Thousands of images, no alt text: the invisible problem of international websites

A quick self-test

Open your company website. Right-click any product image, choose "Inspect", and look at the alt attribute. Does it say anything useful? Does it say anything at all?

If yes: congratulations, you are in the minority. If no: welcome to the club. When we recently analyzed the website of a mid-sized measurement technology manufacturer, we found almost 13,000 image occurrences without alt text on the German main domain alone, spread across roughly 600 pages. And this was a well-maintained, professionally run website. Not a neglected one. The normal case.

Why alt text is not a nice-to-have

Three reasons, sorted by urgency:

1. Accessibility is now the law. People who work with screen readers "see" an image only through its alt text. If it is missing, the content is missing: For a product image, the product itself is simply gone. Since June 2025, the European Accessibility Act (in Germany implemented as the BFSG) obliges private companies to make their digital offerings accessible. Alt text is not optional polish here, it is one of the most fundamental requirements of the WCAG.

2. Search engines, and increasingly AI assistants, read along. Google understands images without descriptions less well, and image search largely ignores them. And if you want ChatGPT, Perplexity and friends to represent your products correctly, you should give them text they can work with.

3. It is simply wasted content. Every product photo, every application image carries information, but only for those who can see it.

So why is it missing everywhere?

Because no CMS enforces it and no editorial team has time for it. The typical flow: An image gets uploaded, the colleague has four other tasks on her desk, the alt field stays empty, the page goes live. The image looks fine, after all. That something is missing is, quite literally, invisible.

On international websites the problem then multiplies mercilessly. Let us do the math with our example: A main domain with roughly 600 pages, plus eleven country subsidiaries with their own instances in eight languages. Even with generous deduplication (the same product image appears on dozens of pages, after all), you are left with more than 1,200 unique images per language version that need alt text. Across all instances: a five-digit number of images, each to be described in the correct local language.

Now the editorial math: A good alt text takes two to three minutes to write manually. Look at the image, understand the context, phrase it, enter it into the CMS. For 15,000 images that adds up to more than 600 working hours, conservatively estimated. Four person-months. For a mandatory field. In practice this means: it never happens. As they say in East Westphalia, where we are from, people keep it short: nobody does that.

What automation can do today

This is exactly where something fundamental has changed in the past two years. Multimodal AI models can not only recognize images, they can describe them in the context of their page. And this context is the decisive difference between a useful and a useless alt text.

An example from the real project: A photo shows a person holding a gray device at a pipeline. A model without context writes: "Person holding a device." Correct, but worthless. Our approach gives the model everything the page has to offer: page title, the heading next to the image, surrounding text, caption, even the file name. The result: "Inspection camera [model name] in use for damage localization in small pipes." That is an alt text that actually helps screen reader users and search engines.

This is the pipeline we build for it:

1. Collect instead of guess. All pages of all language instances are gathered via the sitemaps. This works with every major CMS (TYPO3, WordPress, Drupal, Shopware and others) without us touching the system.
2. Deduplicate. Images are grouped across all pages. In the example project, 13,000 image occurrences shrank to 1,246 unique images, more than 90 % less effort.
3. Classify. Not every image needs text: Decorative icons, flags and tracking pixels deliberately get an empty alt="" recommended, in line with the WCAG. That too is a decision that deserves documentation.
4. Generate context-aware and multilingual. Every image is described with its page context, in the language of the respective instance. The Dutch site gets Dutch alt text, the Japanese site Japanese. Not translated, but written directly in the target language.
5. Human review. The results land in a review report with preview images and in an export file for the CMS. The editorial team checks and enters them. Or we write the texts directly into the CMS asset metadata via API, so the text automatically applies everywhere the image is used.

Perhaps the most surprising number, to finish

The complete German instance from our example, 1,200 images described with context, caused less than one US dollar in model costs and finished in under twenty minutes. The 600 editorial hours from above shrink to a manageable review effort: scroll through, correct the outliers, approve.

As a side effect, the run also found seven images that were simply broken links on the live site. That too only becomes visible once you look systematically.

Conclusion

Missing alt text is not an editorial failure, it is a scaling problem. On international websites it is realistically unsolvable by hand, but with context-aware, multilingual generation and a clean review process it becomes a matter of days instead of person-months. And with the accessibility legislation in force, "we will get to it eventually" is no longer an option.

If you want to know how many images on your website are missing alt text: The analysis is the fastest part. Get in touch, we will take a systematic look.