AI for Patient Education Materials

Health literacy experts and organizations such as the AMA and NIH commonly recommend writing patient materials at about a sixth- to eighth-grade reading level. That is because typical hospital handouts read well above what many adults can easily follow. Two free tools help judge quality: the CDC's Clear Communication Index, and AHRQ's Patient Education Materials Assessment Tool (PEMAT), which scores understandability and actionability.

A sound AI workflow: start from vetted content, such as your institution's guidelines or clinician-written text, and tell the model to use only that source; Specify the audience, the reading level and the format: headings, short bullets, one idea per sentence, and clear actions such as 'Call 911 if...'; Measure readability with a formula such as Flesch-Kincaid or SMOG; and Have a clinician fact-check every dose, threshold, warning sign and timeline. Simplifying can quietly turn into inaccuracy. The model might drop 'unless your doctor told you otherwise', add 'take with food' when that isn't required, or blur an urgent warning sign into general advice.

Readability formulas count sentence length and syllables. They cannot judge accuracy, tone or cultural fit, and a text full of short words can still confuse readers.

For other languages, machine translation quality varies. It is generally weaker for languages with less training data, and it is riskiest for dosing instructions and medical terms. In the United States, the 2024 rule under Section 1557 of the Affordable Care Act says that when covered entities use machine translation for critical content, a qualified human translator must review it. Back-translation and testing with native speakers catch further problems.

Two misconceptions are common. The first is that a lower reading level means 'dumbing down'. In fact, clear writing helps readers at every literacy level. The second is that AI translations are ready to hand out without review.