AI visibility of German websites: 1,000 checked
By Philipp Kant 7 min read
AI visibility is how easily AI answer engines such as ChatGPT, Perplexity, and Google’s AI Overviews can reach a website, read its content, and tell who is behind it.
On 26 October 2026 we ran the 1,000 most-linked German websites (the top .de domains in the Majestic Million) through our own AI Visibility Check. 852 of them could be audited. The median score is 72 out of 100. Access and content are mostly in order. The weak spot is identity: 60.4% of the sites never state in machine-readable form who runs them.
| Signal | Share of top German sites |
|---|---|
| Organization or Person schema | 39.6% |
| sameAs links to other profiles | 26.5% |
| Block at least one AI crawler in robots.txt | 27.0% |
| Block GPTBot (OpenAI training) | 21.8% |
| Block OAI-SearchBot (ChatGPT search) | 5.3% |
| Under 80 words of text without JavaScript | 4.0% |
| Exactly one H1 on the homepage | 60.1% |
| Complete Open Graph tags | 46.5% |
| llms.txt | 5.9% |
60% of top German sites never say who runs them
Only 337 of the 852 homepages carry an Organization or Person node in their structured data. We counted every schema.org subtype, so a newsroom with NewsMediaOrganization, a university with CollegeOrUniversity, or a ministry with GovernmentOrganization all count. Half of the sites (50.1%) have no JSON-LD at all.
The links that tie a company to its other profiles are rarer still: 26.5% use sameAs to point at LinkedIn, Wikipedia, Wikidata, or similar.
This matters because an answer engine has to attribute what it quotes. Microsoft has said that Bing uses schema markup to help its language models, Copilot included, understand page content (Search Engine Land). Every German site has an Impressum that names the operator, but that is a page of prose for humans. One JSON-LD block says the same thing in a form a machine reads without guessing.
27% block at least one AI crawler, mostly the training bots
Of the 833 sites that serve a parseable robots.txt, 225 (27.0%) fully block at least one AI crawler, and 32.2% address AI bots in some form. We used the rule from our scan of the 800 most-linked websites worldwide in August, so the two samples compare directly.
| AI crawler | Purpose | Top German sites | Top global sites |
|---|---|---|---|
| GPTBot | OpenAI training | 21.8% | 19.2% |
| CCBot | Common Crawl | 21.7% | 20.8% |
| Bytespider | ByteDance | 20.3% | 18.2% |
| ClaudeBot | Anthropic | 19.4% | 18.5% |
| Google-Extended | Gemini training | 17.4% | 16.6% |
| meta-externalagent | Meta AI | 17.2% | 15.9% |
| Applebot-Extended | Apple training | 16.4% | 14.6% |
| PerplexityBot | Perplexity search | 11.9% | 15.0% |
| Amazonbot | Amazon | 11.6% | 13.5% |
| OAI-SearchBot | ChatGPT search | 5.3% | 8.0% |
German sites block the training crawlers a little more often than the global top and the search crawlers less often. 138 of them (16.6%) block GPTBot while leaving OAI-SearchBot open. That is the deliberate split: no training on the content, citations in ChatGPT search still possible.
96% of the sites serve their text without JavaScript
The median homepage delivers 1,241 words of readable text in the raw HTML. Only 4.0% serve fewer than 80 words, the empty shell a site leaves when it renders everything in the browser.
That is good news, because the major AI crawlers fetch JavaScript files but do not run them (Vercel and MERJ, 2024). A homepage that only fills itself in the browser is close to blank for GPTBot and ClaudeBot, however good it looks to a visitor.
A quarter of the homepages have no H1
The basics that help a machine find the main topic are missing more often than the advanced ones:
- No H1: 24.3% of homepages. Another 15.6% have several.
- No meta description: 13.5%.
- No Open Graph tags at all: 28.2%. Only 46.5% set all three (title, description, image).
