Private accessibility review

Alt Text Checker and HTML Image Audit

Inventory image alternatives in pasted HTML and review one draft against its intended purpose. Nothing is fetched, rendered, scored, or sent away.

Context decides the right alternative. A missing attribute is mechanically detectable; whether alt="" or descriptive text is correct depends on what the image does and what nearby content already communicates.

Audit pasted HTML

The bounded lexical scanner inventories img, input type="image", and area elements without creating a DOM or loading their sources.

Maximum 200,000 UTF-16 units, 400,000 UTF-8 bytes, 20,000 markup tokens, depth 100, and 500 image records. JavaScript-generated DOM and external context are not present.

Review one authored alternative

Leave this exactly empty only when the selected decorative intent is correct.

Choose the image purpose
  • Automated wording checks cannot see the image or determine its purpose.

HTML audit report

0records
0needs action
0human review
0no mechanical flag

Paste HTML and run an audit. Referenced images will not be loaded.

Browser-local: no HTML, alt text, URL, image, or report is transmitted or persisted by this tool. CSV cells that could be interpreted as spreadsheet formulas are prefixed defensively.

How to use the HTML audit

  1. Paste source HTML or a rendered fragment. Do not paste credentials or private page data into tools you do not trust.
  2. Run the audit and fix definite markup problems first: omitted alternatives, whitespace-only values, and unnamed image controls.
  3. Review every empty and nonempty alternative in its real page context. The report’s ancestor text is only a clue.
  4. Re-run after edits, test the actual page with its accessibility tree and keyboard, then include people who use screen readers in meaningful review.

Choose by purpose, not by a score

  • Informative: replace the information or purpose the image contributes.
  • Decorative or redundant: use the null attribute alt="", not a missing attribute, space, filename, or “decorative image.”
  • Functional: name the action or destination, not merely the pixels.
  • Complex: use a concise identifier and provide the data or explanation nearby or through an appropriate linked description.

What the diagnostics mean

Needs action is reserved for mechanically strong problems such as a missing attribute, blank-but-not-null value, or an image control without alt. Human review covers contextual or editorial signals: null alternatives, filenames, duplicated surrounding text, generic wording, and unusually long text. No mechanical flag means only that the scanner found no supported pattern; it is not an accessibility pass.

There is no universal alt-text character limit. The 200-grapheme prompt is an editorial review cue, not WCAG. Correct detail depends on purpose and whether a longer description is available.

Standards and limits

Start with WCAG 2.2 Success Criterion 1.1.1, the WAI alt decision tree, the HTML image alternative requirements, the HTML-AAM image-input name order, and the Accessible Name and Description Computation.

Accessible-name evidence is intentionally bounded. The scanner uses the first recognized WAI-ARIA 1.2 role token, distinguishes its widget-role list from roles whose descendant content can supply a name, and recognizes native buttons, linked anchors, summaries, and explicit or wrapping labels for image inputs. For an image input it reports the supported author-source order: aria-labelledby, aria-label, associated label text, nonempty alt, then title. A browser may supply a localized submit name when none is usable; that fallback is not author evidence. Hidden descendant text is excluded from ordinary clues, while hidden text directly referenced by aria-labelledby is included. Whitespace-only references are absent. Unresolved or duplicate IDs remain a human-review uncertainty because the pasted fragment may omit their targets; the tool does not present its clues as a computed accessibility tree.

This lexical audit is not an HTML validator or accessibility-tree engine. It recovers numeric character references and a small common named-reference set; uncommon named references remain literal, and ambiguous missing semicolons are reported rather than guessed. It cannot see pixels, generated DOM, CSS-driven visibility, background images, shadow DOM, framework source syntax, external page fragments, recursive naming relationships, language nuance, or whether nearby text truly replaces the image. It does not check live pages, SEO rankings, image search, legal compliance, or screen-reader output.