AI for image handling: tagging, resizing and alt text
Which parts of managing images on a website and in a catalogue are now automated well, where AI proposals need a person, and how to set the flow up.
The short answer
Managing images on a website and in a catalogue is two kinds of work. The mechanical kind, resizing, converting to modern formats, compressing, cropping to a focal point, generating the sizes every device needs, has been solved automation for years and should never be done by hand. The descriptive kind, alt text, tags, captions, finding the right image in a library, is where AI has changed things: a model describes what is in an image well enough to draft alt text and tags, and to make a library searchable by what the pictures contain. What the model does not know is why an image is on a particular page, which is exactly what good alt text conveys, and alt text is an accessibility requirement with legal weight for many organisations. So the flow is: upload one original, process everything mechanical automatically, let AI draft the descriptions, have a person approve them with the page’s purpose in mind, and serve the right size and format to every visitor.
Mechanical and descriptive work
| Task | How it is done | Person needed? |
|---|---|---|
| Resizing to every needed size | Pipeline, automatically, from one original | No |
| Modern formats and compression | Pipeline | No |
| Cropping to a focal point per layout | Pipeline, with the focal point set once or detected | Set the focal point on key images |
| Lazy loading and responsive delivery | The site’s image component | No |
| Alt text | AI drafts; person approves against the page’s purpose | Yes, approval |
| Marking decorative images | Person decides; AI can suggest | Yes |
| Tags and categories for a library | AI proposes; person confirms in bulk | Light review |
| Captions | AI drafts; person edits | Yes |
| Searching a library by content | AI-generated descriptions indexed | No |
| Duplicate detection | Automatic | No |
| Catalogue consistency: backgrounds, crops | AI-assisted with review | Review |
| Content moderation for user uploads | AI flags; person decides | Yes, on flags |
Setting up the flow
- One original upload into the library, never resized by hand.
- Automatic processing: sizes, formats, compression, focal-point crops.
- AI drafts alt text, tags and a caption on upload.
- A person approves alt text on every published image, asking what it is for on this page, and marks decorative ones.
- Bulk-confirm tags for the library; correct the wrong ones.
- Serve responsively through the site’s image component.
- Audit periodically: images without approved alt text, oversized files, duplicates.
Why it matters beyond tidiness
Alt text is what a blind visitor hears and what search engines read; it is required by accessibility standards that apply to many organisations by law. Image weight is where page speed is won or lost. A searchable library saves editors hours and stops the same photo being uploaded repeatedly. And consistent catalogue images sell more. The mechanical automation and the AI drafts make all of that routine rather than aspirational.
What this means for you
Let the pipeline do everything mechanical with images and let AI draft the descriptions, then approve alt text with the page’s purpose in mind and mark decorative images. Make the library searchable by content and audit for gaps. The site gets faster and accessible, editors find images in seconds, and nobody resizes a photo by hand again.
Frequently asked questions
Can AI write our alt text?
It can draft it: a description of what is in the image, often accurate and sometimes better than what people write when rushed. What it cannot know is why the image is on the page, which is what good alt text conveys. A photo of a technician on a roof might need the alt text to say what the technician is doing for this article, or be marked decorative if the text already says it. The draft saves time; the approval, by someone asking what the image is for, makes it right.
What about resizing and formats?
Entirely automatic and has been for years. The pipeline takes one uploaded original and produces every size and modern format the site needs, with the browser choosing the right one per device. Nobody should be resizing images in an editor or choosing formats by hand; if they are, the site is missing a basic piece of its build.
How does AI help with a large image library?
By making it searchable by content rather than by filename: find the photos with a van, the product shots on white, the images with people wearing our uniform. Tags and descriptions generated once let editors find the right image in seconds, and duplicate detection finds the same photo uploaded six times. For a catalogue, consistent crops and background handling can be automated with review.
Sources
- W3C Web Accessibility Initiative: An alt Decision Tree (accessed 2026-09-12)