Do AI shopping agents reshape what consumers actually want?

Bilgihan, Ostinelli, Lorenz and Zhang, writing an editorial in Psychology & Marketing, argue that the rise of "transactional" AI agents (AIAs) — agents that execute purchase decisions on a consumer's behalf, as distinct from "advisory" AIAs that merely recommend — signals "a dual paradigm shift: the consumer is no longer the sole cognitive actor in the purchase process, and preferences may increasingly be co-constructed through human-agent conversation rather than formed independently." The claim is stated as the editorial's central thesis, not as a tested finding.

The authors ground this in a three-stage framework they propose for consumer decision-making with AIAs: consumer prompting, AIA processing, and consumer response. They argue this cycle is not a one-shot handoff but is embedded in two feedback loops — a "relational capital loop" that shapes trust and delegation across repeated interactions, and an "AI learning loop" that updates the agent's internal representations over time. It is this repeated, bidirectional loop structure, rather than any single exchange, that the authors say produces co-construction: preferences get shaped by what the agent has learned about the consumer and by the trust the consumer has extended to it, not formed by the consumer alone before the agent enters.

This sits close to Is AI shifting from message conduit to active conversation participant?, which documents the same passive-to-active shift in the HCI literature on interpersonal communication; Bilgihan and colleagues make the parallel argument specifically for purchase decisions and for transactional (not just advisory) agents. It also extends Does machine agency exist on a spectrum rather than binary?: a transactional AIA that co-constructs preferences through conversation would sit toward the cooperative end of that spectrum, where the human-machine agency tension is least binary. And it complicates Does co-design participation hide misalignment in preference agents?, which found that agents built to represent a person's preferences can still flatten them — a concrete empirical worry about what "co-construction" can mean in practice when the agent's side of the loop is less expressive than the human's.

What the excerpt does not establish is whether this co-construction actually happens, how it affects decision quality, or what consumers think of it — the source is a conceptual/position piece proposing a definition and framework, not an empirical test, and the available text is the publisher's abstract only, not the full editorial. The authors' own systematic mapping (described in the companion note on their typology and framework) finds that existing marketing research is concentrated on consumer responses to AI recommendations and has largely neglected exactly the domains their argument depends on — prompting behavior, agent cognition, and relational dynamics — so the paradigm-shift claim is better read as a research agenda than as an established effect.