Most patient information fails in one of two directions. It either sounds more certain than the evidence allows — “your symptoms are caused by a virus, rest at home” — or it hedges so heavily that the reader cannot tell what to do next. Both are safety problems. False certainty delays care. Vague hedging gets ignored, which delays care in a different way.
Uncertainty is not a defect in clinical information; it is usually the honest state of the evidence at the moment a patient reads it. Simpkin and Schwartzstein made the point in the New England Journal of Medicine in 2016: tolerating uncertainty is a core clinical skill, not a weakness (https://www.nejm.org/doi/full/10.1056/NEJMp1606402). Products, however, often write as if the skill does not exist. This article sets out how to communicate uncertainty in patient-facing text so that readers stay accurately informed and still know what to do.
Table of Contents
- Why uncertainty is a safety issue, not a tone issue
- Three kinds of uncertainty, three different sentences
- Wording patterns that hold up under review
- A worked example: rewriting one unsafe paragraph
- A note for pediatric products
- A review checklist you can reuse
- What a content review is — and is not
- Frequently asked questions
Why uncertainty is a safety issue, not a tone issue
A patient reading your product at 2 a.m. is making a decision: seek care now, wait and watch, or stop worrying. Every sentence about certainty pushes that decision. “This is usually harmless” nudges towards waiting. “This could be serious” nudges towards acting. Neither is neutral, so the question is never whether to influence the decision, but whether the influence matches the evidence.
Two specific harms follow from getting this wrong. The first is false reassurance: wording that closes a question the evidence leaves open. The second is alarm without direction: wording that raises anxiety but gives no time frame and no action. Our earlier piece on why accurate medical content is not enough for a safe AI product covers the first harm in depth; uncertainty wording is where it most often hides, because a statement can be factually correct and still imply a certainty it does not have.
Three kinds of uncertainty, three different sentences
Reviewers find it useful to separate uncertainty into three kinds, because each needs a different sentence structure.
- Diagnostic uncertainty — what is causing this? Early symptoms overlap across conditions, so the honest statement names the likely causes and the ones that must not be missed. Pattern: “The most common causes are X and Y. Less commonly, this can be a sign of Z.”
- Prognostic uncertainty — what happens next? The honest statement gives a typical course and a time frame. Pattern: “Most cases settle within three to five days. If it is not improving by day five, contact a clinician.”
- Action uncertainty — what should I do given that I cannot be sure? This is resolved not by more information but by safety netting: named warning signs and a specific route back into care. We covered the structure of that advice in how to review red-flag and escalation advice in a symptom checker.
Most weak patient text collapses all three into a single hedge — “it might be nothing, but see a doctor if you are worried” — which answers none of them.
Wording patterns that hold up under review
Four patterns recur in patient information that survives clinical review:
- Name the likelihood in words, anchored to a base rate where one exists. “Most”, “commonly”, “rarely” are acceptable only if the underlying frequencies support them. Where a reliable figure exists, give it: “about 1 in 10 children with this infection will…” Numbers need a source and a review date.
- Pair every uncertainty with its time boundary. Uncertainty without a clock leaves the reader alone with it. “We cannot tell yet” becomes safe when followed by “this usually becomes clearer within 48 hours; if it has not, do this.”
- State what would change the advice. “If the fever rises above…, if breathing becomes…, if your child stops drinking…” Conditional triggers convert vague worry into a watchable list.
- Keep one register throughout. Mixing a confident headline (“It’s just a cold”) with a cautious body confuses readers about which to believe. Readers remember the headline.
Health-literacy guidance points the same way: the AHRQ Health Literacy Universal Precautions Toolkit recommends plain language and explicit next steps for all readers, not only those who ask (https://www.ahrq.gov/health-literacy/improve/precautions/toolkit.html). Our guide to how to evaluate AI-generated patient instructions shows how to test readability and actionability together.
A worked example: rewriting one unsafe paragraph
Original (typical of generated content):
“Your child’s cough is most likely caused by a virus and does not need antibiotics. It should get better on its own. If you are concerned, speak to your doctor.”
Problems: it asserts a cause the product cannot know, implies a prognosis with no time frame, and ends with a trigger (“if you are concerned”) that transfers the clinical judgement to the parent.
Rewrite:
“A cough like this is most often caused by a viral infection, which antibiotics do not treat. It usually improves over one to two weeks. Contact a clinician the same day if your child is breathing faster than usual, is struggling to drink, has a fever that lasts more than five days, or seems unusually drowsy. If none of these happen but the cough is no better after two weeks, book a routine appointment.”
The rewrite does not claim more knowledge. It separates the three kinds of uncertainty, attaches clocks and named triggers, and gives the parent a watchable list instead of a worry.
A note for pediatric products
In pediatric content the reader is rarely the patient. A parent is interpreting symptoms they can observe but cannot verify, often with broken sleep and a low threshold for alarm — appropriately so, since young children deteriorate faster and report less. That changes the writing in two ways. First, observable signs beat internal states: “fewer wet nappies than usual” is actionable; “seems unwell” is not. Second, age bands matter: what is watch-and-wait at age six can be same-day assessment at six weeks. Any uncertainty statement that ignores age is incomplete. We set out the wider pattern in seven pediatric risks adult-focused digital-health teams commonly miss, and the pre-launch checks in ten clinical-safety checks before launching a healthcare chatbot include uncertainty wording as an explicit gate.
A review checklist you can reuse
- Does the text distinguish diagnostic, prognostic and action uncertainty, or collapse them into one hedge?
- Is every likelihood word (“likely”, “rarely”, “usually”) supported by evidence the team can point to?
- Does every uncertain statement carry a time boundary?
- Are the triggers for seeking care named, observable and age-appropriate?
- Do the headline and the body use the same level of confidence?
- Could a tired reader, skimming, still find the one action that matters most?
- Does each clinical claim carry a source and a review date?
What a content review is — and is not
Reviewing uncertainty wording is clinical-content review: it checks what the product says against current evidence and good practice in risk communication. Since Montgomery v Lanarkshire Health Board [2015] UKSC 11, UK law has framed risk disclosure around what a reasonable patient would want to know, which is one reason wording choices carry legal as well as clinical weight (https://www.supremecourt.uk/cases/uksc-2013-0136). But a content review is not regulatory approval, certification or a guarantee of safety. Whether a product needs regulatory assessment, and under which framework, is a separate question for qualified regulatory advisers. A review can find unsafe wording; it cannot approve a product.
Frequently asked questions
Why not just tell patients to see a doctor if they are worried?
Because it transfers the clinical judgement to the person least equipped to make it. Safe uncertainty wording names the specific signs to watch for and the time frame for acting, so the patient is observing defined triggers rather than judging their own risk.
Should patient information use percentages or words like ‘rare’?
Either can work, but each needs support. Percentages need a reliable source and a defined population; words like ‘rare’ or ‘likely’ need the underlying frequencies to match. Mixing the two inconsistently across a product is a common review finding.
Is communicating uncertainty the same as a disclaimer?
No. A disclaimer limits legal liability; uncertainty wording is part of the clinical content itself. A product can carry a perfect disclaimer and still falsely reassure patients in its main text.
Does reviewing this wording count as regulatory approval?
No. Clinical-content review checks what a product says against evidence and good practice. Regulatory approval or certification is a separate legal process. A review can identify unsafe wording; it cannot approve a product.
Building a healthcare-AI or pediatric digital-health product? DigitalProved provides an independent review of clinical content, patient-safety risks, escalation advice and user communication. Receive a prioritised written report and recorded walkthrough within five working days. Request a Product Review
