Alt text is generally 125 characters or less that describes an image so that learners using screen readers can benefit from the content of the image the same way that those who can see the image do. AI can help with writing alt text, but needs guidance to make it effective.

• Prompt: “Write a 125-character alt text for this image.”
• Result: “Several cars are parked in an outdoor parking lot beneath three large green trees against a bright, lightly clouded blue sky.” (Generated by Gemini)

What makes this description lower quality is that it does not consider intent, context or audience. If a prompt includes these considerations, the output is greatly improved. When prompted with more information, Gemini suggested alt text that was tailored to the lesson focus:

• For Urban Forestry & Shade Benefits: Three mature oak trees growing in parking lot medians, providing a dense shade canopy over parked cars on a sunny day.
• For Species Identification (e.g., Willow Oak): Three fine-textured Willow Oaks (Quercus phellos) with dense, light-green canopies planted in an urban parking lot.
• For Planting Environments & Root Space: A row of oak trees planted in narrow soil islands surrounded by asphalt parking spaces under a clear sky.

TIP: If you have multiple images within the same lesson or document, establishing a continuous conversational thread with your chosen AI tool means that you can provide lesson-wide context just once, further increasing your efficiency.

When teaching established concepts using standard graphics, describing the literal visual appearance of the image may not be the best approach. Instead, a clear text description of the underlying process or concept, rather than a description of the graphic, may be better. Let’s take the example image below.

For a complex diagram of the Krebs cycle like the one above, the brief alt text might state: “Complex diagram illustrating the Krebs cycle.” A link to a more detailed text description of the process provides an accessible equivalent if the steps are not covered in surrounding text or transcripts.

The prompt,  “Generate a detailed description of the Krebs cycle, including each step and key molecules,” might be a good starting point for an equivalent text description, and the next step would be to review the output for accuracy and instructional alignment. If desired, you might prompt further to home in on a specific focus area, add observations, or reframe the description from another perspective.

Unique complex images that do not depict standard established processes or concepts require targeted prompting to translate instructional content into clear text equivalents.

Sample Prompt for complex images:

I am providing an image from a college course slide or document. Your job is not to describe the image visually, but rather to translate the instructional content and concepts in this graphic into standalone, fully realized text. A student reading the description should gain complete conceptual mastery without needing to know an image was used to present it.

Context:

• [Enter course & topic]
• [Enter lesson context]
• [Enter learning objective] 

Instructions and guidelines:

• Write direct instructional prose that teaches the underlying concepts shown in the graphic.
• State the central theory, phenomenon, or mechanism directly. Name and define each variable, node, or element, explicitly stating what role it plays in the larger concept.
• Explain how these components interact, influence one another, or flow from input to output in sequential, logical order.
• Do not use visual meta-language like “In this diagram,” “The image shows,” “On the left side,” or “Represented by a red arrow.”
• Convert all visual relationships (arrows, clusters, layers, colors) directly into logical or causal relationships.
• Present any explicit formulas, conditional rules (e.g., “If X increases, Y decreases”), or critical takeaways embedded in the material.”

Longer descriptions can be refined by instructing the AI model to omit information covered in accompanying lecture text. Final descriptions can be integrated alongside the graphic, within lecture notes, or in an accessible document appendix.

One approach is to prompt AI to describe the general concept or features being tested rather than asking it to describe an exemplar image. Ask for key diagnostic traits to ensure the text description provides the necessary details without giving away answers:

• Prompt: “How would a layperson describe a white oak leaf in enough detail for identification using 125 characters or less?”
• Result: “A green tree leaf with 7–9 smooth, rounded, finger-like lobes, no sharp points, and a pale underside.” (generated by Gemini)

This approach ensures that students relying on text descriptions are evaluated on the same domain knowledge as those viewing the visual image. You must review the description to ensure it highlights the features you have discussed in your course, and that it is equivalent to the image you’re providing for sighted students.

In some instances, prompts must explicitly restrict specific terms that reveal assessment answers. For example, this leaf shows a fungus called powdery mildew, but it’s challenging to describe the image without using the word “powder” or “powdery.”

