Who Is Asking, Who Is Answering?

The company where I work has just launched a site for an American audience. On 4 November I opened it and stopped, within a minute, at the FAQ block. Six questions. The fourth one reads: "How fast will I see results?"

The third one reads: "Which AI systems do you work with?" That sentence can be taken two ways. A visitor may be asking the company. Or the site may be asking the visitor, the way the rest of the page talks to the reader about the reader's own business. Both readings fit. After the second, the fourth question turns the conversation around: the site was addressing the reader as "you", and now an "I" appears, as if someone else had taken the floor.

My rule is short. A block like this keeps one formula. "I" works well in an FAQ, because these are the questions a client asks in their own head, and the page only gives them a place to be written down. If the site speaks to the reader as "you", it speaks that way to the end. A switch in the middle means the site has started to speak in the client's place.

Coherence has always been counted among the norms of clear speech. The voice is part of it: a text that changes who is speaking has broken its own logic, even when every sentence in it is correct.

The suggestion is to write the fourth question without a pronoun, "How fast to expect results?" Other ways are "How soon will the results be visible?" and "How soon do results appear?" The one thing it must not be is "will I see". Any single voice would do. What matters is that there is one.

The same six questions also sit in the page's markup, as a list of questions and answers for machines. A model reads them without the layout, one pair after another, and has only the pronouns to work out who is speaking. A person on the page has the layout to help, and still stumbles for a second.