Can AI Write Alt Text for eLearning Images?
Your course has 180 images, every one needs alt text before it can pass an accessibility review, and nobody budgeted time for it. Somebody suggests letting a tool write them, while somebody else says that will never meet standards. Both have a point, and the truth sits in the middle: AI alt text is useful, but only if you know what to feed it and what to check. Can AI write alt text for eLearning images well enough to use?
Yes, AI can write good first-draft alt text for most eLearning images, but a person who knows the course still needs to review every one. Multimodal assistants such as ChatGPT, Claude and Gemini can look at an image and describe it accurately, and Microsoft Office apps can suggest alt text automatically. What they can't know on their own is why the image is on the screen. Good alt text depends on purpose and context: the same photo of a forklift needs different alt text in a safety lesson than on a welcome screen. Give the tool the context, mark decorative images yourself, and review the drafts against the learning objective.
What Makes Alt Text Good in a Course?
Good alt text gives a screen reader user the same information a sighted learner gets from the image, in as few words as possible. Under the Web Content Accessibility Guidelines (WCAG), non-text content needs a text alternative that serves the equivalent purpose, and that word "purpose" is the whole game.
WebAIM's guide to alternative text sums it up well: think about the content and function of the image, not just its appearance. A few rules follow from that:
- Skip "image of" or "picture of." Screen readers already announce that it's an image.
- Describe what matters for the lesson, not every detail.
- If an image is a button or link, describe what it does.
- If an image is purely decorative, mark it decorative so screen readers skip it.
- If the image contains text, include that text.
How to Prompt for AI Alt Text That Fits the Lesson
The best AI alt text comes from prompts that include the lesson context, not just the image. Without context, the tool describes what it sees, but with context, it describes what the learner needs.
Here's a simple process for each image or batch:
- Give the course audience and the learning objective for the screen.
- Paste the on-screen text or narration that appears with the image, so the tool doesn't repeat it.
- Ask for short alt text (many teams aim for roughly 125 characters or fewer), or a short alt plus a longer description for complex images.
- Ask it to say "decorative" if the image adds nothing beyond the text.
- Ask it to note anything it can't identify with confidence.
A sample request: "This screen teaches new forklift operators to check the load backrest before lifting. Narration already says to inspect the backrest. Write alt text for this photo in under 125 characters, focused on what the learner should notice, and tell me if anything in the image is unclear."
A typical result might be: "Forklift with a bent load backrest, the top bar visibly pushed back on the left side." That beats "A yellow forklift in a warehouse," which is accurate but useless for the lesson.
Where AI Alt Text Goes Wrong
AI drafts go wrong in a few predictable ways, and each one has a quick fix. This table is the checklist I'd hand a reviewer:
| Image type | Common AI tendency | Reviewer fix |
|---|---|---|
| Photos of people at work | Guesses age, mood or identity | Describe only what matters to the task |
| Equipment and parts | Uses a generic name or the wrong part name | Swap in the exact term the course teaches |
| Charts and graphs | Describes the colors, not the finding | State the key trend and put data in a table nearby |
| Screenshots | Lists every menu item | Name only the control the learner should use |
| Decorative art | Writes a long, poetic description | Mark it decorative |
| Images with text | Misreads small or stylized text | Copy the text from the source file |
None of this is a reason to avoid the tools, since reviewing and correcting a draft is usually much quicker than writing from a blank field, especially across hundreds of images.
What About Complex Diagrams and Charts?
Complex diagrams and charts need a short alt text plus a longer description somewhere the learner can reach it. A single line can't carry a process flow with eight steps or a chart with twelve data points.
AI tools are genuinely helpful here. Ask for two outputs: a one-sentence alt text that names the diagram and its main point, and a structured long description, such as a numbered list of steps or a small data table. Put the long version on the same screen, in a transcript panel or in a downloadable text version. Then have your subject matter expert (SME) check every value and step, because a misread number in a chart description is an accuracy problem, not just an accessibility one.
What About Confidential Images?
Treat images the same way you treat text: only upload them to tools your organization has approved for that content. Training images often show more than people realize, such as unreleased equipment, a customer name in a screenshot or a badge on a lanyard.
A quick pre-check helps. Crop or blur anything sensitive before the batch goes out, and keep proprietary diagrams in an approved enterprise tool. If an image can't leave your network at all, write that alt text by hand, which usually affects only a small share of the list.
A Workflow for a Course With Hundreds of Images
For a large course, batch the work and split the roles. Here's a workflow that holds up:
- Export an image list with screen numbers, on-screen text and the lesson objective for each screen.
- Mark decorative images first. This often removes a large share of the list.
- Run the remaining images through your AI tool in batches, with the context from step one.
- Paste drafts into a review sheet next to a thumbnail of each image.
- Have an instructional designer review for purpose, and an SME check any technical terms.
- Add the approved alt text in the authoring tool, then test a sample with a screen reader such as NVDA or VoiceOver.
Keep the review sheet, because when the course is updated, you'll already have approved alt text for every image that didn't change.
Video gets the same treatment. If your course includes narrated clips, the review habits from this workflow carry straight over to AI captioning, where the draft is fast and the cleanup is where quality happens.
AI Alt Text Is a Fast Draft With a Human Finish
AI alt text takes the slowest part of image accessibility, writing the first description, and turns it into a quick batch job. The tools interpret images well, while your team supplies the purpose, the exact terminology and the final judgment.
Pick the most image-heavy module in your current course and run it through the workflow above. Give the tool real context, review every draft against the objective, and test a few screens with a screen reader. You'll likely find the work goes noticeably faster, and the alt text gets better because the reviewer finally has time to think about purpose instead of typing descriptions from scratch.