AI Summary
Alt text is the text representation of an image inside an AI retrieval pipeline. When an AI crawler indexes your page, it processes each image as a chunk. If alt text is present and descriptive, that chunk carries a text signal that can be matched to a query. Google’s official image SEO guidance on developers.google.com states: avoid filling alt attributes with keywords, which results in a negative user experience and may cause your site to be seen as spam. WCAG Success Criterion 1.1.1 (Level A) requires that all non-text content presented to the user has a text alternative serving the equivalent purpose. Both rules point to the same practice: describe what is actually in the image.
What alt text does in an AI retrieval pipeline
When an AI crawler fetches a page, it processes the full HTML as a series of semantic chunks. Each image element is one chunk. The alt attribute is the text portion of that image chunk. Without alt text, the image chunk contains only the filename and surrounding markup, which is a weak signal for vector embedding and retrieval.
Multimodal models powering Gemini and ChatGPT can read image pixels directly. That capability does not make alt text redundant. Alt text is the reliable text-layer signal because it is always present in the HTML the crawler indexes, it travels separately from binary image data, and it is processed identically to any other text chunk. The multimodal vision pass is a supplementary signal; the alt text is the primary indexable text. Where published crawler documentation from Google, OpenAI, Anthropic, and Perplexity is silent on the exact interaction between vision processing and alt text embedding, we treat the text layer as the safe bet.
GPTBot from OpenAI, ClaudeBot from Anthropic, and PerplexityBot from Perplexity all follow robots.txt and process standard HTML. Their published documentation describes HTML text processing. Alt text is HTML text. That is the documented basis for treating alt text as a retrievable signal.
Google’s verified image SEO guidance
Google’s Search Central Image SEO Best Practices page at developers.google.com contains two rules we apply on every client site. First, write descriptive alt text that accurately represents the image content. Second, avoid keyword stuffing in alt attributes, which Google explicitly states results in a negative user experience and may cause the site to be seen as spam. The Google SEO Starter Guide on the same domain reinforces this: “Add descriptive alt text to the image. Alt text is a short, but descriptive piece of text that explains the relationship between the image and your content.”
Google’s guidance has been consistent since at least 2007. What changed is the downstream use. The same alt text that used to feed Google Image Search now also feeds into multimodal retrieval pipelines. The standard was right then. It is even more consequential now. For schema-marked content, alt text on images is treated as anchor text for any links the image contains, and as descriptive metadata for the entity the image represents.
WCAG 1.1.1: the accessibility requirement that doubles as AI hygiene
W3C WCAG 2.1 Success Criterion 1.1.1 (Non-text Content, Level A) states: all non-text content that is presented to the user has a text alternative that serves the equivalent purpose. The W3C’s Understanding document explains the intent: text alternatives allow information to be rendered through any sensory modality, including synthesized speech, braille, or visual display. This is the accessibility baseline for every image on every public website.
The WCAG decorative exception is equally clear. WCAG identifies Situation F: if non-text content is pure decoration, is used only for visual formatting, or is not presented to users, it must be implemented so that assistive technology can ignore it. The implementation is alt=”” (empty string). A missing alt attribute fails this criterion. A non-empty alt on a decorative image like alt=”decorative” or alt=”background” also fails it by delivering noise instead of meaning.
The accessibility and AI retrieval cases align without conflict. Accurate alt text satisfies WCAG 1.1.1 and gives the AI crawler a usable text chunk. Empty alt on a decorative image satisfies WCAG 1.1.1 Situation F and tells the AI crawler there is nothing to index. E-E-A-T signals from demonstrated site quality, including accessibility compliance, feed into how Google assesses content for AI Mode citations. The dual benefit is real and documented at both the WCAG and Google Search Central levels.
Our house practice: describe what is in the image
Every diagram we publish on this site has alt text that literally describes what the image shows. That is a non-negotiable rule in our publishing workflow. The diagram on this page has alt text listing each decision node visible in the flowchart. If Gemini or ChatGPT retrieves this page on a query about alt text for AI search, the alt text of the diagram is itself a citable chunk.
We apply the same rule in client work. Before any image ships, we ask: if someone read only the alt text, would they know exactly what information the image conveys? If the answer is no, the alt text is wrong. The check takes five seconds per image. It is the single highest-leverage text-layer fix most content teams can make without developer involvement.
A common failure pattern we find in client audits: a marketing team uploads a diagram and puts the campaign tagline in the alt field. The tagline appears nowhere in the visual. An AI model reading that page encounters a contradiction between the text layer and the visual. Our GEO audit work consistently shows that pages where alt text drifts from the actual image content produce weaker semantic consistency signals than pages where alt text accurately mirrors the visual.
