AI Alt Text: Accessible, SEO-Ready Product Images

A product image without alt text is invisible to screen reader users and mute to search engines. This is exactly where AI alt text comes in: image language models (vision models) analyse your entire product image library and suggest alternative text at scale - relevant for accessibility under WCAG 2.2 and for image SEO alike. The catch: 53.1% of all home pages contain images without alt text (WebAIM). This guide shows how a pipeline of vision analysis, a brand glossary, special-case handling and - crucially - human final review makes your online shop both accessible and visible.

Why alt text decides both accessibility and SEO

Alt text (alternative text) is the link between an image and everyone who cannot see it. Screen readers read it out, search engines treat it as a content signal, and on slow connections it appears when the image fails to load. Without alt text, a product photo remains entirely invisible to blind and low-vision users. The WebAIM Million Report 2026 analysed the one million most-visited home pages: missing alternative text is the second most common accessibility failure at 53.1% of pages (WebAIM) - right after low text contrast at 83.9% (WebAIM) and ahead of missing form labels at 51% (WebAIM).

The scale is larger than it first sounds. On average 10.8 alt texts are missing per home page (WebAIM), and 16.2% of all images on those pages carry no alternative text at all (WebAIM). Add poor descriptions - a file name or the bare word image in the alt attribute, which applies to 10.8% of labelled images (WebAIM) - and more than one in four images on popular home pages has missing, questionable or repetitive alt text (WebAIM). For a shop with thousands of articles, that adds up to a structural barrier.

Accessible image descriptions also pay into revenue. Product images are one of the strongest drivers of the purchase decision in online retail - anyone who cannot see them needs the description as a full substitute. Alt text that names material, fit and condition reduces uncertainty and follow-up questions and thus supports conversion. Accessibility is therefore not a mere compliance line item but a contribution to the user experience for everyone - including people on slow connections or with situational limitations, such as strong sunlight on a smartphone display.

What WCAG 2.2 and the BFSG actually require

The legal basis is unambiguous. Success Criterion 1.1.1 (Non-text Content) of the Web Content Accessibility Guidelines (WCAG) 2.2 requires a text alternative that serves the same purpose for any non-text content (W3C). This applies to product photos just as much as to icons, diagrams and charts. In Germany, the German Accessibility Strengthening Act (BFSG) makes this binding for e-commerce services - it has been in force since 28 June 2025 (BFSG). As the central authority, the Bundesfachstelle Barrierefreiheit points to BITV 2.0, which in turn references the WCAG success criteria of conformance levels A and AA (Bundesfachstelle Barrierefreiheit).

The key point: not every image needs the same alt text - and some need none at all. In its Images Tutorial, the W3C distinguishes by the function of the image. The overview below maps the most common shop image types to their requirements:

How vision models analyse product images

Image language models (vision models) combine image processing with language generation. They recognise objects, read text in images, determine colours and materials and describe scenes in natural language. For a product image library with thousands of articles, that means: instead of writing every alt text by hand, you can generate a first draft for the entire library. This is what makes AI-supported processes for accessibility scalable in the first place - manually maintaining ten thousand images simply never happened in many shops.

The strength of these models lies in volume and consistency. Combine the image analysis with existing catalogue data - for example from a PIM or from data enrichment - and you get descriptions that cleanly merge model number, colour and category. That way accessibility and image SEO in AVIF- and WebP-optimised shops reinforce each other.

The pipeline: from image analysis to approved alt text

A reliable process is not a single model call but several coordinated stages. Only the interplay of analysis, context knowledge and control produces alt text that satisfies both the standard and the brand:

1. Inventory: All images are catalogued and classified as informative, functional, decorative or complex - because only informative and functional images need meaningful alt text.
2. Vision analysis: The model produces a raw draft from objects, colours and recognised text for each relevant image.
3. Context enrichment: The raw draft is merged with catalogue and brand data from the brand and context glossary.
4. Rule check: Length, banned filler phrases (such as image of), duplicates and empty alt attributes for decorative images are checked automatically.
5. Editorial final review: A human reviews, corrects and approves - especially for sensitive categories and text-in-image.
6. Write-back and monitoring: Approved alt texts flow back into the shop system and PIM, and new images run through the same pipeline.

For quality to hold up over time, a robust pipeline also needs versioning. Every alt text is given a status - draft, reviewed, approved - and can be attributed to an editor. That keeps it traceable which description a model suggested and which one an editorial team ultimately took responsibility for. This documentation is at the same time evidence to market surveillance authorities that accessibility is not left to chance but handled systematically - a point that carries weight in an inspection.

