I read the order before the claim

Three browser windows, a quiet table, and ten versions of the same buyer question: that is usually where the work starts. One prompt says “best supplier,” another says “reliable partner,” another adds “for French mid-market companies,” and the order begins to move. I keep notebooks for that movement. Rows of prompts, answer positions, source fragments, language shifts and competitor names matter more than trophy screenshots. The pattern is often dull at first. Then one phrase repeats, and the dullness becomes useful.

I am from western France, and I have spent sixteen years close to the places where language turns into commercial trust: search strategy, editorial planning, brand evidence audits, B2B positioning, prompt-run comparisons and category-language analysis. I have worked around service companies, industrial suppliers and specialist retailers, the kinds of businesses that may be solid in the market and still strangely thin inside an AI answer. My habit is to read the visible sentence and the evidence behind it at the same time. Why does one brand get “recommended”? Why is another only “also available”? Why does an English answer sound confident while the French answer loses the company completely?

AI prominence is not a magic layer sitting above reputation. It is mechanical enough to be traced, and annoying enough to resist simple fixes. Models reward public evidence that is repeated, current, specific and easy to place beside competitors. They also carry old sources, directory fragments, review surfaces and vague category pages into the answer. My work is to make the order legible: mentions counted separately from recommendations, first-position placements separated from polite inclusion, French prompts tested apart from English ones. A single clean screenshot is a weak story. A repeated answer pattern is where the work begins.