A useful UpToDate Expert AI prompt starts with the clinical question you need to resolve. State the relevant population and setting, identify the distinction that matters, and ask for the supporting explanation. A long prompt is useful only when its extra detail changes the question.
UpToDate Expert AI supports natural-language questions, follow-up exchanges, stated assumptions and inline links to its reference content. Those features make it possible to refine a question as your understanding develops. They also give you several ways to ask for clarification before accepting an answer.
The seven patterns below are original editorial templates, designed for adaptation rather than presented as tested product outputs. They also help clinicians formulate questions for other source-linked reference tools, including iatroX. This article is published by iatroX; product documentation was checked on 5 November 2026.
Build a small question before writing a large case
Start with four elements: population, setting, uncertainty and requested output. For example, your uncertainty may concern the scope of a recommendation, the reason two approaches differ, or the findings that would change an assessment. Naming that uncertainty usually improves the question more than adding an entire history.
Use hypothetical or appropriately de-identified information for learning. For real clinical information, use the organisation's approved account and permitted workflow. UpToDate's terms make the handling of protected health information dependent on the applicable organisational agreement and permitted use. An ordinary reference login is not sufficient evidence of approval to paste a patient record.
Choose one of these question patterns according to the work you want to complete.
1. Define the population and setting
Template: For [population] in [setting], explain the recommended approach to [specific question]. State which population and care setting the answer applies to, and link the relevant source section.
This is useful when an otherwise sensible answer could concern the wrong context. A recommendation written for an outpatient population may address a different decision from one made in an acute setting. Similarly, an adult reference question should not silently become a paediatric one.
For a learning example, ask how the assessment framework for a symptom differs between primary care and an acute service. Keep the request focused on the framework you are studying. The aim is to make the boundaries visible before working through details.
After the response, compare the population named in its assumptions with the population in the linked topic. A follow-up can correct the scope directly: Please reconsider this for [the intended setting] and identify which parts of the explanation change.
2. Compare two approaches using a defined decision
Template: Compare [approach A] and [approach B] for [defined situation]. Organise the answer by the decision they address, relevant benefits and limitations, and circumstances in which each is considered. Link the supporting sections.
An open-ended comparison can produce an impressive table without answering your actual question. Decide first whether you are comparing diagnostic roles, monitoring requirements, evidence quality or practical delivery. Ask for those columns explicitly.
For instance, a clinician preparing teaching may want to compare the purposes and limitations of two investigation strategies. Asking for the clinical question each strategy addresses is more informative than asking which is simply best.
Use a follow-up to resolve one row that remains unclear. If a distinction is important enough to influence your interpretation, it deserves more attention than a single table cell.
3. Ask which finding would change the answer
Template: What additional information would materially change this explanation? Identify the missing details, explain why each matters, and distinguish what can be answered now from what depends on them.
This pattern helps expose an apparently complete answer that rests on an unstated assumption. It is particularly useful when your original question intentionally leaves part of a fictional case unspecified.
The important follow-up is why the information matters. A list of possible missing facts can be long; a useful explanation connects a fact with the decision it could alter. Ask the system to prioritise the few details most relevant to your stated uncertainty.
Then supply only the additional context you actually know. A learning exercise can deliberately explore an alternative scenario, but keep that change explicit so that assumptions do not become fictional facts about a real person.
4. Separate the recommendation from its rationale
Template: Explain the reasoning behind [recommendation or distinction]. Separate the recommendation, the evidence or mechanism supporting it, and the important limitations. Link the source passage for the recommendation.
This is a strong pattern for teaching and exam preparation. Recognising a recommendation and understanding its rationale are different achievements. The second makes it easier to explain why a superficially similar scenario may require a different approach.
Ask for the explanation at the appropriate level: an undergraduate overview, a postgraduate discussion or a short account for a teaching session. If the response uses a term you cannot explain, ask about that term before requesting more detail across the entire subject.
A good endpoint is a concise explanation in your own words. That is more useful for learning than retaining an elaborate paragraph you have not worked through.
5. Ask where a claim comes from
Template: Which linked source section supports [specific claim]? Explain how the source relates to that claim, including relevant qualifications or exceptions.
Expert AI's documented inline links and supporting-content panel give this question a practical destination: open the reference section. Ask about a particular substantive statement rather than requesting an undifferentiated list of sources.
This matters when a response mixes a source recommendation with contextual explanation. The source may support the general rule while a detail in the answer depends on an assumption. Reading the relevant section helps you identify that boundary.
If you need a more systematic way to inspect the response, our guide to checking the evidence behind clinical AI answers explains a separate evaluation method. Prompt composition and answer evaluation work together, but complete different tasks.
6. Request a useful uncertainty statement
Template: Where does the available guidance leave uncertainty for this question? Distinguish an evidence gap, variation between recommendations and missing context in the question. Explain what each means for interpreting the answer.
The word uncertainty can hide several different problems. A reference may describe limited evidence, recommendations may differ because they address different populations, or your prompt may omit an essential detail. These situations call for different next steps.
Ask for the relevant distinction rather than a generic confidence percentage. A precise explanation of what remains unresolved gives you something useful to investigate. It also makes a stronger starting point for a discussion with a colleague or supervisor.
7. Turn an answer into a focused learning task
Template: I want to understand [one distinction]. Explain the core idea, then suggest a short self-explanation exercise. Keep the exercise focused on applying the distinction rather than recalling an isolated phrase.
This pattern is useful after the immediate reference question is resolved. Choose one point you are likely to encounter again. Trying to revise an entire topic from a single conversation can obscure the specific misunderstanding that prompted the search.
For example, write down what would make two similar presentations different, then explain which detail carries that distinction. Revisit the source if your explanation depends on a rule you cannot justify.
Where iatroX makes the next step easier
For UK reference and learning, Ask iatroX and the clinical guidance library provide a direct route into source-linked questions and concise NICE guidance. The particularly useful advantage is how that question can connect with deliberate practice.
After identifying the learning need, attempt a relevant question independently. iatroX Tutor then works from the attempted question: explore the reasoning through guided discussion or request a direct explanation. You can keep the resulting insight in an editable basic or cloze card and export it to Anki.
That structure gives a clinician something concrete to do with the answer. Reference resolves what you need to know; an attempted question reveals how you apply it. One iatroX plan includes every paid bank, Tutor, mocks, study planning, available simulations and CPD at £29/month or £99/year in the UK; US$29/month or US$99/year elsewhere.
Up to 10 new sample questions per day, shared across exams and subject to availability, with explanations and no account required.
One complete guided Tutor session is included with a verified free account; continued Tutor use is included in all-access.
For the subscription and product-name distinctions behind Expert AI, see our current UpToDate plans guide.
Frequently asked questions
Is there one best prompt for UpToDate Expert AI?
The useful prompt depends on the question. Name the population, setting, uncertainty and requested output, then use a follow-up to resolve the part that remains unclear.
Are these prompts copied from UpToDate?
No. They are original editorial templates informed by the product's documented question, follow-up and source-linking features. They are not transcripts of a licensed-account test.
Can I use the same question patterns in iatroX?
Yes. Clear scope, a specific uncertainty and a request to inspect the supporting guidance are useful ways to formulate referenced questions. For structured practice after an attempted exam question, use iatroX Tutor.