I read the sources behind the answer

On my desk there is often a German product page, an English company profile, and an AI answer that has welded the two together at the wrong seam. A composite manufacturer becomes a reseller because an old distributor page used stronger wording and still has a half-broken logo in the header. A composite regional service company becomes "local-only" because the English export page never explained its cross-border work and the only map citation points to the head office. A composite specialist supplier becomes generic because the clearest source was a directory category, not the company's own site. That is usually where I begin: with the hinge sentence.

I am from northern Germany, and I have spent seventeen years around search content auditing, B2B website editing, technical page review, directory cleanup, multilingual evidence mapping, and marketing diagnostics for export-oriented firms. The work has made me suspicious of neat summaries. German pages often carry the precise technical proof. English pages often carry the market language. Procurement portals may carry the official category. Local press may carry the reputation. AI search can pull from any of them, and it will not explain which language it trusted more unless someone records the answer carefully.

My current work is built around a bilingual citation ledger. Every answer is saved with the exact query, language, date, and engine. Then I reduce it to source, claim, language, and missing proof. I do not treat a citation as a victory just because the company name appeared. The cited source has to support the claim being made. If it does not, the repair is usually practical: rewrite the entity description, align German and English pages, clean directory language, clarify schema, or move the strongest proof onto a page that answer engines can actually read. AI SEO, to me, is evidence work. The business becomes visible when the record becomes hard to misread.