ARTICLE
Why Does AI Make Terrible Training Videos?
Why AI-generated videos often fall short in training, where accuracy, consistency, and real-world application matter.
When AI Doesn't Get the Job Done
In employee training, the one constant is change. We're always exploring different ways to present, organize, and format content to achieve the best outcomes for learners. Sometimes that's driven by what works best for the learner. Other times, it's because a client wants to try the shiny new thing.
So when a client wants to pivot from a simple Articulate Rise 360 course to short, AI-generated training videos, we get to work figuring out how to deliver what they want while maintaining the goal of creating training that helps people retain information and put it into practice.
After a couple of days filled with terrible AI-generated videos based on already-written scripts and basic prompts, we started looking more closely at how to write better prompts to get the desired results. When AI gets close, but not quite right, how do we keep it from throwing out the good stuff when we ask for a small modification?
As the frustration mounted, I decided to ask Google, "Why is AI so bad at creating training videos?" (Because honestly, what was being produced wasn't worth the water and energy it took to generate it.)
The answer pointed to three major challenges: accuracy, consistency, and spatial logic. Generative video tools can struggle to reproduce real-world actions, maintain consistent visuals from one shot to the next, and accurately demonstrate processes. In training, even a small visual error can teach someone the wrong thing.
These concerns also come up in discussions about AI video limitations, instructional design experiences, and the challenges of AI-generated video.
Why Training Videos Break AI
Poor physics and spatial logic: Video models predict pixels rather than accurately simulating real-world space or mechanics. Tools can morph mid-action, buttons can shift, and hands or steps can look wrong.
Poor shot-to-shot consistency: Characters, interfaces, and environments often change appearance across clips or even within short generations. Training requires consistent visuals so learners can recognize the exact tool, equipment, or workflow being demonstrated.
Limited precision and control: Prompting rarely provides exact control over camera framing, timing, or step-by-step accuracy. Fixing small mistakes across multiple clips can take more effort than recording the demonstration directly.
Uncanny avatars and generic filler: Synthetic presenters and stock-like AI scenes can feel stiff or distracting, taking attention away from the actual training. Opinions on AI video limitations are mixed, with some people seeing rapid progress and others pointing to ongoing consistency and realism gaps.
Additional perspectives on these issues include Oliver Emberton's discussion of AI video production, IMA Studio's overview of AI video shortcomings, and Quora discussions of the disadvantages of AI-generated video.
These limitations matter more in training than they might in other types of video. A visual that is slightly off in a marketing video might go unnoticed. In a training video, it could teach someone the wrong way to perform a task.
Instructional Design by Real People is the Key
This takes us back to where we started: we still need human expertise and time to do the work.
As instructional designers, we can use AI to help create training videos, but it needs to be an enhancement, not the workhorse. We need to understand what learners need to know and do, determine how best to demonstrate it, and make sure the finished product accurately represents the work.
AI doesn't automatically understand the difference between a visually appealing video and an effective learning experience. That's where instructional design expertise matters.
Where AI Actually Helps
AI can be useful in several parts of the video development process:
Drafting and editing: Use AI to outline scripts, generate rough storyboards, clean up voiceovers, or remove dead air from real screen recordings and camera footage.
Real captures over synthetic demonstrations: Record actual software interactions, equipment, and workflows yourself. Then use AI to assist with captions, pacing, or basic editing rather than asking it to invent the demonstration from scratch.
For examples of these approaches, see this video editing demonstration and this practical discussion of AI video editing.
The key is knowing where AI adds value and where human expertise is still essential.
In the world of training, stakeholders often want training faster. That can lead to pressure to bypass the more expensive and time-consuming process of having people do the work. But as the saying goes, you can have it fast, cheap, or good, but you can't have all three.
AI-generated training videos may offer a faster, less expensive starting point. But getting them right still takes time, expertise, and quality control.