Part five

What we don't know yet

This book leans on evidence that's young, and much of it comes from firms with something to sell. These gaps matter most.

Habitual buying. No study I found shows AI changing which brand people buy on habit.
Memory and machines. No published study tracks a brand from building its name to being recommended more by AI. The evidence is correlational, or uses invented brands. Nor has anyone shown that moment-based memory measures predict sales better than prompted awareness.
How many runs are enough. No independent study says how many AI answers make a stable rate. The margins in Chapter 14 and the appendix are arithmetic, not evidence.
Which answers decide. No study measures which questions settle a purchase. Nobody has tested whether answering price, risk or comparison questions changes how often AI recommends you.
Consistency and the one truth. No study links consistency across sources to AI recommendation or to sales. The best papers suggest machines pick one version rather than flag a conflict. A structured brand record improves the answers you generate yourself, but nobody has shown that outside assistants reward it. Studies of what AI cites disagree widely, from about 1% to 44% for brands' own sites.
Reasons. No study links a reason that customers repeat to a higher win rate. Nor do we know whether assistants repeat a brand's own reason, or whether the reason a machine gives is the one that decided.
The last step. The size of "no decision" in business buying comes from one proprietary dataset. Most guarantee evidence comes from lab studies of intention, not from real purchases.
Keeping and the machine. No study links a firm's churn or service failures to a later fall in AI recommendation. I found no business-to-business evidence on AI use at renewal.
Challengers and small firms. I found no verified case of a challenger displacing a bigger rival in AI answers through better evidence. Tally comes closest, and its numbers are its own. I found no data on how owners choose a bookkeeper, an agency or a lawyer, with or without AI.
Cause and effect. No randomised experiment yet links an AI recommendation to an actual purchase. No vendor or buyer has described, on the record, a large deal won or lost because of what an AI said.
The score itself. Which stages carry each mode is my inference. The weights in Chapter 30 are judgement, and not yet fitted to sales.
The score against share. Nobody, including me, has yet shown that a higher Selectability Score goes with more market share or more sales. Chapter 30 sets out the test, and we're running it.
How independent AI answers are. The margins in this book assume each answer is a separate draw. Answers to the same phrasings, from the same assistants, aren't. Nobody has measured by how much that widens the real margin.
Most of the world. The company stories here come from North America, Europe, China and Australia. I found nothing verified elsewhere.

Most of the surveys cited were run by vendors, and each chapter's note says whose evidence it is. Almost everything known about how marketers are faring comes from surveys of opinion, not measured results. The evidence was reviewed in October 2026, and it'll date quickly.

Every gap on this list can be closed, and most need data no single firm holds. If you score your moments, share what you find. Send your scores beside your sales through rocmode.com/research, and the findings will be published. The worksheet and every tool in this book are free at rocmode.com/playbooks.