Source: Wikimedia

• Prompt: “Describe the visual appearance of powdery mildew on a leaf without using the word ‘powder'”
• Result: A dull white or pale gray haze on plant surfaces, closely resembling a dusting of flour, fine chalk, or light wood ash.

Static images depicting movement are another example of a case where getting help describing the movement in words is better than asking for an image description. In the example below, asking for alt text generated descriptions that were not usable in a quiz:
Image source: Wikimedia

• Prompt: “Write alt text for this image.”
• Results from Gemini:
  • Image description: A black-and-white line drawing depicting a person’s upper body with their left arm extended diagonally downward from the shoulder. A thick black arrow loops in a circular path around the hand and wrist, illustrating circumduction—the conical movement of a limb around a joint.
  • Alt text: Line drawing of an extended arm with a circular arrow around the hand indicating circumduction movement.

Neither of these results from Gemini are good alt text. The first is a literal description of the image, which is not as aligned with the probably learning objective of knowing terms that describe a movement. The alt text Gemini offered gives away the term.

A better result came from the prompt, “Describe circumduction of the arm in 125 characters or less using layperson terminology.” This resulted in, “It’s swinging your arm in a big circle from the shoulder, making your hand trace a cone shape in the air,” which could be easily refined to create good alt text for a quiz.

It’s not uncommon for screenshots of data tables to make their way into instructional materials. As image of text, these should be converted to true text for accessibility, but these are tedious to recreate manually. AI models can transcribe text from table images into structured formats which can even be copied into spreadsheets as CSV data. However, models may occasionally err or introduce unprompted edits, such as inserting extra header rows or symbols, making human verification necessary.

AI can also enhance image resolution and visual quality. One instructor described the time-consuming task of creating quality graphics himself, and found that AI could greatly reduce the time it took to do so:

• Prompt: “Generate a cleaner version of this image.”
• Result: New image generated by Gemini

AI can assist in establishing accessible color contrast and visual distinction.

• Prompt: Please remediate the color on this graphic, applying high-contrast, WCAG 2.1AA – compliant palette, set a white background, and add distinct patterns or textures to data series so the chart does not rely solely on color for interpretation.
• Result:

Careful verification of AI output remains critical. For example, in one instance an AI tool was prompted to improve color contrast on a grayscale line graph, and generated a visually polished result. Closer inspection revealed inadvertently swapped line color designations mid-series.

When granular data analysis is required, AI tools can extract underlying data points from a chart image to generate an accessible data table. While dense or overlapping graphs may require manual adjustments, simple visual charts are easier for models to digest and represent as tables.

Platforms like Zoom and Panopto use AI to auto-generate video captions. While generally reliable, auto-captioning systems often struggle with specialized technical vocabulary or non-standard accents, substituting everyday words for domain-specific terminology.

AI can help correct and refine auto-generated captions. A typical workflow involves downloading the initial caption file with timestamps, using AI to correct inaccuracies, and uploading the revised caption file back to the video platform.

When using AI for caption editing, providing clear context and instructions is essential.

• Provide a list of technical terms: Supply key vocabulary and acronyms used in the video to prevent default substitutions.
• Address common mis-transcriptions: Instruct the chatbot on recurring phonetic errors specific to your course or platform.
• Decide how to handle filler words: Specify whether to retain or remove spoken filler words based on your editing needs.
• Protect timestamps and sequence numbers: Explicitly instruct the model not to alter timestamps, line sequence numbers, or indexing formats, as changes will break file synchronization.

Sample prompt for correcting captions: “You are a copy editor specializing in accessibility and academic caption correction for an upper-level undergraduate course on [Subject]. Correct the attached caption file and use this list of key terms, acronyms, and proper nouns to resolve phonetically mistranslated jargon. Replace phonetically misheard words with the correct technical terminology based on the context in the glossary. Do not alter, add, delete, or reformat any timestamps or line sequence numbers. Maintain clean verbatim speech. Fix glaring grammatical errors caused by auto-captioning, but preserve the speaker’s natural tone and phrasing. Provide only the corrected caption file contents without introductory text, conversational remarks, or markdown wrappers.”

