Image text-alternative coverage
What it checks
An image with no text alternative has no accessible name. It appears as an unnamed node in the accessibility-tree snapshot that agent toolkits send to a model: Playwright MCP, Claude-in-Chrome read_page, Chrome DevTools take_snapshot. It also carries no subject matter for Google Images, which states it uses alt text to understand what an image shows.
This measures the accessible name, not the alt attribute alone: aria-labelledby, aria-label, alt and title, in the order accname ranks them. Images marked decorative (alt="", role="presentation") or hidden from assistive technology (aria-hidden="true") are excluded, because correct markup is not a defect.
A multimodal agent that fetches the image bytes can caption it without a text alternative. A text-only crawler or a snapshot-driven agent cannot.
Why it matters
The alt attribute is the native text-alternative source in the accessible-name computation. It therefore becomes the accessible name of an <img> node in every accessibility-tree snapshot. It is also the only representation of the image for text-only crawlers, which neither execute JS nor run vision models over page images. Google separately states it uses alt text as an input to understanding image subject matter. An image with missing alt is an unnamed node an agent cannot refer to; an image with alt=‘’ is mapped to presentation/none and intentionally removed from the tree.
Evidence
Image alt text as the machine-readable representation…
Direct vendor statement: ‘Google uses alt text along with computer vision algorithms and the contents of the page to understand the subject matter of the image’ [google-image-seo-docs]. The mechanism is standardised: accname (W3C Recommendation, 2018) lists HTML alt among the native host-language text-alternative sources ranked below aria-labelledby/aria-label [w3c-accname], and HTML-AAM maps img[alt] to the image role and img[alt=‘’] to none/presentation [w3c-html-aam]. Vercel’s crawler-log data shows the AI crawlers that matter here do not execute JavaScript at all [vercel-ai-crawler-study], so server-rendered alt is what they get.
The ads experiment gives the behavioural corollary: agents across GPT-4o, Claude 3.7 Sonnet, Gemini 2.0 Flash and OpenAI Operator ‘ignore purely visual calls to action, clicking banners only when semantic button overlays or off-screen text labels are present’ [machine-readable-ads-paper]. Baseline: 69% of images pass the alt audit and ~8.5% of alt values are just filenames [web-almanac-2025-accessibility].
Limits
The ‘multimodal AI’ framing is where this overreaches. Neither OpenAI nor Anthropic documents consuming alt text anywhere, and Google’s AI-features page says no special optimizations are needed for AI Overviews [google-ai-features-docs]. Vercel’s data shows ClaudeBot spends 35.17% of its fetches on images [vercel-ai-crawler-study]. Image bytes are therefore being retrieved, and can plausibly be captioned by a vision model with no alt at all. Capable multimodal systems can substitute for alt in a way they cannot substitute for a missing heading.
So grade A rests on Google’s explicit statement, not on a general ‘all AI reads alt’ claim; the audit should say so. Also note alt=‘’ is correct, not a failure, for decorative images (30% of alt attributes are legitimately empty [web-almanac-2024-accessibility]) — an audit that flags empty alt as missing alt is wrong.
How it scores
Google states it directly: “Google uses alt text along with computer vision algorithms and the contents of the page to understand the subject matter of the image.” The mechanism is standardised rather than conventional. accname, a W3C Recommendation, lists alt among the native text-alternative sources, ranked below aria-labelledby and aria-label. The attribute therefore has a defined role in the tree every agent snapshot is built from. A vendor statement plus a ratified specification is the grade-A bar. The grade does not extend to the multimodal framing this audit used to carry: neither OpenAI nor Anthropic documents consuming alt text, and Google’s AI-features page says no special optimisation is needed for AI Overviews.
Sources
- Image SEO Best Practices — Google Search Central — Google, vendor-doc (verified 2026-08-20)
- Accessible Name and Description Computation 1.1 — W3C, spec (verified 2026-08-20)
- HTML Accessibility API Mappings 1.0 — W3C, draft-spec (verified 2026-08-21)
- The rise of the AI crawler — Vercel / MERJ, study (verified 2026-08-21)
- Machine-Readable Ads: Accessibility and Trust Patterns for AI Web Agents interacting with Online Advertisements — arXiv (Nitu, Mühle, Stöckl, 2025), study (verified 2026-08-20)
- Web Almanac 2025 — Accessibility chapter — HTTP Archive, dataset (verified 2026-08-20)
- Web Almanac 2024 — Accessibility chapter — HTTP Archive, dataset (verified 2026-08-20)
- AI features and your website — Google Search Central, vendor-doc (verified 2026-08-21)
- The rise of the AI crawler — Vercel, dataset (verified 2026-08-21)
- AI features and your website — Google Search Central — Google, vendor-doc (verified 2026-08-21)