From clinical language to a clear explanation

Create patient education materials with AI: a brief, a draft and a clinical edit

A practical template for clinician-reviewed patient information: define the audience, preserve approved facts, create a visual brief and check translations.

The short answer

Start with approved source material and a specific audience. Use AI for the first draft and visual ideas, then check the meaning, instructions and translation before handing the material to a patient.

12 October research: test the language you will actually publish

A 12 October 2026 scoping review in Communications Health mapped 806 studies across 62 non-English languages and found uneven evaluation coverage. Its search ended in February 2026. We reviewed the publisher’s accessible abstract and excerpts, not the complete methods; this is not a MedSafeAI evaluation.

Here is our practical editing exercise: take one approved, non-identifying paragraph and specify the target language, country and audience. Before generating anything, mark the facts and qualifications that must survive. Keep the source version fixed when comparing drafts. The fictional examples below are checks of meaning, not instructions to give a patient.

Fictional source wordingReject this changeWhat the reviewer checks
Bring your medication list if available.Bring your medication list.The condition ‘if available’ remains explicit.
A review is planned in two weeks.You must attend tomorrow.The planned timing is preserved; no appointment is invented.
The benefit remains uncertain.This will help you.Uncertainty has not become a promise.
Use the clinic contact details supplied with this sheet.A generated phone number or local service.Contact details come from the approved source, never inference.

Request an inspectable translation, then record corrections

Try this prompt with a short source you are authorized to use: ‘Translate only the supplied text into [language and country variety] for [audience]. Preserve every number, time interval, condition, negation and statement of uncertainty. Do not add advice or local contact information. Return the draft, followed by a table with source passage, translated passage and any term needing human review. Mark missing information instead of filling it in.’

Ask a suitably qualified reviewer who understands both the language and the clinical content to check each row against the original. A back-translation or another model’s agreement can identify questions to inspect, but cannot approve the document. Log the exact correction, reviewer, source version and review date. If the source changes, reopen its translations for review.

Start with one approved paragraph in MedSafeAI’s AI Library. Use an available model for the draft; Ask All AI and Consensus can help expose differences between candidate phrasings. Check the original responses and the fixed source yourself. Keep a draft only after the responsible team approves it in its intended language. Track corrected meaning errors and editing time separately; we have not tested this exercise as a clinical benchmark.

October 2026 example: check understanding, not just wording

Published on 7 October 2026, AYA-RISE combined clinic visits with researcher-written chatbot content and improved knowledge of cancer risk by age 30. Its materials did not use generative AI. This is not a MedSafeAI evaluation.

Our proposed drafting exercise: take one approved explanation and identify the single distinction a reader must retain. For a numerical risk statement, that may be the population, denominator and time horizon. Create a short explanation and an optional detail panel, then compare both against the same source. These are editorial checks we propose, not a reproduction of the trial or a validated intervention.

Try in MedSafeAI with a document in AI Library: “Using only this approved passage, draft a short explanation and an optional detail box for [audience, language]. Keep every number, denominator, time horizon and uncertainty unchanged. Add three questions that invite the reader to explain the message in their own words, with a source-linked answer key for the clinician. Flag missing information. Do not invent risks, screening instructions or evidence that this draft improves understanding.”

  • Check the answer key before the draft: can every expected answer be traced to the approved source?
  • If a reviewer cannot distinguish two time horizons or identify whom a number describes, revise that sentence before polishing the layout.
  • Use the questions to discuss understanding with an appropriate reviewer or reader; an AI-generated answer to them is not a patient comprehension test. Keep a named clinical reviewer and version date for the final material.

Define the reader before requesting the text

‘Explain this simply’ leaves too many choices to the model. State who will read the material, what they already know, their language and the one idea you want them to understand. A one-page explanation before an appointment has a different purpose from a poster for a waiting room. Choose the format before generating the draft.

Provide a clinician-approved fact sheet, guideline passage or institutional document. Identify the facts and instructions that must remain unchanged. MedSafeAI’s AI Library can keep your source documents available for this work, while specialty assistants help you frame the topic. Ask for unclear or missing points to be returned to the clinician, not completed by invention.

Fill in this short production brief

This template makes the assignment concrete. Avoid identifying patient details when a general description will do. If the material includes individual care instructions, those instructions must come from the treating team and remain subject to its review.

Brief fieldWhat to specify
AudienceLanguage, approximate reading level and relevant prior knowledge
PurposeOne thing the reader should understand after reading
Approved sourceDocument title, date and the sections to use
Fixed informationStatements, numbers and clinician instructions to preserve
FormatFor example: one page, short headings and one explanatory diagram
ReviewThe clinician or team responsible for the final version

Ask for a draft and a checking table together

‘Using this approved source, draft a patient explanation for [audience] about [topic]. The main message is [message]. Use short paragraphs and explain necessary medical terms. Preserve the listed facts and clinician instructions. Do not add treatment advice, doses, warning signs or contact details that are absent from the source. After the draft, provide a table linking factual statements to source passages, and list questions that require clinician review. Length: [limit].’

Read the draft aloud and remove wording that sounds reassuring but says little. Check every number, conditional statement and instruction against the source. The CDC Clear Communication Index offers a structured way to assess communication materials; use it as an additional review aid. A requested reading level is a drafting goal, not proof that every reader will understand the result.

Give the illustration one explanatory job

MedSafeAI can create medical visuals. Begin with a visual brief: ‘Explain [one relationship or process] in a simple labelled diagram for [audience]. Include [essential structures]; omit decorative elements that distract from the message. Supply a caption and a short text description.’ Keep the diagram tied to the approved explanation rather than adding a new clinical claim through an image.

Review anatomy, labels, arrows and proportions. Check that the visual does not imply a guaranteed outcome or a treatment instruction the text never makes. Retain a readable text explanation so the material remains useful when the image is small, unavailable or inaccessible to the reader.

Review each language as its own patient document

Translate the reviewed source text and specify the country or language variety when terminology differs. A second model can flag possible changes in meaning, but it does not replace a qualified language review. Have a suitable reviewer check important instructions, familiar wording, numbers and local contact details. Save the source version and translation dates together. Publish or hand out only the version the responsible clinical team has reviewed.

Sources & further reading

Follow the original sources to check the details. Product pages describe their own products; they are not independent clinical evaluations.

  1. Communications Health: non-English clinical AI evaluation review (12 October 2026) nature.com
  2. CDC: Clear Communication Index cdc.gov
  3. MedSafeAI: medical AI workspace and product features medsafeai.com
  4. JAMA Network Open: AYA-RISE randomized trial (2026) jamanetwork.com

For healthcare professionals. AI responses and agreement between models do not replace source verification or clinical judgment.

Keep exploring

Journal

PUT IT INTO PRACTICE

Bring your next question to MedSafeAI.

Explore the models, compare perspectives and keep the evidence in view. Use MedSafeAI on the web, iPhone, iPad and Android.

See plans