Why AI decides who gets chosen

For twenty years search produced lists: ten results, and the person did the evaluating. Today someone asking “which is the best window company in Milan” does not get ten links: they get two or three names, with a reason. The comparison has already been made, by someone else, before the user could take part in it.

An answering system does not sort pages: it builds a synthesis from the sources it considers clear and reliable. A page can rank first on Google and never be cited, because it is written to be skimmed by a human eye rather than extracted by a model.

• A list rewards position; an answer rewards quotability.
• A list tolerates ambiguity, because the user digs deeper; an answer discards it, because it must be short.
• A list shows ten options; an answer shows three, and the rest simply do not exist.

AI does not “visit” a site the way a person would: it extracts verifiable statements. Pages that say precise things, explicitly, in one place, are at an advantage.

• Clarity of offer: what you do, for whom, where. Written, not implied.
• Structure: real headings, structured data, information not buried inside images.
• Consistency: the same data repeated identically across site, listings and external citations.
• Self-sufficiency: every paragraph must stand up even when pulled out of its context.

When an answer excludes a company, it does not penalise it: it makes it invisible. The user does not know it exists, so does not compare it, so does not choose it. And because every answer reinforces the sources it already cites, the gap tends to widen on its own over time.