Alt text has been part of HTML since the beginning, and it has been neglected for just as long. The WebAIM Million report, which audits the top one million home pages every year, found in its 2026 edition that more than half of all websites still ship at least one image with no alt text — on average, more than ten images per page that say nothing to anyone who cannot see them. For years that read as an accessibility footnote. In 2026, two deadlines turned it into a visibility problem.
What changed this year
The first is regulatory. The European Accessibility Act became enforceable in June 2026, and image descriptions are part of what it covers. A site selling into the EU now has a legal reason to care about the alt attribute, not just a good-citizenship one.
The second is AI search. The engines answering questions directly — and deciding which pages to cite — read pages the way crawlers always have, and an image without alt text contributes nothing to how they understand the page. The images on a page are part of its evidence; alt text is how that evidence gets entered into the record.
"But AI can see images now"
The obvious objection is that modern crawlers run vision models. They can identify objects in a photo, read text inside a screenshot, and describe a scene without any help from the markup. So why would a text description still matter?
Because a vision model's read is a probabilistic guess, and alt text is an explicit statement. A model might see a person in an office; the alt text says this is the product's founder demonstrating the feature the page is about. When the model's guess and the page's statement line up, confidence in the page goes up — and with it, the odds of being surfaced. The function of alt text has shifted from describing the image to a system that cannot see to confirming the image to a system that can.
The same shift cuts the other way. Keyword-stuffed alt text used to be a cheap tactic with a small upside. Now the vision model can check the claim against the pixels, and a mismatch — alt text promising one thing while the image shows another — reads as low-quality content. Stuffing has gone from mildly useful to an active liability.
What good alt text looks like now
The rules are short, and they have barely changed — what changed is that machines now enforce them:
- Describe what the image actually shows, specifically: subject, key objects, setting. "Website analytics dashboard showing organic traffic growth" works; a string of keywords does not.
- Keep it around 125 characters. Screen readers pause near that length, and AI summaries truncate past it. Detail beyond that belongs in the surrounding paragraph or a caption.
- Match the image's function on the page. A product photo, a diagram, and a screenshot each deserve a description of what they contribute, not a generic label.
- Give decorative images an empty alt (
alt="") so screen readers skip them — that is a deliberate choice, distinct from omitting the attribute.
None of this is hard for one image. The problem is knowing which of a page's forty images are missing it.
Finding the gaps in one pass
Checking alt text by reading the page source is miserable work, and the gaps are invisible in the rendered page — an image without alt looks identical to one with it. Sukat Inspector makes the invisible part visible: open it on any page and it lists every image with its alt text alongside dimensions, file size, and format, flagging the ones where the attribute is missing. The flagged filter narrows the list to just the problems, so a page-by-page cleanup becomes a checklist instead of an excavation. The audit runs locally in the browser — nothing uploads, nothing is logged.
The same pass surfaces the rest of the image story — oversized files, outdated formats, broken sources — so the alt-text fix usually rides along with a size cleanup handled by Sukat itself.
A small attribute with three jobs
One attribute now serves three audiences at once: screen-reader users who depend on it, regulators who have started requiring it, and AI engines that weigh it when deciding what to cite. Half the web still leaves it blank. Run Inspector on the pages that matter most, fix the flagged images, and the page becomes more legible to every one of those audiences in the same afternoon.
Frequently asked questions
Does alt text still matter now that AI can see images?
Yes. A vision model's read of an image is a probabilistic guess; alt text is an explicit statement. When the two line up, confidence in the page goes up. The job of alt text has shifted from describing an image to a system that cannot see, to confirming it to one that can.
How long should alt text be?
Around 125 characters. Screen readers pause near that length and AI summaries truncate past it. Detail beyond that belongs in the surrounding paragraph or a caption.
What should I do with decorative images?
Give them an empty alt attribute (alt="") so screen readers skip them. That is a deliberate choice, distinct from omitting the attribute entirely, which leaves the image undescribed.
How do I find images missing alt text on a page?
Sukat Inspector lists every image with its alt text and flags the ones where the attribute is missing; a filter narrows the list to just those. It runs locally in the browser, with no upload.
About Sukat
Sukat builds free, privacy-first browser tools for compressing images and verifying published content. Everything runs in your browser — nothing is uploaded.