Alt text patterns by image type
A single formula applied to all image types produces mediocre results. Here is how we approach each type:
| Image type | What to describe | Word range | Common failure |
|---|---|---|---|
| Product photo | Product name, key visual attribute (color, material, configuration) | 8-20 words | Product name only, no visual context |
| Chart or data visual | Chart type, subject, key finding or trend with metric | 12-25 words | “Bar chart” with no data meaning |
| Diagram or flowchart | What the diagram shows and its key structure or conclusion | 10-20 words | “Diagram” with no further detail |
| Screenshot | Which tool, what state or action shown, key number visible | 8-15 words | Generic “screenshot of” with no specifics |
| Infographic | Main topic and the primary statistic or claim shown | 12-20 words | Campaign tagline not visible in the image |
| Decorative image | Nothing. Use alt=”” | 0 words | Keyword list or alt=”decorative” |
Product images
E-commerce product images carry commercial intent queries. The alt text pattern: product name first, then the key visual attribute, then use case if relevant. “Matte black stainless water bottle with wide mouth lid, 32oz, shown on a hiking trail” is accurate. “Water bottle stainless steel BPA free 32oz hydration outdoor sports best” is keyword stuffing. The first helps a screen reader user and an AI retrieval system. The second harms both.
Charts and data visualisations
Charts are the highest-leverage alt text opportunity for content-heavy sites. In our May 2026 study of 153,425 citations, the mean cited position was 37% through the document and 74.9% of cited sentences appeared in the first half of the page. Chart alt text that encodes a specific data point creates an additional citable chunk from that same visual content.
The W3C’s WCAG Understanding document gives this as its first example of proper alt text for a data chart: “A bar chart compares how many widgets were sold in June, July, and August.” That structure, chart type plus subject plus key data, is the minimum. Add a key finding or directional change when the visual shows one. Keep under 25 words. Longer reads as stuffing to both screen readers and AI crawler parsers.
Screenshots
Screenshots need context about which product, which interface state, and what is being shown. “Google Search Console Coverage report showing 38 pages with Crawled, currently not indexed status” is complete. “Screenshot” is not. For tutorial screenshots, include the action: “Google Analytics 4 property settings with data retention dropdown set to 14 months.”
Decorative images
Background patterns, dividers, abstract illustrations with no informational content: use alt=”” (empty string). This is the correct WCAG implementation for Situation F and the correct signal for AI crawlers. Do not use alt=”decorative” or alt=”background image”. Those are non-empty strings announcing the image without describing meaningful content, which is a WCAG 1.1.1 failure.
Common alt text failures
In our crawler control audits, five patterns account for most alt text failures across sites:
- Missing alt attribute entirely. Screen readers announce the filename or “image”. AI crawlers get no text signal. Fix: add alt=”” for decorative, write a description for informational images.
- Filename as alt (IMG_4729.jpg, product-image-final-v3.png). Garbage tokens that lower chunk quality. Fix: describe image content, not the file name.
- Keyword list. Google’s image SEO guidance explicitly flags this pattern as spam. Five or more keywords in alt text serves neither users nor retrieval. Fix: describe what is visible in the image.
- “Image of” or “photo of” prefix. Wasted tokens before the actual description. Screen readers already announce the image type. Fix: start with the noun or the metric.
- Caption copy-pasted as alt text. If the caption is visible to all users and matches the alt text exactly, screen readers read it twice. Fix: alt text describes what is shown; caption explains why it matters.
Run a crawl with Screaming Frog or a similar tool. Export the Images tab. Filter the Alt Text column for blank, for strings matching filename patterns, and for strings over 25 words. Most sites find 20 to 40 percent of informational images need alt text work. Prioritise pages that carry your core entity claims and that appear in your current performance audit.
Alt text in the broader text-layer stack
Alt text is one layer of the text-layer hygiene stack that AI retrieval depends on. It sits alongside semantic HTML structure, atomic sentence writing, and readability calibration. None of these factors works in isolation. A page with accurate alt text but thin prose and no schema will not outperform a page that executes all layers well.
Our March 2026 study of 42,971 citations across 520 queries established that AI systems preferentially cite pages where the text content is structured for extraction at multiple levels: paragraph text, list items, table cells, and headings. Alt text is the one extraction layer that most sites handle worst. Fixing it improves visibility for AI crawlers and accessibility for users at the same time.
For scaling across large image libraries, AI-assisted generation is viable with constraints: use a multimodal model to draft alt text from each image, provide context in the prompt (page title, surrounding paragraph, image type), set a word limit of 10-20, and require human review of every generated alt before shipping. Generated alt text without human review produces generic descriptions that miss brand entities and key data points. The human review step is the quality gate.