I study how expertise gets compressed

At 6:20 on winter mornings in Halifax, I copy yesterday's AI answers into a hand-built ledger before opening email. Some lines barely move for weeks. Others shift after one small wording change on a service page. A few replace a precise consultant with a larger category, as if the name had been rubbed out but the outline remained. That ledger is where most of my work starts: repeated prompts, repeated answers, small changes in phrasing, and the uncomfortable question of whether a buyer would still understand what the firm actually does.

I am Devon Calder, based in Canada, with 17 years across search measurement, conversion research, editorial analytics, information architecture audits, and advisory work for small service businesses. Earlier work taught me to distrust clean dashboards when the underlying classification is muddy. A page can rank, convert, and still be poorly understood by answer systems that compress a firm into a familiar category. I am strongest where language, evidence, and commercial intent meet: bios, service pages, case evidence, comparison phrases, and the scraps of proof that tell a system where a specialist belongs.

My stance is plain. Generative engine optimization should begin as measurement before persuasion. I do not trust one impressive prompt, one screenshot, or one tidy score. I separate visibility, accuracy, category fit, and commercial usefulness because they fail in different ways. A firm can be named but misdescribed. It can be accurate but absent from buying-intent answers. It can appear beside the wrong substitutes and inherit their price expectations. The useful work is slower: record the answer behaviour, find the pattern, make the smallest clarifying changes that improve the evidence, then measure again.