To write alt text with AI, start by explaining why the image appears on the page. A description of everything visible is not always the description a reader needs. Give the AI the image, its surrounding text, and any link destination, then review the proposed wording before using it.
This guide combines a visual review of an actual blog image with fictional image briefs and editorial sample answers. The fictional exercises remain labeled; the visual review identifies the asset and the checks performed. You can use the same process with your own images in an AI tool that accepts image input.
Visual review: an image from this blog
For this update on September 9, 2026, the AI assistant inspected the actual generated background from our thumbnail tutorial. The image below is the reviewed asset. The check records visible features and a context-specific description; it is not a screen-reader test.
Review brief: The image appears in a tutorial explaining where to put a headline. Describe the composition that makes that placement possible. Do not infer an application, a file status, or readable words from decorative marks.
| Detail | Review result |
|---|---|
| Objects | Three blue folders and a white sheet are visible. |
| Placement | The objects occupy the right side; the left half is open. |
| Text | The sheet has horizontal marks, with no readable headline or file name. |
| Unsupported inference | The illustration does not show files being backed up, encrypted, or successfully renamed. |
Reviewed alt text for the layout tutorial: “Three blue folders and a white sheet on the right, leaving the left half blank for a headline.” The empty area matters because the surrounding lesson is about headline placement.
Different context: If the same illustration is only decoration beside a complete text explanation, an empty alt="" may be appropriate. If it is the only content of a link, describe the destination or action instead. This follows the W3C guidance on decorative images; the correct choice depends on how the image is used.
1. Decide what the image contributes
Alt text is a text alternative for an image. The W3C WAI alt decision tree starts with the image’s content and purpose: does it convey information, contain necessary text, or provide a link or control? An informative image needs a brief alternative conveying its relevant meaning. Important words that appear only inside an image need a text equivalent too.
Before drafting, write down the page title, the nearest paragraph, and the reason you included the image. These details give you something concrete against which to judge the answer. “Make this accessible” provides much less direction than “This photo helps readers identify which cable to connect.”
If you are still creating the visual itself, start with our AI blog thumbnail walkthrough. Return to the finished image when writing its alternative; the generation prompt may describe details that did not appear in the final result.
2. Work through an informative-photo example
Fictional image brief: a laptop sits on a desk. A cable connects it to a small external drive. A label on the drive reads “Backup.” A mug and a plant are also on the desk. The photo appears in a guide explaining how to identify the connected backup drive. There is no caption describing that connection.
Weak draft: “A beautiful modern workspace with a laptop, a fast encrypted SSD, a coffee mug, and a green plant.”
This draft introduces speed, encryption, and a specific drive type that the brief does not establish. It also gives space to decorative desk objects instead of the connection the reader needs to understand.
Reviewed alternative: “External drive labeled Backup connected to a laptop by a cable.”
This wording is specific without claiming that a backup succeeded. A connected drive is not evidence of a completed backup. When reviewing an actual image, check that the cable connection and label are visible; do not fill gaps using what you expect the scene to show.
3. Give AI the image and page context
Help draft alt text for the attached image. Page title: [title] Nearby text or caption: [paste exact text] Purpose of the image: [what readers need from it] Is the image a link or button? [yes/no] If yes, destination or action: [verified destination/action] Does the same link or button already contain a visible text label? [details] Use the image and supplied context. Do not infer identities, hidden specifications, causes, or outcomes. If a relevant label is unreadable, flag it instead of guessing. Return: 1. A suggested treatment: informative, decorative, functional, or complex 2. Proposed alt text, or an explicit empty-alt recommendation 3. Details I need to verify Keep the proposed wording concise. Put review notes outside it.
If the tool cannot accept images, provide a verified description and tell it that it is editing your description. Do not ask it to pretend it has inspected a file. The fictional brief above can be used for that text-only exercise.
For a batch of images, give each one a unique identifier and its own context. Require one result per identifier. Review a small sample before processing the rest so that a mistaken assumption does not spread across an entire gallery.
4. Change the treatment when the context changes
The same desk photo used only as decoration: suppose a page uses it as a visual break, with no instructional information depending on it. An empty alt attribute may be appropriate. W3C’s decorative-image guidance explains that alt="" lets assistive technology ignore such images. Omitting the attribute is different; some screen readers may announce the filename instead. Do not type “decorative image” into the field as a substitute for empty alt.
The same photo used alone as a link: suppose selecting it opens a page titled “Backup setup guide,” and the link has no other accessible name. An alternative such as “Backup setup guide” communicates the destination. “Laptop on a desk” does not. If the image shares a link with visible text that already names the destination, repeating that text may be unnecessary. W3C’s functional-image examples distinguish these situations.
This is why a filename-level description should not be copied into every placement without review. Keep a record of the page and image location alongside the approved wording. A later layout change can change what the reader needs even if the image file stays the same.
5. Give charts more than a short label
W3C’s complex-image guidance describes a short identification plus a longer text equivalent for information-rich visuals. A chart’s data and relationships often need nearby text or a table, not an overloaded alt field.
Fictional chart brief: a bar chart compares support requests over three weeks: Week 1 has 12, Week 2 has 18, and Week 3 has 9. There is no information about why the counts changed.
Short alternative: “Support requests by week; values and summary follow.” Use this wording only when the promised explanation actually follows the chart.
Nearby text equivalent: “Support requests: Week 1, 12; Week 2, 18; Week 3, 9. The count rises in Week 2 and falls in Week 3, which has the fewest requests.”
An AI draft saying “The new help center reduced support requests” would add a cause the data does not demonstrate. Ask it to separate the displayed values from interpretation. Check every value against the source chart or underlying data before accepting either the short alternative or the longer explanation.
6. Review the wording and its placement
- Match every concrete claim to the image or supplied context. Remove unsupported names, specifications, emotions, and outcomes.
- Check labels and numbers manually. If text is too small to read, use a clearer source rather than accepting a plausible guess.
- Read the nearby paragraph and proposed alternative together. Confirm that the necessary information survives without unnecessary repetition.
- Confirm that review notes and uncertainty labels have not been copied into the alt field. Resolve important uncertainties first.
- Check the rendered page: the right image should have the approved alternative, and an image-only link should have a meaningful name.
A visual preview alone does not show how an alternative is exposed to assistive technology. Include a screen-reader check in your publishing review where possible. For a deliberately decorative image, verify that the resulting image has empty alt rather than assuming a blank editor field produced the intended result.
Keep the final choice with a short reason: “describes the cable connection,” “decoration only,” or “names the linked guide.” That record makes later reviews faster and gives an editor a clear basis for changing an AI suggestion. The useful output is a reviewed alternative that fits this page, not simply a polished sentence about the picture.