Bar length = frequency of AI mentions (per 1,000 words of the FY2025 annual report); segments = the context in which the board writes about AI. Hover a bar for details.
Click a company row to expand the list of mentions (quote ≤25 words, page, term) — every number is auditable. * = context mix computed on a systematic sample.
| Company | Mentions | /1,000 words | Dominant context | Substance |
|---|
At N=10, a single company or the choice of estimator can shift a result several-fold. So every number goes through an estimator swap, leave-one-out, and stratification by document type — and the result is published alongside the report.
| Claim | Status |
|---|---|
| Sector gradient — digital > industrials > financials > resources | robust holds separately within each document type |
| Extremes spread | robust follows directly from the data |
| Risk and regulation as the dominant context | robust in the data caveat: annual reports carry mandatory risk sections |
| DAX↔WIG20 ratio | fragile ranges depending on method; document type is also almost perfectly confounded with country in this edition |
What we measure. Frequency and context of AI mentions in FY2025 annual reports — the full Geschäftsbericht / integrated report (DAX companies) and the management board activity report or full annual report (WIG20 companies). Documents were retrieved exclusively from official, public investor-relations websites; one language version per document (local language preferred). Multilingual EN/DE/PL dictionary (AI, KI, künstliche Intelligenz, sztuczna inteligencja, machine learning, GenAI, LLM…); word-boundary matching, overlapping hits counted once; normalised per 1,000 words.
Context classification. Each mention (±2 sentences) → one class: revenue/growth, cost/efficiency, risk/compliance, talent, buzzword; plus a "substance" flag (a number, project name or concrete result). Classification by an LLM; a human reviews a random ≥10% sample before publication. For documents with very many mentions, the class mix is computed on a systematic sample (every k-th mention) — frequency is always computed on the full text.
Limitations. N=10, purposive sampling (largest companies, approximate DE↔PL sector pairing) — this is a signal, not a benchmark; results do not generalise to whole indices. Document type is almost perfectly confounded with country in this edition (the German companies supplied full annual reports, most Polish ones management board reports), so the country effect cannot be separated from the document effect; that is why the DAX↔WIG20 comparison is reported as a range rather than as a result. Reporting-culture differences may affect normalisation. Siemens' fiscal year ends in September 2025. "Mentions AI rarely" ≠ "is behind" — the report describes, it does not judge. Short quotes with attribution (fair use); no redistribution of full texts. Full methodology: metodyka.md (Polish) · robustness test: analiza-wrazliwosci.md (Polish) · source data: dataset.json (every mention with its quote, page and document checksum).