AI Narration in CME: Review the Lesson Learners Will Hear

A clinically approved script and an approved audio lesson are different production artifacts. If AI supplies the narration, the CME production lead needs a review that reaches the recording learners will actually hear. Script approval alone leaves that step unresolved.

AOE Consulting's August 26 question about using an AI voice generator in accredited CE makes the distinction timely. For a team considering synthetic narration, the useful question is specific: what did the tool change, and who approved the resulting lesson?

Start by separating uses. Reading an approved script aloud, rewriting it for spoken delivery, and generating new clinical explanations are different contributions. Record which occurred. Otherwise, a statement that “AI was only used for narration” may conceal text changes that need their own clinical review.

Follow approval through to the audio

ACCME's January 30, 2026 AI guidance keeps responsibility with the accredited provider. For fixed AI-generated materials, it calls for approval by named, qualified people before dissemination, checks for accuracy and bias, and traceability of who reviewed what and when. Applying that guidance to narration means identifying the final output under review, rather than treating approval of an earlier script as the whole record.

The underlying content obligation predates synthetic narration. ACCME's Standards for Integrity and Independence, released in December 2020, require patient-care recommendations to rest on science, evidence, and clinical reasoning and present balanced options. An attractive voice does not settle whether those conditions are met.

For the recording, ask the clinical reviewer to listen alongside the approved script. Check whether medical terms, abbreviations, numbers, units, negations, and qualifications remain unambiguous. A production editor can check pacing and presentation; the person responsible for clinical accuracy needs to confirm that the delivery preserves the intended meaning. These are proposed review steps, not a voice-specific procedure prescribed by ACCME or evidence that synthetic speech has a particular error rate.

There is older production experience to draw on. A 2024 Write Medicine conversation about CME script writing emphasized flow, pauses, and a clear pronunciation guide shared across the team. The discussion concerned human voice work. It supports preparing a script for speech; it does not validate an AI narrator or establish learner outcomes.

If a passage needs a new pronunciation, pause, or wording, regenerate that passage and review it again in the assembled lesson. Keep the approved audio version with the script it implements. The next editor should be able to tell which pair is ready for learners.

Let the learner distinguish the roles

ACCME's same AI guidance recommends disclosing AI use, including its purpose, tool details, and human verification. A generic “AI-assisted” label gives little help when the central question is whether a clinician spoke the lesson, wrote it, reviewed it, or did some combination of those things.

For synthetic narration, explain that the voice was generated and describe the human contribution accurately. If a faculty member reviewed the content, say that; do not turn that review into a claim that they recorded the audio. Identify the tool and relevant version/date information in the activity's accessible production notes, alongside the review details. This is a transparency recommendation, not evidence that a disclosure will increase trust.

Our earlier discussion of human accountability for AI-assisted writing asked who could take responsibility for the work. Narration adds a delivery artifact to that question. The accountable person still needs to know what the learner received.

Before release, make the clinical reviewer and accreditation lead's decisions concrete: which final audio was checked, whether the narration preserved the approved content, and whether the explanation of AI and human roles is truthful. A convincing voice can deliver a lesson. It cannot approve one.