Why I Went All In on Agentic Commerce

It started with a frustrated message from a friend who runs a one-person e-commerce store selling homemade cookies online.

He had built his business on organic search. Years of work. Good rankings, consistent traffic, steady sales. Then slowly, over the past several months, his search traffic started declining.

He knew why traffic was dropping in general. He had read the same articles the rest of us had. People are shifting from Traditional search engines to ChatGPT, Perplexity, and Gemini for product research. Traditional search is losing share to LLMs. That part made sense to him.

What he couldn’t figure out was this: why wasn’t his store showing up in those LLM recommendations either?

He had good products. Strong reviews. A clean website. He was doing everything right for Google. But when he tested his own products in ChatGPT and Perplexity, his store was invisible. Competitors he had outranked on Google for years were getting cited. He wasn’t.

“I’m losing traffic from search,” he told me, “and I’m not picking it up anywhere else. It’s like I’m being erased.”

I didn’t have an answer that day. But I started digging.

What I found changed how I think about the entire future of selling online.

The Funnel You Built Is Becoming Invisible

For two decades, e-commerce worked like this: get ranked on Google, pull a human to your product page, and win the conversion through great design, strong copy, and a fast checkout. That pipeline made a lot of people a lot of money.

Here’s what nobody is saying clearly enough: the buyer is changing.

Not the demographic. Not the intent. The actual buyer.

On February 16, 2026, OpenAI launched native checkout inside ChatGPT for U.S. users. A shopper types “find me a tent for a rainy camping trip next week under $150,” and ChatGPT browses merchant databases, evaluates specs, checks return windows, and processes the transaction without the user ever visiting a website. The feature runs on the Agentic Commerce Protocol, co-developed with Stripe.

At the same time, Google began expanding its Universal Commerce Protocol across Search AI Mode and Gemini.

The transaction rails for autonomous AI purchasing are live. Not coming. Live.

Here’s what this means for my friend’s store: he wasn’t being penalized. He was being ignored. LLMs couldn’t confidently recommend his products because his data wasn’t structured for machine consumption. His return policy was buried in paragraph text. His product specifications were incomplete. His feeds weren’t built for the schemas these systems actually read. The agents scanning for products like his found competitors with cleaner, more complete data and cited them instead.

His Google traffic was declining because human search behavior was shifting. His LLM traffic was zero because he had never been visible there to begin with.

What the Market Data Is Actually Saying

I spent weeks pulling apart the numbers to understand whether this was hype or something real.

The agentic AI retail market hit $60 billion in 2026. The broader agentic commerce sector is projected to scale from $136 billion in 2025 to $1.7 trillion by 2030. Morgan Stanley estimates agentic shoppers alone could drive $190 billion to $385 billion in U.S. e-commerce spending by 2030.

ChatGPT alone now processes an estimated 50 million shopping-related queries every single day, according to a study by OpenAI’s Economic Research team and Harvard economist David Deming. That’s roughly 2% of the 2.5 billion prompts flowing through ChatGPT daily.

But the number that stopped me cold was this one: early brand implementations of agentic commerce are reporting 3 to 4 times lift in conversion rates. Adobe Analytics reported that AI-referred traffic converted 42% better than non-AI traffic in March 2026, an 80-percentage-point swing from a year earlier when AI traffic converted worse than standard traffic.

These are not edge cases. These are the early signals of a structural market shift.

And then there’s Amazon. Amazon has explicitly blocked ChatGPT and OpenAI crawlers in its robots.txt. The reason is straightforward: Amazon generates roughly $56 billion a year from advertising, and that business depends on shoppers browsing Amazon.com rather than buying through ChatGPT. That means every Amazon product listing is invisible to these conversational shopping recommendations right now. Independent stores using open, API-first checkouts can capture this wave completely unhindered by the largest marketplace on earth.

That window won’t stay open forever.

The Shift Nobody Talks About: From Storefront to Data Source

Here’s the mindset shift that took me the longest to accept.

Your website is no longer primarily a destination for your customer. It is a data source for the machine that shops on behalf of your customer.

AI agents do not respond to emotional copy. They do not notice your hero banner. They are not persuaded by social proof widgets or scarcity timers. They scan for structured, machine-readable data and make probabilistic decisions in milliseconds based on completeness, accuracy, and confidence.

If your product data is clean and complete, the agent recommends you.

If it isn’t, the agent moves on without leaving a trace in your analytics.

This is why my friend’s traffic was declining in search but not recovering in LLMs. Traditional SEO optimizes for crawlers that index text and rank pages by authority signals. LLMs don’t work that way. They pull from structured data, product schemas, policy information, and verifiable specifications. Clean, complete, machine-readable data is what gets you cited. Aspirational copy and a high domain authority score don’t move the needle here.

He was invisible in LLMs not because his products were bad, but because LLMs couldn’t read his store confidently enough to recommend it.

What Agent-Ready Actually Looks Like

Once I understood the problem, the fix became clear. It’s not about rebuilding your store. It’s about making your existing data legible to machines.

There are four things that matter most right now.

Product feeds built for machines, not just marketplaces. Every product needs exact identifiers like GTIN, UPC, or MPN values. The OpenAI specification accepts product data updates as frequently as every 15 minutes. If your inventory changes and your feed doesn’t, an agent will generate a cancellation error, and that error gets associated with your store’s reliability score.

Policies as structured constraints, not paragraph text. Your return window, shipping thresholds, and geographic restrictions need to exist as data points, not buried in a help center article. An agent reads your policies early in the evaluation process. If it can’t parse them, it flags your store as high-risk and moves on.

Zero variant ambiguity. Size charts, compatibility tables, and configuration options must be deterministic. If a product comes in three sizes but your data doesn’t specify which size ships in what timeframe, the agent can’t complete the purchase with confidence. It requests human clarification, which increases drop-off. Competitors with clean data win that transaction.

Dual-protocol implementation. Right now, both ACP and UCP are gaining traction simultaneously. Data shows that stores running both protocols capture 40% more agentic traffic than single-protocol stores. You don’t need to pick a winner yet. You need to be present in both ecosystems.

None of this replaces good product photography or strong copy. Human shoppers still matter enormously. But the brands that build for machine legibility now will have a compounding advantage as agent-driven purchasing scales.

What I Told My Friend

I called him back after three weeks of research.

“Your traffic problem isn’t an SEO problem,” I said. “It’s a machine readability problem. LLMs are evaluating your store and can’t find enough structured, confident data to recommend your products. You’re invisible in the channel that’s replacing the one you built your business on. You won’t see it in your rankings. You’ll see it in traffic that keeps declining while your content stays exactly the same.”

He went quiet for a moment.

“So how do I fix it?”

The answer is the same for any store owner reading this: start by auditing what an AI agent actually sees when it encounters your store. Check your product feeds for completeness. Convert your policies to structured data. Eliminate any variant ambiguity. Then get your checkout onto an open protocol.

This is not a distant transformation. It is happening in the market right now. The retailers who treat agent readiness as an infrastructure priority today will be the ones capturing the demand that scales over the next three years.

The ones who don’t will keep watching their conversion rates drift and wonder what changed.

If you’re thinking about where to start with agent readiness for your store, I’m happy to talk through it. The checklist above is a reasonable first pass, but the specifics depend on your stack, your feed setup, and how your policies are currently structured.