When the Buyer Is an AI Agent, Who Are You Marketing To?

The Buyer Has Left the Building

Google declared in February 2026 that "agentic commerce is no longer just a concept, it's reality." That sentence should unsettle every marketer who still thinks of a purchase as a moment between a human and a screen.

The thesis is simple and uncomfortable: a growing share of buying decisions are now being made by software agents, not people. Marketing built to persuade humans, compelling copy, emotional triggers, striking creative, may not reach the entity actually choosing. If an AI agent is filtering, comparing, and selecting on a buyer's behalf, the levers that worked for decades stop working. Not because they're bad. Because the audience changed.

None of this is a future threat to plan for. The infrastructure is already live. Google launched its Agent Protocol 2 and Universal Commerce Protocol to handle secure identity, payments, and agent-to-merchant connections, and launched Business Agent and Direct Offers features within AI Mode in January 2026. Microsoft Advertising published guidance in May 2026 telling brands to get "agent-ready." Accenture put it plainly: "nearly all companies will need to structure their offerings to be seen and chosen by AI agents."

Brands that treat this as a future problem are already behind.

What AI Agents Actually Evaluate

Here's what makes agentic commerce genuinely different from anything that came before. An AI agent shopping on a user's behalf doesn't browse the way a human does. It doesn't respond to a clever headline. It doesn't feel urgency from a countdown timer. It reads structured data, matches product attributes against stated requirements, checks that what your feed says matches what your page says, and makes a decision based on compatibility with the user's stated or inferred preferences.

Microsoft Advertising was direct about this: the core requirement is that "what the AI sees in your feed matches what it sees on your page." That sounds like basic hygiene. In practice, most brands fail it. Stale feeds, missing attributes, inconsistent categorization. These aren't minor technical issues. They're disqualifying. Koddi frames this as a shift from competing for visibility to competing for inclusion within recommendation systems. You can't be considered if you're not readable.

That represents a fundamentally different performance problem than anything in the traditional marketing funnel. Traditional performance marketing optimized for clicks, CTR, conversion rate. The agent-era equivalent is whether your product data is complete, accurate, and structured in a way that machine reasoning can evaluate. That's not a creative problem. It's an infrastructure problem.

eMarketer identifies the four priorities brands need right now: Generative Engine Optimization (GEO), AI ad format experimentation, protocol monitoring, and first-party data investment. Notice what's missing from that list: ad creative, copy testing, audience segmentation. Not because those things are irrelevant, but because the prerequisite layer, machine-readable product truth, has to come first.

The Brand Problem No One Is Talking About

Fix your feed, make your data agent-ready, and you've solved about half the problem. The other half is harder.

AI agents don't just retrieve product data. They synthesize brand understanding from across the internet. They learn from reviews, forum discussions, third-party comparisons, and press coverage. When an agent is deciding between two products that both have complete, accurate feeds, it falls back on something much closer to brand reputation. And that reputation was built, or not built, long before any agent entered the picture.

Patrick Gilbert's treatment of Answer Engine Optimization in Never Always, Never Never makes a point that applies equally here: LLMs don't need links to learn about your brand. They learn from the consistency, specificity, and volume of signals across independent sources. A brand with a vague or muddled story gets smoothed into the background. A brand that is distinctive, consistently described, and well-represented across third-party sources gives the agent something concrete to work with.

Mental availability, the probability that a brand comes to mind in a buying situation, was always a human cognitive phenomenon. The Ehrenberg-Bass Institute built an entire framework around it. Byron Sharp's research established that brands grow by reaching more buyers and being mentally and physically available when those buyers are ready to purchase. Agentic commerce doesn't eliminate mental availability. It extends it to machines. The agent needs to recognize your brand as a legitimate option for a given context, which means the signals it learned from need to be consistent and unambiguous.

If the internet is inconsistent about what you do and who you serve, an agent won't resolve that ambiguity in your favor. It will either describe you incorrectly or omit you entirely.

AdVenture Media's team encountered this firsthand when a prospective client discovered the agency through ChatGPT, only to receive a confident but fundamentally wrong description of what the agency does. The model had stitched together partial truths from across the internet into a coherent-sounding narrative that didn't match reality. That's not a hallucination problem. It's a brand signal problem. And it illustrates why AEO marketing isn't a search tactic. It's a brand consistency problem with new consequences.