- No sitemap at
/sitemap.xmlor in robots.txt: 25.4%. - Links that are not links: 35.0% have more than three anchors
without a usable href (
#orjavascript:), which an AI agent browsing the page cannot follow.
llms.txt is on 5.9% of German top sites
50 of the 852 sites publish an /llms.txt file, against 12.9% among
the global top 800 in August. Google’s guide to its generative AI
features says Google Search ignores the file and that it “will neither
harm nor help” a site there
(Google Search Central).
Treat it as optional. The Organization schema above does more.
The 100 biggest sites score lower than the rest
Size does not buy AI readability. The 89 audited sites among the 100 most-linked score a median of 70, against 72 for ranks 101 to 1,000. 36.0% of them carry Organization schema (rest: 40.0%), and 7.9% serve an almost empty HTML shell (rest: 3.5%).
How we measured
- Sample: the 1,000 highest-ranked .de domains in the Majestic Million (a ranking by referring subnets), retrieved on 26 October 2026.
- Audit: each homepage was fetched once by the engine behind our
free AI Visibility Check, along with
/robots.txt,/llms.txt, and/sitemap.xml. Same code, same scoring as the public tool. - Coverage: 852 sites were audited. 148 were not: 68 answered our self-identified checker with HTTP 403, 65 did not respond in time or failed at DNS or TLS, and 15 returned other errors (429, 404, 401, 410, 503).
- AI crawlers: measured on 833 parseable robots.txt files. A
crawler counts as blocked when its own User-agent group contains
Disallow: /. BlanketUser-agent: *rules are not counted, as in the global scan. - Schema: JSON-LD only. Microdata, RDFa, and JSON-LD that a script injects after load are not read, which is also what most AI crawlers see.
- llms.txt: counted when
/llms.txtreturns a non-HTML file, the rule from the global scan. - Limits: homepage only, one fetch per site. The global comparison uses a different rank range (the top 800 worldwide), so treat the differences as indicative.
- Our own fixes: running the study surfaced three measurement errors in our tool (pages that leave a tag unclosed, HTML error pages counted as llms.txt, and schema subtypes such as NewsMediaOrganization). All three are fixed in the public tool, and every number here comes from the corrected version.
- Raw data: ai-visibility-german-websites-2026.csv, one row per domain.
What to fix first on a German business site
- Say who you are. One Organization JSON-LD block with name, url, logo, and sameAs links to LinkedIn, Wikidata, or Crunchbase. It is a single block of code, and it is the gap 60% of the top sites still have.
- Decide training and search separately. Keep OAI-SearchBot, PerplexityBot, and Claude-SearchBot open if you want to be cited. Block GPTBot or CCBot only if your content is your product.
- Fix the basics. One H1, a meta description, and Open Graph tags on every important page.
- Read your raw HTML. If the homepage has fewer than 250 words before JavaScript runs, AI crawlers see an empty page.
Product pages carry the same problem one level deeper; see Product data for AI shopping assistants.
FAQ
How many German websites block AI crawlers? Of the top 1,000 German websites, 27.0% of those with a readable robots.txt fully block at least one AI crawler. GPTBot is blocked by 21.8%, OpenAI’s search crawler OAI-SearchBot by 5.3%.
Do German websites use Organization schema? Mostly not. 39.6% of the audited top German homepages declare an Organization or Person in JSON-LD, and half have no JSON-LD at all.
How many websites have an llms.txt file? 5.9% of the top German sites in our scan, and 12.9% of the global top 800 in August 2026. Google Search ignores the file.
Does a high AI visibility score mean ChatGPT will cite my site? No. The score measures whether AI systems can reach and understand the page. Whether they cite it depends on what others say about you and how well the page answers the question.
Can I check my own site? Yes. The AI Visibility Check runs the same checks on any URL, free and without signup.
Making a site readable for AI is the first step of the AI integration work we do: the systems behind the site come next.
Ask an AI about this post
