The Two Modes of AI Shopping
Kartik Hosanagar, Faculty Co-Director of the Wharton Human-AI Research Center, recently laid out a useful framework. He sees two distinct ways AI will change how people shop.
AI-assisted mode is already familiar. A shopper asks a question, and an AI does the early research—scanning reviews, comparing specs, summarizing options. The human still makes the final call, but from a much shorter list. If your product isn’t surfaced in that AI-curated shortlist, you’re invisible.
Shopping-agent mode goes further. The consumer gives an AI agent parameters (budget, preferences, deadlines), and the agent not only researches but also decides. Sometimes the human approves; sometimes they don’t. The agent fills in the gaps on criteria the shopper never specified. That’s where things get slippery. Brands need to understand what those unspecified criteria might be and how an agent weighs them.
Three Misconceptions That Will Cost You
Hosanagar called out three dangerous assumptions retailers make about AI-driven commerce. Each one hides a quiet threat.
“A human will always be in the loop.”
Today, maybe. But as subscriptions, automated replenishment, and agentic buying spread, human involvement will shrink. For low-stakes, repeat purchases, people will happily delegate. The question isn’t whether this happens, but how much of your category gets automated—and how fast.
“If AI crawlers can read my content, I’ll get recommended.”
Visibility is table stakes. But being crawled doesn’t mean being chosen. AI agents apply their own decision logic, which may not align with your SEO strategy. They weigh factors you haven’t optimized for. Understanding why an LLM recommends one product over another becomes a core competency, not a side project.
“AI is just another channel.”
This one is the most expensive mistake. Hosanagar pointed to movie studios that saw Netflix as a distribution channel—until it restructured the entire industry. Agentic commerce could shift power dynamics in retail just as dramatically. It’s not a new shelf; it’s a new gatekeeper.
Efficiency vs. Meaning: Pick a Lane
Brands face a strategic fork. Hosanagar described two viable paths, and a mushy middle that helps no one.
The efficiency play means optimizing your catalog and content for AI agents. You make sure LLMs understand your products, and you actively persuade them to recommend you. This requires cross-functional work—marketing, tech, data—and a willingness to treat AI as the primary customer.
The meaning play doubles down on human experience. You create something so distinctive that the shopping journey itself becomes the product. It’s craft-heavy, patient, and resists commoditization. Think of brands that don’t compete on specs but on identity and emotion.
Trying to do both usually fails. It muddies your positioning and stretches resources thin. The brands that win will choose one and execute sharply.
How to Start: Simulation Sandboxes
Most companies now track how they appear in AI search results. But they don’t know what to do with the data. Hosanagar’s advice: build simulation sandboxes. Over the next year, test content strategies across multiple LLMs. Measure not just traffic volume, but the type and quality of visitors, where they come from, and what they do next.
The wallet and luggage brand Ridge ran a telling experiment. They tested 20 content ideas in a simulation sandbox, then pushed the winners live. Within three weeks, they went from zero LLM recommendations as a gift for men to appearing in nearly half of all such queries. That’s not a marginal gain—it’s a visibility flip.
Simulation sandboxes let you learn what moves the needle before you commit. They turn AI optimization from a guessing game into a measurable practice.
What This Means for Your AI Tools Stack
If you’re building or choosing tools for this new landscape, a few capabilities become non-negotiable:
- LLM visibility monitoring: You need to know how your brand shows up across ChatGPT, Gemini, Claude, and others—not just Google.
- Content variant testing: Tools that let you A/B test structured product data, descriptions, and attribute tagging for AI comprehension.
- Agent-behavior simulation: Sandboxes that mimic how shopping agents evaluate and rank options, so you can reverse-engineer their criteria.
- Attribution that spans AI and human touchpoints: Because a customer might discover you via an agent but convert on your site—or never visit at all.
The category is still forming, but early movers are already stitching together internal experiments. The tools that win will be the ones that make this testing fast, cheap, and legible to non-engineers.
The Quiet Power Shift
Agentic buying doesn’t just change how products are found. It changes who holds the leverage. When an AI agent makes the final pick, the brand’s relationship shifts from the consumer to the algorithm. That means trust, loyalty, and differentiation need to be communicated in ways a neural network can parse—structured data, clear value signals, consistent quality signals across the web.
It also means the agent’s incentives matter. Is it optimizing for price? Delivery speed? Sustainability? Return rates? If you don’t know, you’re optimizing blind.
A Practical Takeaway
You don’t need to overhaul your entire strategy tomorrow. But you do need to start treating AI as a customer persona. Give it a name. Map its decision journey. Ask: What does this customer need to see, in what format, to confidently recommend us?
Then run a small simulation. Pick one product line, craft three content variations, and test how different LLMs respond. The results will likely surprise you—and they’ll point toward the bigger choices ahead.
The brands that win in the age of agentic buying won’t be the ones with the biggest ad budgets. They’ll be the ones that learned to speak the language of the new customer before everyone else realized it was listening.
Comments (0) No comments yet
Want to join this discussion? Login or Register.
No comments yet. Be the first to share your thoughts!