The first, from researchers at King's Business School and Emory's Goizueta School, maps out three types of AI-agent interactions brands will face. There are brand agents — AI systems a company builds to engage with human customers directly (think Capital One's Auto Navigator, which checks dealership inventory, schedules test drives, and answers financing questions). There are consumer agents — AI systems that act across many brands on behalf of a shopper. And then there's the emerging frontier of full AI intermediation, where agents on both sides of a transaction negotiate with each other.

The same researchers outline three stages of adoption that every brand needs to think through. First, decide whether you even need an agent. Not every product category benefits from automation — Lamborghini's CEO made the point that people buy a Lamborghini to drive it, not to be driven in it. Some customer journeys are the product. Second, if you do build or participate in agent ecosystems, figure out how to get customers to actually use them. Sephora is a strong example here: their proprietary data across 140,000 skin tones and 34 million loyalty members makes their AI tools genuinely useful, and customers who use those tools are three times more likely to purchase. Third — and this is the big strategic shift — make other AI agents choose your brand even when no human is directly involved in the decision.

The second framework, from a PwC team writing in HBR, focuses on trust. Their research found that 64% of consumers need at least one concrete guarantee (like a money-back promise) before they'll let an AI agent buy on their behalf. Their five-step approach includes structuring product data for machines (not just humans), defining clear boundaries around what agents can and can't do, protecting customer data visibly, monitoring how your brand appears in agent ecosystems, and planning for recovery when things go wrong, because they will. As the authors put it, the response matters more than the error.

The Walmart Lesson

One of the most instructive real-world examples so far is Walmart's partnership with OpenAI. Walmart embedded its shopping agent into ChatGPT, making it possible for users to discover and explore products without ever leaving the conversation. But here's the thing: when OpenAI first tested in-chat checkout, conversion rates ran roughly three times lower than Walmart's own website.

The takeaway? AI platforms are excellent for discovery — helping people find and evaluate products through conversation. But when it comes to actually completing a purchase, consumers still prefer the familiarity and control of a brand's own environment. OpenAI subsequently pivoted ChatGPT toward product discovery rather than trying to own the entire transaction.

For marketers, this suggests a practical split: use AI platforms to be found and recommended, but keep the conversion experience in your own environment where trust is already established.

Where to Start

If this feels like a lot, here's a grounded sequence of steps.

Right now, do the same diagnostic exercise that works for GEO: ask AI assistants the kinds of questions your customers ask, and see whether your products come up, how they're described, and whether the information is accurate. This is free, takes an hour, and the results are almost always illuminating.

Over the next one to three months, focus on making your product catalogue machine-readable. That means structured attributes, detailed specifications, clear availability data, and consistent formatting. Adobe found that the average US retail product page scores just 66% on AI readiness — meaning a third of the information agents need to make good recommendations is missing or poorly structured.

Over the next three to six months, decide your strategic posture: are you building your own branded agent, optimising to be chosen by third-party agents, or both? Stand up new metrics — AI citation share, agent-referred conversions, agent interaction completion rates — and assign cross-functional ownership.

The Bigger Shift

Underneath all the tactics, there's a more fundamental change worth naming. For decades, marketing has been about winning the attention of humans — through creativity, emotion, storytelling, and persuasion. Those skills aren't going away. But a new audience has entered the room: the algorithmic intermediary. And that audience doesn't respond to clever headlines or beautiful imagery. It responds to structured data, accurate specifications, verifiable claims, and machine-readable trust signals.

The brands that will thrive in this environment aren't the ones that choose between marketing to humans and marketing to machines. They're the ones that learn to do both — and understand when each matters.