Protocol Wars and Why They Matter

Google's Universal Commerce Protocol isn't a product feature. It's an attempt to become the standard infrastructure for how AI agents transact. If UCP becomes the dominant protocol for agent-to-merchant connections, then being compatible with it is roughly equivalent to being indexed by Google in 2005. Opt out and you cease to exist for a large portion of agent-mediated commerce.

This is the competitive battleground that most marketing teams aren't watching. eMarketer and Deloitte both flag protocol adoption as a strategic decision, not a technical afterthought. Amazon and Walmart are already testing sponsored prompts and AI search ads, building their own monetization layers on top of agent-mediated discovery. The platforms are moving fast. Brand infrastructure teams are not.

For AI marketing strategy, budget allocation needs to account for a layer most marketers haven't funded: agent readiness. That means clean, complete, consistent product data. It means monitoring which protocols your distribution channels support. It means first-party data investment, because as agents reduce reliance on traditional tracking, brands that own their customer data maintain signal quality while others go dark.

None of this replaces brand building. It runs alongside it.

The Measurement Problem Is Going to Hurt

Here's the part of agentic commerce that almost no one has a good answer for yet: how do you measure marketing effectiveness when the buyer is an agent?

Traditional attribution traces a human through a funnel. Click, visit, conversion. Even the incrementality vs attribution debate assumes a human making a decision at some point. Agent-mediated transactions may not leave the same trail. An agent that purchases on a user's behalf might bypass ad tracking entirely. Sponsored prompts on AI platforms are early experiments in building new measurement frameworks, but they're experiments. The industry hasn't solved this.

Les Binet and Peter Field spent years documenting how brand investment pays back over time in ways that short-term attribution can't see. Their work through the IPA DataBank showed that the long-term effects of brand advertising are systematically undervalued precisely because they don't show up in direct response metrics. The 60/40 brand performance split they recommended was partly a corrective for this measurement blind spot.

Agentic commerce creates a new version of the same problem. If an agent selects your product based on brand signals it learned from third-party content, reviews, and consistent positioning, and that selection never touches a trackable ad unit, the contribution of your brand investment becomes invisible to performance dashboards. Marketers who cut brand spend because it's "hard to measure" will eventually face a world where their product data is agent-readable but no agent considers them worth recommending.

A new attribution model won't fix this. Investing in brand clarity and reach before the measurement problem fully arrives is the answer, accepting that some of the return will only become visible later.

What to Actually Do

Agentic commerce doesn't require abandoning everything that works. It requires adding a layer and protecting what you already built.

On the data and infrastructure side, the immediate priorities are clear. Audit your product feeds for completeness and accuracy. Make sure your on-page content and feed data match. Monitor which agent protocols your major channels support and plan for compatibility. Invest in first-party data now, before tracking degradation makes external signals unreliable. These aren't marketing decisions. They're operations decisions that marketing leaders need to push for.

On the brand side, the work is the same as it always was, but the urgency is higher. Distinctive brand assets need to be clear enough that a machine learning from the internet can learn them too. Your positioning needs to be consistent enough across independent sources that an agent synthesizing signals from reviews, forums, and press coverage arrives at an accurate understanding of what you do and who you serve. A muddled brand story in the human-readable internet will be equally muddled in the machine-readable internet.

Klaviyo documented a concrete example of what the agentic layer can do when it works: Naked Wardrobe's AI customer agent resolved 86% of customer queries and 94% of product recommendation requests over a 90-day period. That's not a marketing outcome in the traditional sense. It's a distribution outcome, a brand present and readable at the moment an agent needs to make a recommendation.

That's the frame that makes agentic commerce legible. Physical availability has always meant being easy to find and easy to buy. AI agents are becoming a new distribution channel. Being available to them requires the same logic: show up where decisions are made, in a form that the decision-maker can process.

Increasingly, that decision-maker is a machine. Build accordingly.

---

Patrick Gilbert covers the intersection of AI-driven search, brand visibility, and the evolving answer stack in [Never Always, Never Never: Strategic Marketing in an AI World](https://neveralwaysbook.com). The chapters on the AI Answer Stack and the future of AI-driven search are directly relevant to anyone building strategy for the agentic era.