BioShield AI alongside, not instead of, those visits.
• Mental-health crises beyond a basic risk-flag. The AI will surface crisis resources and recommend emergency care when warranted, but it is not a therapist and not a crisis line. For acute mental-health emergencies, call 988 in the United States, your local crisis line, or 911.

How it handles uncertainty

Health questions almost always carry uncertainty. The AI is built to surface that uncertainty rather than mask it. When a description is ambiguous, it asks clarifying questions before guessing — about timeline, measurements, and household context. When a symptom could be benign or could be serious, it gives you the watch criteria for both branches and names the specific signs that would tip the balance toward escalation. The output is closer to "here is how to think about this for the next 24 hours" than "here is what you have." For deeply unfamiliar scenarios — including the speculative ones explored on the unknown pathogens page — the AI will not pretend to identify the cause; it stays on the same calm escalation logic it uses for ordinary illness.

How it handles bias and edge cases

The AI is intentionally tuned conservatively. When in doubt, it suggests the higher-care tier rather than the lower one. The reason is asymmetric cost: a wasted urgent-care visit is a few hours and a copay, while a missed serious diagnosis can be far worse. That conservative tilt increases for higher-risk groups — infants under 3 months, pregnancy, older adults, immunocompromise, complex chronic disease — because their margins for error are smaller and their atypical presentations are more common.

Known failure modes

• Long context loss. In very long conversations, important early details can drift out of focus. If you have been chatting for a while, restate the essentials before asking the final question.
• Ambiguous symptom descriptions. "Stomach hurts" covers dozens of possibilities; the AI will ask follow-ups, but vague input produces vaguer output.
• Regional variation in care availability. Urgent care, telehealth, and after-hours options vary widely by location. Recommendations assume a typical U.S. care landscape and may not match a rural area or a different country.
• Language nuance. Translated descriptions, idioms, and culturally specific symptom language can occasionally be misread. Plain literal descriptions help.

What you can do to get a better response

The single biggest lever is specificity. Share an actual timeline ("started Monday afternoon, fever climbed Tuesday, this morning is the worst") rather than "a few days." Share measurements when you have them: highest temperature, lowest oximeter reading, blood pressure, weight in kilograms or pounds. Share who else is in the home, including ages and any chronic conditions, since household context changes the framing. Share what you have already tried and how it went. The more concrete the input, the more useful the output.

If you disagree with the AI

Trust your gut and your clinician over the chat window. Parents and partners notice things that no symptom description captures: a child who is not quite right, a parent who is unusually quiet, a spouse who looks pale in a way that worries you. If the AI says monitor and your instinct says go in, go in. If the AI says urgent care and your clinician's nurse line says it can wait, follow the clinician. The AI is a structured second look, not a verdict.

Related: How Risk Guidance Works · Editorial Standards · Medical Disclaimer.