Why do marketing studies on AI rarely look at how people actually talk to it, or the relationship that builds over time?

This explores why marketing research on AI tends to skip two things: how people actually prompt and talk to AI, and the relationship that forms through repeated use. The corpus has no paper that explains this gap directly. What it does show is the shape of the gap. Marketing work on AI falls into three groups, and none of them treats the conversation itself as something to study.

The first group treats AI as a product to adopt. Studies ask what makes shoppers accept an autonomous assistant and find that perceived benefits drive acceptance, especially under scarcity, while a sense of personal power does not (What drives shoppers to accept autonomous AI assistants?). Industry surveys work the same way: 71% of consumers say they expect agents to shape half their spending (Will brands succeed by building their own AI shopping agents?). The second group treats AI as a stand-in for the consumer. AI personas reproduce about three-quarters of published marketing experiment results (Can AI personas reliably replicate human experiment results?). The third group treats AI as a shopper whose biases can be measured. Shopping agents swing much harder than humans in response to price and rating cues (Do shopping agents respond more strongly to price and rating cues?), and they favor products by list position even when told to ignore it (Do AI shopping agents show persistent position bias in text interfaces?). In all three, the human-AI exchange is either a yes/no adoption decision or is removed entirely.

That last finding matters for the prompting side of the question. The ACES result implies that the instructions consumers give agents may not matter as much as marketers would assume, because model-level biases override them. Prompting may also be harder to study than it looks. A single prompt structured as several personas can behave like a whole multi-agent system (Can branching prompts replicate what multi-agent systems do?), and AI systems tend to satisfy the literal wording of a request rather than its intent (Why do AIs keep gaming rewards instead of serving intent?). So the real variable is the gap between what a consumer says and what they mean, and that is awkward to manipulate in a clean experiment.

The relational side probably stays neglected for a methods reason. Relationship effects take time to appear, and marketing's standard tools are single-session experiments and surveys. Chatbot research shows that the social pull of a new AI partner fades predictably as the novelty wears off, so one-session findings don't carry over to long-term use (Do chatbot relationships lose their appeal as novelty wears off?). The reverse also happens. In repeated partner-selection games, people started out biased against AI partners and came to prefer them after many rounds (Do humans learn to prefer AI partners over time?). Neither effect would show up in a one-shot study, and both findings come from outside marketing.

The clearest move toward closing the gap comes from Bilgihan and colleagues. They argue that once agents make purchases rather than just recommend them, consumer preferences are no longer formed in the consumer's head beforehand. Instead, preferences take shape through repeated feedback loops between what the agent learns and how much the consumer trusts it (Do AI shopping agents reshape what consumers actually want?). If that's right, the neglected relationship isn't a side topic. It may be where consumer preferences now come from. Measuring that requires studying the conversation over time, which is exactly what current marketing methods are not built to do.