Start by asking what the image contributes on this page and in what context. It belongs to a larger section, where it might help a visitor choose a product, follow an instruction, or understand an idea. Describing it on its own can miss that purpose.
Alt text, short for alternative text, conveys an image’s information or purpose, including for people using screen readers. It isn’t the same as the visible caption beside a photograph.
For example, a photograph of someone tightening a backpack strap might explain a step beside fitting instructions. The same photograph could link to those instructions or add atmosphere to a campaign page. Its subject hasn’t changed, but the text alternative may need to.
Write the information or action a visitor would otherwise miss, rather than an inventory of everything visible. WAI’s alt-text decision tree is a useful reference when the right treatment isn’t obvious.
AI can propose wording, but someone who knows the page must choose the purpose, verify the details, and check how the text is implemented. Agree on that standard before generating a batch of descriptions.
Decide What the Image Does Here
Read the surrounding section’s heading and copy, the caption, and any link text. Ask what the section helps the visitor understand or do, then decide whether the image adds information, performs an action, or repeats meaning already available. Confirm the actual destination of a linked image rather than guessing from its appearance.
Keep alt text concise enough to communicate the relevant meaning. You don’t usually need “image of” or a description of every background object. Don’t guess a person’s identity, emotions, or product properties from appearance.
The examples below are fictional, with their context stated. Their wording is a starting point to adapt, not text to paste onto a different image.
Describe Product Details That Matter to the Choice
Suppose a product photograph shows a fictional Harbor backpack’s roll-top opening and two side bottle pockets. If those are the useful details the photograph adds, a candidate is: “Harbor backpack with roll-top opening and two side bottle pockets.”
“Backpack, bag, travel gear, hiking equipment” adds keywords without conveying that detail. Don’t add “waterproof” or a capacity unless a trusted source establishes it and it belongs in the alternative. A photograph alone doesn’t prove those properties.
Check the nearby description too. The text alternative should preserve the image’s useful contribution without unnecessarily repeating information already available in the same context.
Describe an Action—or Name the Link’s Action
For an unlinked instructional photograph, “Pulling the chest-strap webbing to tighten the fit” can convey the demonstrated step. Check that this is the action actually shown and that it fits the surrounding instructions.
If the photograph is instead the only content of a link to fitting instructions, “Backpack fitting instructions” describes its destination. If adequate text already names that same link, the image may need an empty alternative to avoid repetition, provided it adds no other essential information.
A logo works similarly. An unlinked logo reading Harbor can be “Harbor.” If it alone links to the homepage, “Harbor home” identifies the destination. WAI’s functional-images guidance explains why purpose matters more than the shape of an icon or photograph.
Treat Decorative Images Intentionally
A generic mountain photograph used only for atmosphere may add no information to a service page. On a trail-condition page, the same subject could be important evidence. Decide from its use, not from the fact that it is scenery.
For a decorative HTML image, an intentionally empty alternative is different from a missing alt attribute. Missing alt can lead to a filename being announced instead. Don’t enter the word “decorative” as the description.
Use the documented decorative-image treatment in your publishing system. A blank field alone doesn’t prove the correct result, so ask the developer or accessibility specialist to verify the output if uncertain. WAI’s decorative-images guidance covers that distinction.
Give Charts and Important Words Enough Room
Complex images such as charts may need a short identifier plus a fuller accessible explanation of their essential data, relationships, and qualifications. “Chart showing growth” isn’t enough if readers need values, units, or the pattern to understand the point. Use the underlying data rather than asking a model to guess unreadable numbers.
Provide an accessible description or table alongside the chart where appropriate, preserving any essential meaning a table alone wouldn’t explain. WAI’s complex-images guidance shows approaches. Ask a specialist to check how the short and full alternatives work together.
If a graphic says “Applications close Friday” and that information appears nowhere else, provide those words as text. Prefer ordinary page text for important messages when possible. A photographer-credit caption doesn’t replace an informative description of the image.
Use AI for an Optional First Draft
You can write the alternative yourself using the examples above. If you want AI assistance, use a service approved for the work and confirm permission from your organization or client before sharing the image and its context.
Don’t submit customer information, confidential material, or unreleased business details as a casual demonstration. Consumer accounts, managed workspaces, and APIs can have different data rules.
These consumer apps document image inputs:
- Claude: its file-upload help describes image attachments. Check your account’s supported formats and current limits.
- ChatGPT: its image-input FAQ explains image uploads and warns that descriptions can be incorrect.
- Standalone Grok: the website and apps FAQ documents image understanding. This isn’t a claim about Grok inside X.
Check the applicable Claude privacy guidance, ChatGPT data controls, or standalone Grok consumer guidance. Turning off model training doesn’t mean zero retention or no authorized human access. Don’t assume deleting a chat deletes separately saved files.
Agree on a Reusable AI Drafting Standard
Use a short brief like this with an approved image. Replace the bracketed details and provide verified context rather than asking the model to infer the page’s purpose:
“Draft a text alternative in [language]. The page helps [audience] do [task]. The image’s purpose is [information or action]. Nearby text says [copy and caption]. The image is [unlinked, or linked to this confirmed destination]. Relevant verified details are [details].
“Preserve the useful meaning without repeating nearby text unnecessarily. Don’t invent identities, emotions, product claims, or unreadable values. Ask if the purpose is unclear. Flag uncertainty separately from the proposed wording. If decorative treatment or a fuller description may be needed, explain why instead of forcing a caption.”
Keep the standard concise, but don’t impose an arbitrary character limit that removes necessary meaning. WAI’s writing tips can help the team judge the result consistently.
Check the Proposal Before Publishing It
Compare every assertion with the image, trusted product information, and actual page context. Verify the action or destination with the page owner, remove invention and needless repetition, and check that complex information has an adequate full equivalent.
Then ask whoever maintains the site to confirm that the publishing system produces the intended alternative, including intentional decorative treatment. A polished AI sentence is a draft to review, not an accessibility certificate.
For an uncertain result, send the page address, image location, proposed wording, and intended purpose to whoever maintains the website. Ask: “Will a screen reader receive this wording for this image or link, and will images marked decorative be ignored rather than announced by filename?”