For recurring course remediation, creating a custom chatbot trained on a course glossary and precise formatting instructions saves time by avoiding the need to re-train generic models in each new chat session. Many paid AI tools provide the ability to create and save custom bots.

When converting caption files into polished, stand-alone transcripts for reading, consider the following parameters:

• Strip metadata: Remove sequence numbers and timestamps to clean up visual clutter.
• Smooth sentence structure: Adjust spoken phrasing into proper written sentence structure without summarizing or losing original wording.
• Decide how to handle audio cues: Decide whether nonverbal descriptions (e.g., background music, laughter) are essential to context or should be omitted.
• Decide how to handle speaker labels: Include speaker identification if dialogue involves multiple participants and tracking attribution is necessary for understanding.

Instructional content presented solely through visual elements must be conveyed verbally during recording or captured through supplemental text. Proactively narrating screen text, diagrams, or animations during presentation recording minimizes the need for separate descriptions.

When existing videos lack adequate verbal narration, one approach is to create an annotated transcript that integrates screenshots and descriptive text or links to extended descriptions.

AI can help with generating these descriptions in much the same way it helps with static images. Supply AI with the video screenshots as well as the surrounding presentation script. An example of this approach is viewable in this example annotated transcript.

The resulting comprehensive transcripts that benefit all learners, including students who are blind or have low vision.

AI models allow instructors to convert screenshots, handwritten notes, or legacy documents into accessible math formats. Proceed with caution, however!

• AI can produce outputs that appear visually correct while lacking underlying accessibility structure.
• Models may attempt to solve equations, alter expressions, or reformat surrounding text unless explicitly restricted by prompt guidelines.

When prompting AI for math remediation, specify precise requirements, such as adhering to WCAG 2.1 AA standards or outputting MathJax-compatible LaTeX or MathML.

Specialized tools designed for math accessibility include:

• Bulk Remediation Tools: Platforms like MathPix, JamA11y, and Grind EQ process entire documents to extract and render accessible math.
• Spot-Fixing Tools: Equatio directly converts selected math images or text into clean MathML or LaTeX.
• Conversion Tools: Pandoc serves as an open-source tool for converting LaTeX source files into accessible web or document formats.

Creating course glossaries manually can be time-consuming, making it an ideal task for AI assistance. For example, in a horticulture course covering numerous plants, an instructor might provide a plant list and prompt AI to generate detailed, precise, pedagogically descriptions of each plant.

• Prompt: Write a 2-sentence visual glossary entry for a tomato for an introductory undergraduate home horticulture class. Use accessible layperson terms focused purely on visual identification. Keep the total entry under 50 words and avoid botanical taxonomy.
  • Sentence 1 (Exterior): Describe key outer features, including shape range, skin/surface texture, color variations, and distinctive physical markers.
  • Sentence 2 (Interior): Describe the cross-section when sliced open, detailing the flesh, core, seeds/pit, and internal layout.
• Result: A smooth, glossy fruit with a star-shaped green stem cap, ranging from tiny cherries to large, ribbed globes in red, yellow, orange, or dark purple. Slicing it reveals a fleshy central core surrounded by juicy, gel-filled pockets packed with small, edible seeds.

The prompt and learning objectives for the course will determine the specific guidelines in the prompt. Key things to note about this prompt, which may be broadly applicable include

• Asking for a certain type of vocabulary. In this case, a person seeing a tomato would not need to know botanical terms to identify the vegetable. Layperson terminology makes an equitable textual experience. Other contexts (like higher-level courses) may rely on more advanced terminology.
• Asking for a specific structural breakdown of sentence 1 and 2. This ensures standardization across the glossary.
• Noting visual identification as the key learning objective. Other learning objectives would require a different prompt (cultural habit, harvesting processes, plant physiology)
• Requiring a strict word count cap ensures a manageable glossary

Once you have a glossary, you can refer to terms within the glossary in your alt text which may help simplify description, Standardized glossary descriptions can also be reused as baseline alt text across course materials, ensuring consistent terminology throughout the curriculum.