Using AI in medical presentations without losing trust.

Make the work inspectable.

A polished slide can still contain an unsupported claim. A plausible-looking eye can still imply the wrong anatomical relationship. A clickable chart can still confuse its audience. Evaluate the teaching claim behind each element as carefully as its appearance.

In this guide, “AI slop” describes work released with unresolved factual, visual, or instructional problems after automated production. That is a production failure you can address: keep the evidence, inspect the output, and record the decisions that make it suitable for its audience.

Our approach starts with four questions: What will the learner understand? What supports the claim? What might the visual imply? Who is responsible for the version being delivered?

Give AI a bounded production task.

Useful tasks have defined inputs and an output you can check. Ask for a draft evidence table from supplied papers, alternative teaching sequences for a chosen objective, or chart code for a fixed dataset. Keep missing information visible so it becomes a review task.

A reusable workflow records the input, instructions, draft, corrections, and final artifact. A collection of prompts alone cannot show whether the process produced something accurate or useful.

Choose visuals by their teaching role.

A clinical photograph shows a real observation. A diagram simplifies a relationship. A chart encodes measurements. Decorative artwork sets a tone. Write down which job an asset is doing before commissioning or generating it.

Use approved references and qualified review for scientific relationships. Work with a medical illustrator when the visual requires specialized anatomical judgment. AI can help draft a brief, arrange labels, or implement motion around approved geometry; each resulting state still needs inspection.

Ask a viewer what the image appears to claim without its narration. If a decorative image is likely to be read as anatomical evidence, its original intent does not resolve the ambiguity.

Give the learner something to work out.

A meaningful interaction makes a prediction, comparison, or explanation possible. A reveal animation is useful when it supports that task. Extra motion and extra clicks can also consume attention without helping someone understand.

For a first presentation, use a small set of authored questions and reviewed feedback. Prepare an equivalent static sequence. A live AI system generating clinical answers introduces different review and monitoring needs from a pre-reviewed interactive lesson.

ACCME's September 2026 guidance draws that distinction for its accredited activities. Confirm the requirements of the provider and event you are working with.

Explain the process accurately.

Document what AI contributed and what review actually occurred. Name the source locations behind claims, the origins of assets, the responsible owner, and the version that was tested. Keep unresolved questions out of language that implies approval.

ACCME recommends disclosing substantive AI use in educational content creation, alongside human oversight. COPE's adopted standards address valid content and independence. Those requirements belong to their respective accredited activities; a Pretectal worksheet does not confer accreditation.

For company work, establish whether the activity is promotional, nonaccredited education, or independent accredited education at the start. Have the responsible organization set the applicable review and disclosure process.

Start with one complete segment.

Choose one topic you already teach and one audience. Rework a short segment, test it with intended learners, and ask them to explain the main finding and its limit. Record production effort, review corrections, and comprehension separately.

The BLINK example shows a prediction, a source-linked outcome comparison, and an interpretation check. Its production notes describe what the example demonstrates and what has not been independently evaluated.