A discharge summary, radiology report or lab result may contain specialist terms, abbreviations and reference ranges. The safest first step is not to upload it to a chatbot. Keep the original document, use an official glossary for general terminology and write specific questions for the care team that issued or uses the record.
Do not delay contacting the care team or local emergency service while trying to understand a document. If the service marked a result urgent, or you have an urgent symptom or safety concern, use its official clinical or emergency route now. Do not ask AI to decide urgency.
Before you copy, paste or upload anything
Do not upload an identifiable medical document, scan, photo or excerpt to a general-purpose AI service. Removing a name is not enough: dates, record numbers, rare conditions, medication combinations and narrative details can identify a person.
Turning model training off does not separately answer whether content is retained, appears in history, can be reviewed by people, is shared with connected services or is subject to legal access. Use AI with real medical data only where a care provider or organization has explicitly approved a protected environment for that exact data and purpose. Follow its access, retention and review rules.
Otherwise complete the medical document literacy worksheet manually. WHO’s AI-for-health principles emphasize human control over medical decisions, privacy, confidentiality and accountability (WHO).
Step 1: Preserve the exact source text
Work section by section. Copy the exact wording into the first column or refer to a line in the original without transcribing sensitive content into another service. Record the document title, date and issuing organization in your private notes.
Do not replace the original with a summary. A single qualifier such as “possible”, “no acute” or “compared with” can change the meaning.
Step 2: Use official glossaries for general words
Look up general terms in resources such as the MedlinePlus medical-words tutorial, medical encyclopedia and lab-test guides. Record the general definition in the paraphrase column and keep it visibly separate from the document’s finding.
Do not ask AI to expand an abbreviation. The same letters can mean different things across specialties and institutions. Mark the abbreviation unresolved and ask the issuing care team what it means in that document.
Step 3: Do not interpret reference ranges
Reference ranges can differ by laboratory, method and population. A value outside a printed range may or may not indicate a health problem, and a value inside a range does not by itself settle the clinical question. MedlinePlus advises interpreting results with health history, examination, other tests and the laboratory’s own range (MedlinePlus).
Copy the value, unit, flag and printed range exactly. Do not ask AI to decide whether the result is normal, concerning, diagnostic or actionable.
Step 4: Separate four kinds of information
Use four columns:
| Exact original text | General-language note from an official glossary | What remains uncertain | Question for the care team |
|---|---|---|---|
| Preserve wording, value, unit and qualifier | Define a general term only | Mark ambiguity or missing context | Ask one specific question without proposing an answer |
Examples of useful questions include:
- “What does this abbreviation mean in this report?”
- “Which reference range applies to this result?”
- “How does this finding relate to the reason the test was ordered?”
- “What follow-up, if any, do you recommend, and when?”
- “Who should I contact if the situation changes?”
The worksheet is preparation, not a second medical record. Bring the original document to the conversation.
Optional AI practice with a fictional example
If you want to practise turning uncertainty into questions, use a clearly fictional sentence with no real medical facts. Do not substitute text from a real document in a general-purpose tool.
This is a fictional writing exercise with no real patient's information.
Given the placeholder sentence "[FICTIONAL DOCUMENT SENTENCE]", produce
only two questions a reader could ask the issuing care team. Do not expand
abbreviations, interpret findings or reference ranges, infer a diagnosis,
recommend action, add facts or take actions.
The prompt boundary is not a security control. Only the fictional input makes this suitable for a general-purpose service.
When the document needs a professional explanation
Complex pathology and imaging reports, ambiguous abbreviations, interacting findings and any result linked to a care decision need the issuing care team’s explanation. If a glossary does not resolve a general word, leave it unresolved. Do not prompt harder until an answer sounds reassuring.
The NHS notes that abbreviations in records can be difficult to understand and advises discussing concerning results with the GP practice (NHS Digital). AHRQ’s Questions Are the Answer provides a general framework for preparing questions for a clinician.
Common pitfalls
- Uploading a full PDF, photo or scan to a public chatbot.
- Assuming redaction of names makes a medical record anonymous.
- Treating a training opt-out as a complete privacy guarantee.
- Asking AI to expand an ambiguous abbreviation.
- Asking AI to compare a value with a reference range or interpret the result.
- Keeping only the paraphrase and losing the original wording.
- Repeating prompts until the answer becomes reassuring.
Try it today
Choose one section of a document. Keep the original beside you, use an official glossary for general words, mark abbreviations and interpretation unresolved, and write one question for the care team. If the issue may be urgent, stop the exercise and use an official clinical route.
This is document-literacy education, not medical advice, diagnosis, interpretation or triage. The original document and responsible care team remain authoritative.
