An AI tool describes your article image as “a beautiful handmade botanical artwork symbolizing growth and mindfulness.” The sentence is fluent, but the model has added a manufacturing claim and an interpretation. Neither tells a reader what the image contributes to this particular page.
This worked example complements our broader accessibility editing guide. The task here is narrower: take one image description and make an editorial decision about what belongs in its alternative text. The sample AI drafts below are invented teaching examples, not recorded model outputs.
Start with the actual image and the page
Look at this article's hero: a yellow ginkgo leaf sits inside a round blue frame, with a glass paperweight nearby. It is an AI-generated illustration presented in a photographic style. It is not evidence that a particular physical product exists or that an artist used a particular process.
Before asking for a rewrite, write one sentence explaining the image's role. On this page, it supplies the visual subject for an editing exercise. On a framing tutorial, the same composition could illustrate the frame's appearance. On a generic writing announcement, it might merely provide decoration.
The W3C alt decision tree makes purpose and context central to the choice of text alternative. The examples below apply that principle; they are editorial judgments for the stated situations, not universal wording to copy onto every image.
Context one: the image is the subject of discussion
Invented draft: “A beautiful handmade botanical artwork symbolizing growth and mindfulness, elegantly displayed in a premium frame.”
Edited alternative: “A pressed yellow ginkgo leaf in a round cobalt frame stands beside a clear glass paperweight on a rust-red tabletop.”
The revision replaces evaluation with observable content. “Beautiful,” “elegantly,” and “premium” advertise a reaction or quality level. “Handmade” claims a production method the image cannot establish. “Growth and mindfulness” assigns meaning that the surrounding article has not supplied.
Keep a species name only when it is reliably identified in the source context or by someone qualified to identify it. Here, ginkgo was specified in the illustration brief and is visibly consistent with the intended fan-shaped leaf. Without that information, “yellow fan-shaped leaf” would be a more careful description than an unsupported identification.
The colors and nearby object are useful because readers will compare the actual composition with these sample descriptions. This does not mean every image needs every color, background object, or material named. Relevance is doing the selecting.
Context two: the page is about the frame's appearance
Imagine the same illustration in a tutorial about choosing a round frame for a botanical display. The paragraph already explains the leaf and paperweight. The detail the image adds is the relationship between the blue rim and the pale backing.
Possible alternative: “A round cobalt frame surrounds a yellow leaf on pale backing.”
That version removes the paperweight and tabletop because they do not help with the stated task. It keeps shape, color, and contrast. It does not add dimensions, glass type, mounting hardware, price, or claims about durability. Those details would need a separate verified source.
If this were a real product listing, use an accurate photograph of the actual item and verified product information. A generated illustration should not quietly become a product photograph merely because its alternative text sounds plausible.
Context three: the image adds decoration only
Now imagine the picture at the top of an announcement whose text already communicates everything readers need. The leaf and frame do not illustrate a point, identify a product, or perform an action.
For a purely decorative HTML image, the W3C images tutorial specifies an empty alternative, written as alt="". That is an intentional decision. It differs from omitting the attribute and hoping the browser or publishing system supplies a useful fallback.
Do not put the words “decorative image” into the alt field. Also check whether the image is a link: a linked image may need an alternative that communicates the destination or action when there is no other accessible link text. Its function can change the decision.
Give the AI editor a bounded assignment
A useful request is: “This image accompanies a discussion of its visible composition. Shorten the following description. Retain the leaf, frame shape, and relevant colors. Remove aesthetic judgments and unsupported claims. List any detail that needs human verification separately.”
If the tool cannot see the image, give it a description you have checked. It cannot verify a material, read a label, or identify an object from wording alone. If it can see the image, still compare the proposed alternative with the actual asset; a confident visual description is not a verification record.
You can use the AIUndetectable humanizer to explore concise wording, then check every retained detail yourself. A smoother sentence or a detector score does not establish that an alternative serves its reader.
Read the alternative beside the surrounding text
Before publishing, read the paragraph, caption if present, and proposed alternative in sequence. Remove unnecessary repetition while preserving information a reader would otherwise miss. A caption and alt text can serve different purposes; neither automatically replaces the other.
Finally, confirm that your publishing system actually saves the intended value. A filled-in field is not proof of useful text, and an empty field is not proof of a deliberate decorative-image implementation. Your finish line is a faithful alternative for the image's role on this page.