Brand and context glossary: why generic descriptions fall short

A vision model sees a blue shoe - your brand sells the Aerorun in cobalt. Without context, AI alt text stays generic, misses the brand's terminology and leaves valuable search terms on the table. A brand and context glossary closes that gap: it stores product lines, colour names, material names and the brand's tone of voice. That turns the model's raw draft from an arbitrary description into one that fits your shop.

The glossary also covers multilingual output. A shop with a German and an English interface needs alt text in both languages - a mere translation rarely suffices, because colour, size and product names differ by market. Vision models can produce both language versions, but terminology and tone of voice are set by the glossary. That way every description sounds like your brand rather than a generic catalogue, and image SEO takes hold in each target market with the terms actually searched there.

The difference between weak and strong alt text is substantial - for people and search engines alike:

Good alt text does not describe that an image is present, but what a sighted user takes away from it - short, concrete and in the context of the page.

XICTRON editorial team

Special cases: decorative images, complex graphics, text in images

Not every image wants to be described. The W3C Images Tutorial makes clear that a badly set alt text can even worsen accessibility - for example when a purely decorative pattern is read aloud and holds up the screen reader user (W3C). Three special cases deserve particular attention in the shop:

Image SEO: how good alt text boosts visibility

Alt text has always been a ranking signal for image search - and image search is more relevant than many shops assume. According to a clickstream analysis, a considerable share of all Google searches goes to image search (SparkToro/Datos). For visually driven assortments such as fashion, furniture or decor, image search is a channel into the shop in its own right.

• Relevance: Precise alt text explains the image content to search engines and increases the chance of ranking for matching search terms.
• Context: Alt text, file name, caption and surrounding text together form a coherent signal - consistency pays off.
• Accessibility as a ranking neighbour: Search engines assess user experience; clean markup and descriptive images point in the same direction.
• No keyword stuffing: Strings of search terms in the alt text harm accessibility and SEO alike - one sentence per image is enough.

Beyond the alt text, complementary signals help: a meaningful caption, descriptive file names and - for larger assortments - an image sitemap that points search engines to all product photos. Together with structured product data, this creates a consistent picture that image search rewards. The alt text remains the substantive core, because it is the only element that serves accessibility and content relevance at the same time.

For the SEO effect to unfold technically, images must be delivered fast and cleanly. How to combine this with modern formats is described in our article on image SEO with WebP and AVIF; for smooth page transitions on image-heavy category pages, the View Transitions API is worth a look too. Search engine optimisation remains an ongoing part of our SEO work.

Human final review: why no model is enough on its own

As capable as vision models are, they deliver suggestions, not certainty. No model describes an entire image library reliably error-free: it confuses similar products, guesses on poorly lit shots, misses legally relevant details or invents details that are not in the image at all. In a shop where alt text is legally binding, that is not a marginal issue but the reason a human decides at the end.

The limits of these models are well documented and no reason to forgo AI - they are a reason to set the process up properly. Where a model is uncertain, it can flag it: low confidence, recognised but unclear text-in-image, or several plausible product variants. Such cases are routed for review first, while unambiguous motifs pass through faster. That way human attention concentrates where it makes the biggest difference - and the editorial team does not blindly review everything, but specifically the critical items.

• Does the alt text capture the essentials that a sighted user recognises?
• Do product name, colour and variant match the catalogue data?
• Is the text short, concrete and free of filler such as image of?
• Are decorative images correctly marked with an empty alt attribute?
• Has relevant text in the image been carried over in full?

Keeping alt text quality up in day-to-day operation

A one-off run does not make a shop permanently accessible. New products, swapped images and relaunches constantly bring new motifs into the system. So that alt text quality does not slip again, you need ongoing operation with clear checkpoints - otherwise the gap grows back faster than it was closed.

• Automatic trigger: Every newly uploaded image runs through the pipeline before it goes live - no product image without an alt text status.
• Regular sampling: The editorial team reviews a selection of approved texts at intervals to catch drift in model quality early.
• Error report: Images without alt text or with questionable descriptions appear in a report that sets priorities like a BFSG audit.
• Feedback loop: Frequent corrections flow back into the glossary and the check rules so that the suggestions improve over time.

Alt text as part of your accessibility strategy

AI alt text is not an end in itself but a building block of a robust BFSG strategy. It solves exactly the problem many shops have put off for years: the sheer volume of unlabelled images. The path there runs through a well-designed pipeline - vision analysis for volume, a brand glossary for precision, clear rules for special cases and a human final review for responsibility. We set up this process for your shop in an accessibility-compliant, editorially safeguarded way and connect it to your shop system and PIM. Get in touch if you want to make your image library both accessible and visible to search engines.