Germany's listed companies are not ready for the AI revolution in Investor Relations

Why this study?

While we at svrn_alpha are working on proprietary AI tools – systems that autonomously crawl financial websites, extract data and analyze it – we made a surprising discovery: most investor relations websites of German listed companies are not prepared for a world in which AI agents play a central role.

This is not an academic observation. It is a market-relevant finding. Institutional investors, research departments, investment banks and increasingly also private investors are already using AI tools for due diligence and analysis. These tools depend on financial information being machine-readable – structured, consistent and without detours through JavaScript rendering or anti-bot measures.

Together with MPCM (Münchmeyer Petersen Capital Markets), a Hamburg-based financial services company specializing in research, sales & trading and corporate finance for German mid-caps, we systematically analyzed 101 HDAX companies.

Methodology

Our analysis covers 11 evaluation categories that together capture how AI-ready an IR website is. Each category was evaluated through automated crawling processes and AI-powered assessments – systematically, reproducibly and without human interpretation:

1. Discoverability – Can agents find and crawl the website?
2. Static Content – Is content available without JavaScript?
3. JS Rendering – Does the website support dynamic rendering?
4. Data Accessibility – Are downloadable data files (Excel/CSV) or an llm.txt file available?
5. Navigation – Can a bot navigate the page?
6. Performance – Load times and efficiency
7. Anti-Bot & Access Friction – Are bots blocked or hindered?
8. Content Quality – Signal-to-noise ratio
9. Machine-Readable Standards – Schema.org, JSON-LD, semantic HTML
10. Investor Relations Specific – IR-specific best practices
11. LLM Comprehension – Can Large Language Models understand the content?

Sample: 101 HDAX companies (April 2026). For 5 additional companies (4.1 % of 122 attempts) the crawler failed terminally even after retry – this finding alone is telling.

Key Findings

Not a single company achieved an A-tier rating (85+). The top ten companies score between 77-78 points – solid B-tier, but far from true AI-readiness.

Average Score by Category

Score Distribution

The Crawling Problem

Of 122 first-attempt crawls, 21 produced an error. 16 of these were transient network errors that recovered on retry. For 5 sites (4.1 %) access failed permanently even after retry – four with timeouts after 7.5 minutes (suggesting complex anti-scraping or JavaScript-challenge mechanisms), one with an HTTP 502 server error. For AI agents, this means: access is blocked from the start.

Top 10: The Best Prepared

These ten companies perform relatively best. But even they are far from being considered AI-ready. They demonstrate what is possible when the technical foundations are sound:

What the Top 10 have in common: solid technical foundations, good performance, clear navigation and little anti-bot friction. But they too neglect structured data and rarely offer an llm.txt file or downloadable data files.

The Complete Ranking

The table below shows all 101 analyzed companies. Rankings from 11 onwards are displayed anonymized.

The Most Critical Deficiencies

Almost No llm.txt Adoption, XBRL Practically Absent

Only about 10 % of HDAX companies provide an llm.txt file – the emerging standard that helps AI agents understand a website's structure and content. About 32 % of companies offer at least some downloadable Excel (.xlsx) files for items such as earnings dates, dividend calendars, or KPI summaries. XBRL/iXBRL links are present at only 5 of 101 firms (about 5 %), and structured XBRL bundles as an integral IR data stream remain the exception.

This means: every bot must crawl the website, parse HTML and dig through the data with regular expressions. Error-prone, slow and far from state-of-the-art.

Machine-Readable Standards: 22.3 out of 100

Schema.org markup, JSON-LD, semantic HTML: almost nowhere to be found or only rudimentary. When LLMs and specialized financial AI models process a webpage, structured data is essential.

IR-Specific Best Practices: 55.1 out of 100

Earnings dates are hard to find, quarterly reports not organized in machine-readable form, dividend information scattered. This is an IR function that is not fulfilling its core mission: reliably informing investors – whether human or agent.

LLM Comprehension: 66.3 out of 100

Even modern Large Language Models struggle to systematically search complex IR websites. Too much context switching, too little structured information, too much boilerplate.

What This Means in Practice

An AI agent tasked with automatically answering the following questions regularly fails:

• When is the next earnings call?
• What KPIs will the company publish in the Q2 report?
• When was the last dividend paid?
• Were there changes in the board or management?

For a human, a minute of research work. For a bot without structured data: impossible.

What Needs to Happen Now

For IR Departments

1. Implement structured data: Schema.org markup for events (earnings calls), organizations, financial entities. JSON-LD is the standard.
2. Provide an llm.txt file and downloadable data: Publish an llm.txt file so AI agents can navigate your site. Offer key financial data as downloadable Excel/CSV files. It's 2026 – machine-readable data is not a luxury.
3. Optimize static content: Server-side rendering instead of client-side JavaScript. While crawlers can now render JavaScript, it's slower and more error-prone.
4. Semantic HTML: Logical heading structure, <article>, <section> and clear labels for financial metrics.
5. Robots.txt and Sitemap: Allow trustworthy bots, configure comprehensive sitemap with priorities.
6. Performance: Load times under 2 seconds. Bots have short timeouts.

For Investors and Research Teams

• Demand structured data from portfolio companies. Machine-readable IR is part of investor service.
• Deploy AI tools that can handle semi-structured data – but understand that much of the effort is driven by poor website quality.

For Regulation and Exchanges

• Consider minimum standards for IR websites. BaFin, Frankfurt Stock Exchange and DAX governance should put this on the agenda.

Conclusion: The Market Will Not Wait

The ability to communicate with AI agents is no longer science fiction in 2026 – it is reality. Institutional investors are already using these tools. Retail investors will follow. Those who do not prepare their IR websites for this reality will lose attention, liquidity and trust.

The good news: the necessary steps are concrete, cost-effective and often quick to implement. An llm.txt file takes minutes to create. Structured data is not uncharted territory. What is missing is awareness and prioritization – that is exactly what we want to change with this report.