Alt Text Checker for Page HTML

An alt text checker has three states before it has an opinion

Missing, empty and present are different source facts. An omitted alt attribute gives assistive technology no author-supplied alternative and can expose a filename or URL. alt="" is deliberate: it asks assistive technology to ignore the image. Non-empty text proves only that characters exist. It does not prove those characters replace the image’s information or function.

That distinction follows W3C’s current guidance. Decorative images use an empty alternative, while informative images need text that conveys their meaning. For a linked or button image, the alternative names the action, unless text already inside the same control does that job. Only the author can decide what role the image plays in its exact context, so the report never awards a pass for the mere presence of text.

View image alt text on the current page without crawling it

The bookmarklet reads the DOM already open in your browser. It inventories <img>, image inputs and image-map areas, including their source URL, alt state, declared dimensions, loaded image dimensions, rendered size, hidden state and whether an enclosing action exposes other text. It also records headings and a bounded set of meta entries. The collector shows that scope before sending anything.

It does not fetch another URL, log in, open pagination or inspect rows and components the page has not rendered. It sends no body paragraphs, form values, cookies, screenshot or image bytes. Image URLs are printed as text in the report rather than rendered, so opening the checker does not leak a new request to the source site.

What can be flagged mechanically, and what cannot

The deterministic findings are intentionally small: a missing attribute, an empty image action with no other source name, a generic placeholder, a filename used as alt text, or a value so long that a separate detailed description deserves consideration. Empty alt on an otherwise ordinary image is a review item, not an error, because decoration is a human judgment. Normal non-empty text is also a review item: the checker cannot see what the pixels mean without fetching or interpreting the image, and even an image model would not know why the author placed it there.

This is consistent with WCAG 2.2’s Non-text Content explanation: alternatives serve the equivalent purpose, and decoration is implemented so assistive technology can ignore it. The page makes no automated WCAG-conformance claim and does not generate replacement text with AI.

Heading order and meta tags are supporting evidence, not a score

Headings are shown in source order. An empty heading and a jump such as h2 directly to h4 are marked for review; moving from h4 back to h2 can correctly close subsections and is not marked. W3C advises nesting headings by rank and avoiding skipped ranks when opening a subsection, while also noting that structure and purpose—not a mechanical sequence alone— are what matter in the document.

Title, canonical and meta name/property entries are extracted without silently choosing between duplicates. That is useful evidence but not a search-preview or SEO verdict. A pasted fragment may not contain its page head at all, and the report says exactly what was supplied rather than treating absence from the fragment as absence from the live page.

What this report does not test

It does not test contrast, focus order, keyboard operation, responsive reordering, accessible names computed after all ARIA relationships, screen-reader output, captions, form labels, language changes or the rest of WCAG. A real accessibility review combines automated checks, keyboard and assistive-technology use, and human judgment. This tool is the image-source ledger for that process, not a substitute for it.

The CSV contains every image finding and next action; JSON also includes the headings, metadata and unmodified source facts. Both are made in this tab. If your adjacent job is collecting the actual image files rather than auditing their alternatives, use the website image extractor, which keeps its own byte-level success and failure receipt.