Product truth, written for extraction

A product page printed on thin paper sits beside my keyboard when I work. I mark every sentence that an answer engine could lift without embarrassment, then I mark the ones that would collapse if they were separated from the paragraph around them. The second pile is usually larger. It has feature names with no buyer, claims with no boundary, pricing language that hints at enterprise without saying why, and French and English pages that describe the same product with two different levels of courage.

I am from France, and I write under the name Elian Soryve for an artistic-informational project about how B2B and technology firms become legible to large language models. My earlier work moved between technical positioning pages, bilingual product explanations, B2B search journey audits, practical web-structure teaching and advisory work for software teams whose documentation and marketing had started to argue with each other. That background left me with a habit I have not managed to drop: I keep a private quote map of how the same company is described by its homepage, docs, pricing page, changelog and AI answers. The mismatches are rarely dramatic. They are small, repeatable leaks.

My strongest work is where marketing, documentation and retrieval meet. I help teams define what a product does, what it does not do, where it works, who it serves and which claim is safe to repeat. I care less about ornamental phrasing than about sentence load-bearing. A large language model does not owe a company generous interpretation. It will often reward the page that states a boring fact cleanly before the page that performs sophistication. That is not a moral failure of machines. It is a structural warning to writers.