Experiment
Lab experiment
Experiment: AI Video Generators for Patient Education
Can AI video generators produce patient education that is accurate, empathetic, and clear? We recreated the same patient education video five ways (traditional motion graphics, Synthesia, Google Veo, OpenAI Sora, and Adobe Firefly) and asked over 100 patient education leaders which would work in their organization. Speed was easy. Trust was the hard part.
Watch the experiments
3 videos. Click any card to play it here.
Experiment
Creating Microlearning with Synthesia: Healthcare Education Example
Example
Patient Education Microlearning: 8 Examples That Drive Real Behavior Change in Healthcare
What question did we test?
We tested whether AI video generators can produce patient education video that is accurate, empathetic, and clear enough for real healthcare organizations to use. Speed is the obvious benefit of AI video. In patient education, though, the bar is patient understanding and trust, so we judged each tool on tone, clinical accuracy, and visual storytelling, not just production time.
New AI video tools appear constantly, and most promise faster production at lower cost. Healthcare teams have a different problem. A video that looks great but shows something clinically wrong is worse than no video. We wanted a fair, side-by-side comparison using one topic and one set of learning goals.
How was the experiment set up?
We recreated the same patient education video five ways: traditional motion graphics, Synthesia, Google Veo, OpenAI Sora, and Adobe Firefly. Then we asked over 100 patient education leaders which version would work in their organization. The comparison video was published in June 2026, after earlier single-tool tests in February and April 2026.
| Version | Tool | Tested |
|---|---|---|
| Baseline | Traditional motion graphics | June 2026 |
| AI presenter | Synthesia | April and June 2026 |
| AI cinematic scenes | Google Veo | February and June 2026 |
| AI generated video | OpenAI Sora | June 2026 |
| AI generated video | Adobe Firefly | June 2026 |
The earlier tests used a real healthcare topic, post-surgical bruising, and the same design process each time: define learning outcomes, write the script, design the visual style, storyboard, generate, then edit and review [VERIFY: confirm the five-way comparison used the post-surgical bruising topic].
What did we build?
We built five versions of one patient education microlearning video, plus two detailed walkthroughs of the workflow behind the AI versions. The Synthesia walkthrough covers presenter-led video with avatars and voiceover. The Google Veo walkthrough covers generating scenes from structured prompts. Both compare the AI output with a manually animated version of the same topic.
The design steps did not change between tools. For Synthesia, we wrote conversational scripts for an AI presenter, selected an avatar and voice suited to healthcare communication, and used text callouts, visuals, and gestures to support understanding. For Veo, we wrote concise, empathetic scripts and storyboarded to control pacing before generating any scenes.
What happened when we compared the tools?
Each tool did something well, and none replaced human judgment. Traditional motion graphics gave the highest control and accuracy but required significant time and resources. Synthesia offered speed and simplicity. The more realistic tools produced compelling scenes, and that realism introduced risk, because convincing visuals can carry clinical errors patients won’t notice. Adobe Firefly struck a balance, pairing AI generation with human design control.
| Approach | Where it stood out | Where it fell short |
|---|---|---|
| Motion graphics | Highest control and clinical accuracy | Significant time and resources |
| Synthesia | Speed, simplicity, multilingual capabilities | Struggles with emotional nuance in sensitive healthcare contexts |
| Google Veo | Visually stunning, cinematic footage | Can introduce subtle clinical inaccuracies; needs visual validation |
| OpenAI Sora | Pushes realism even further | Still lacks the precision needed for regulated content |
| Adobe Firefly | Balances AI generation with human design control, so teams can scale without sacrificing quality | [VERIFY: limitations noted in the video] |
What the 100+ patient education leaders preferred is shown in the video [VERIFY: summarize the survey result and which version was preferred]. The consistent theme across the tests: AI works best as a creative accelerator with human oversight, not as a shortcut past instructional design.
Should you use AI video for patient education? The verdict
AI video for patient education is worth using when the script is clinically approved first, a reviewer checks every visual, and speed or volume matters. Avoid it for procedures where an anatomical or technique error could cause harm, unless a clinician validates each frame. Motion graphics remain the safest choice when precision matters most.
| Use it when | Avoid it when |
|---|---|
| You need many short videos or frequent updates | The video shows a procedure where a visual error could cause harm |
| A calm presenter explaining next steps is the main need (AI avatar) | No clinician is available to review generated visuals |
| Emotive scenes help reassure patients (Veo-style video) | Patients need precise anatomy or step sequences (use motion graphics) |
| Scripts are clinically approved before generation | The script is still being debated |
| You want to test formats with patients before a full build | Your organization’s policy restricts AI-generated health content |
How can you try AI patient education video yourself?
Pick one low-risk, high-volume topic, like what to expect after a procedure, and build it with an AI tool alongside a version you already trust. Lock learning outcomes and the script with clinical review first. Then compare both versions with patients or patient advocates, not just your team.
- Choose a low-risk topic. Aftercare expectations work better than procedure instructions.
- Write outcomes first. What should the patient do or recognize after watching?
- Get the script clinically approved before any generation.
- Pick the tool by need. Presenter-led (Synthesia), cinematic scenes (Veo), or controlled animation (motion graphics).
- Storyboard to control pacing and keep one message per scene.
- Validate every visual with a clinician, frame by frame.
- Test with real audiences and compare against your baseline.
For finished examples from real healthcare clients, see our patient education video examples and healthcare microlearning examples.
What’s next for this experiment?
The next test is whether patient preference matches leader preference. Our survey asked patient education leaders, and the people who matter most are patients themselves [VERIFY: planned patient-facing test]. We also want to track how quickly these tools change, since results from mid-2026 may not hold a year later.
Related work continues in the Lab, including our test of AI voiceover in eLearning. For a broader view of what AI video looks like in finished training, browse our AI-generated training video examples.
What is the best AI video generator for patient education?
It depends on the job. In our test, Synthesia offered speed and simplicity for presenter-led explanations, Google Veo produced cinematic scenes, and traditional motion graphics still gave the most control and accuracy. No single AI tool won on every measure. Choose by the risk of the content, how much realism you need, and who will review it.
Is AI-generated video safe to use for patient education?
It can be, with clinical review. The main risk is realism: a convincing scene can show something clinically wrong, and patients may trust it because it looks real. Every AI-generated visual should be checked by a clinician for accuracy, and scripts should be approved before generation. Keep a human reviewer in the workflow, every time.
How long should a patient education video be?
Short. Patients are often stressed, distracted, or in pain, so one topic per video and a length of a few minutes or less usually works best. Our post-surgical bruising examples focus on a single situation and what to do about it. See our guide on microlearning length for how to size modules to the task.
Should we use an AI avatar or animation for patient education?
Use an AI presenter when a calm, human explanation is the point and the topic is mostly verbal. Use animation or motion graphics when patients need to see a process, the body, or steps in sequence. Many teams mix both. Test with real patients or patient advocates before committing to one style across a library.
Does AI make patient education video cheaper?
It usually makes production faster, especially for presenter-led videos and updates. Total cost depends on how much design, clinical review, and editing the content still needs, and those steps don't disappear. We have not published a cost comparison for this experiment yet [VERIFY: cost or hours per version]. Our microlearning cost guide covers the main price drivers.
Newsletter
Get new experiments first
We test AI tools on real healthcare learning problems and publish the results, including what clinicians pushed back on.
