The four-step framing
- Intake. The AI asks for symptoms, duration, intensity, exposure context, household members, and any chronic conditions. It will keep asking calibrated follow-ups until the picture is clear enough to be useful.
- Pattern recognition, not diagnosis. It compares the input against common patterns — viral upper respiratory, gastrointestinal, urinary, dermatologic, etc. — and against red-flag patterns. It does not commit to a single illness.
- Vulnerability weighting. Anyone in a higher-risk group raises the recommended tier by one step relative to a baseline-healthy adult.
- Tier and watch criteria. The AI returns a tier (monitor / telehealth / urgent / ER) and the specific signs that should change the tier in either direction over the next 24–72 hours.
What goes into the framing
Symptom dimensions
- Onset (sudden vs. gradual), duration, and trend (improving / steady / worsening).
- Severity in terms of function, not just numbers.
- Specific red-flag presence: chest pain, difficulty breathing, neurological signs, etc.
Exposure context
- Recent crowd events, travel, workplace clusters, household illness.
- Specific exposure types — respiratory, food, water, animal, vector-borne, environmental.
- Time since exposure relative to typical incubation windows.
Household and personal factors
- Age extremes (infants, older adults).
- Pregnancy status.
- Immunocompromise, chronic disease, recent surgery, complex medication regimens.
- Presence of vulnerable family members in the same household.
How the tier translates to action
- Monitor. Rest, fluids, fever-reducer if needed, watch for the named red flags. Re-check in 24–48 hours.
- Telehealth or same-day clinic. Bring a structured summary. The AI will help you compose one.
- Urgent care. Same day, in person. Best for moderate symptoms that need eyes-on but are not life-threatening.
- ER or 911. For red-flag patterns and rapid worsening. The AI surfaces these directly when present and recommends not waiting.
The conservative tilt — and why
BioShield AI is calibrated to err toward escalation rather than reassurance. The asymmetry is intentional: a wasted urgent-care visit is a minor cost; a delayed ER visit can be a major one. That tilt increases for higher-risk people and for symptoms with serious downside risk (chest, brain, breathing, allergic).
What it deliberately avoids
- Naming a single illness.
- Telling you what medication or dose to take.
- Promising any specific outcome.
- Treating "I feel fine now" as proof that the original concern was overblown.
End-to-end example. Adult with three days of fever to 101°F, mild cough, attended a wedding last weekend, lives with an 80-year-old parent. The AI returns: monitor tier, telehealth-ready summary, watch criteria (shortness of breath, fever >72 hours, the parent developing symptoms), and a household isolation playbook. No diagnosis, no prescription — but a clear next 48 hours.
Try the four-step framing on your own situation.
Open AI Risk Guide →Related: What the AI Can and Cannot Do · Editorial Standards · Symptom Hub.
Editorial
Primary sources
- CDC — Centers for Disease Control and Prevention
- MedlinePlus — U.S. National Library of Medicine
- National Institutes of Health (NIH)
- WHO — Health Topics
- U.S. National Library of Medicine
External links open the cited public-health resource. BioShield AI does not control external content; consult a qualified clinician for personal medical decisions.