Vercel Built AEO Tracking for Coding Agents. Here's Why Shopping Agents Are the Bigger Opportunity.

First — Why AEO Is Not Just "SEO for AI"

SEO was about one thing: ranking on Google Search. You optimized keywords, built backlinks, and fought for page-one visibility in a single channel with well-understood rules.

AEO — Answer Engine Optimization (also called GEO or AI Visibility) — is a fundamentally different problem. It is not about ranking in one search engine. It is about how your brand shows up across an entire ecosystem of AI agents — ChatGPT, Gemini, Perplexity, Claude, Copilot, shopping agents, coding agents, voice assistants, and dozens more that don't exist yet. Where and how AI agents get used is still evolving, but the trajectory is clear: this will be far, far bigger than Google Search ever was.

What Vercel Got Right

Vercel — a developer platform — recently published a detailed breakdown of how they built AEO tracking for coding agents. It is one of the most important pieces of AI infrastructure writing this year.

The core insight: when developers use AI coding assistants like Claude Code or OpenAI Codex, the AI doesn't just give advice — it makes the decision and executes it. If a developer asks the agent to "set up a deployment pipeline," the AI picks a platform and immediately configures the project to use it. The brand the AI chose is now baked into the developer's project. There is no "consideration phase." The recommendation is the purchase decision.

Vercel recognized that this was a visibility problem: which brands are AI agents actually picking, and why? So they built a system to systematically test real AI agents with real prompts and track exactly which products get recommended. They call it AEO tracking.

They got the architecture right. The methodology is sound. The problem they're solving is real.

But there is a much larger version of this problem that most brands are completely blind to.

Shopping Agents Are the Same Problem — at 1,000x Scale

Everything Vercel observed about coding agents is now happening in commerce. Except the stakes are exponentially higher.

When a consumer asks ChatGPT "What's the best running shoe for flat feet under $150?" — that AI doesn't return a list of blue links. It returns a curated recommendation with specific brands, specific models, and increasingly, a direct purchase button.

Consider what's already live:

• Google's "Buy for Me" — agentic checkout across Google Search AI Mode and Gemini. The AI agent executes the purchase directly on merchant websites.
• ChatGPT Shopping — OpenAI's product research engine using GPT-5 mini with reinforcement learning. Comparative product guides. Real-time feedback loops.
• OpenAI's Agentic Commerce Protocol — co-developed with Stripe. Shopify is building cross-merchant cart infrastructure on top of it.
• Perplexity Buy — one-click purchasing embedded directly in AI search results.

These aren't prototypes. These are live, shipping products that are actively reshaping how consumers discover and buy things.

The Numbers Tell the Story

The data on agentic commerce is no longer speculative. It is overwhelming:

Metric	Value	Source
AI recommendation conversion rate vs traditional search	4.4x higher	McKinsey
Consumers using AI in shopping journey	73%	McKinsey
Users who prefer AI search over traditional search	44%	McKinsey
Consumers who trust AI results more than traditional ads	41%	IAB
E-commerce orgs with reduced customer acquisition costs via AI	76%	Pattern Group
Brands already deploying AI shopping agents	1 in 3	Pattern Group
Global agentic commerce opportunity by 2030	$3–5 trillion	McKinsey

Compare this with the developer tools market. Vercel's AEO tracking matters — but the total addressable market for "which deployment platform does Claude Code pick" is a rounding error compared to "which running shoe does ChatGPT recommend."

The Attribution Black Hole

Here's the part that should terrify every CMO.

In traditional e-commerce, retailers see everything: impressions, clicks, dwell time, add-to-cart events. The funnel is fully instrumented.

In agent-mediated commerce, the behavioral data stream starts at the add-to-cart moment. The discovery phase, the consideration phase, the comparison — all of that happens inside ChatGPT, inside Gemini, inside Perplexity. Your analytics platform never sees it.

Attribution collapses. The middle of the funnel disappears into a black box you don't own.

Vercel solved this for their domain. They built sandboxes, ran the agents, and captured what happened. But most e-commerce brands have zero visibility into what shopping agents say about them.

Ask yourself:

• When someone asks ChatGPT for the "best moisturizer for dry skin," does your brand appear?
• When Gemini compares your product against competitors, what does it say about pricing, quality, or reviews?
• When Perplexity generates a "top 5" list in your category, are you on it?
• When an AI shopping agent autonomously builds a cart for a consumer, does it include your product — or your competitor's?

If you don't know the answers, you're flying blind in a channel that converts at 4.4x the rate of Google Search.

What Vercel's Architecture Teaches Us About Shopping AEO

Vercel's approach has a clean, generalizable architecture. Let's map their six-step sandbox lifecycle onto shopping agents:

Vercel's Step (Coding Agents)	Shopping Agent Equivalent
1. Install the agent CLI	Configure AI model access (ChatGPT, Gemini, Claude, Perplexity)
2. Inject credentials via AI Gateway	Route prompts through unified gateway for logging and cost tracking
3. Run the agent with a prompt	Run shopping queries: "best [product] for [use case] under [$price]"
4. Capture the transcript	Capture full response: brands mentioned, ranking position, sentiment, links generated
5. Normalize the output	Extract brand mentions, product attributes, competitive comparisons, purchase intent signals
6. Grade brand extraction	Score: AIScore, Share of Voice, sentiment grade, citation quality

The architecture is the same. The domain is different. And the business impact is orders of magnitude larger.

Shopping Agents Are Harder to Track Than Coding Agents

Vercel acknowledged that coding agent transcripts are messy. Shopping agent tracking is even harder, for three reasons:

1. The recommendation surface is wider

A coding agent picks from maybe 20 deployment platforms. A shopping agent picks from millions of products across every category imaginable. The brand extraction problem is combinatorially more complex.

2. Context changes everything

Ask a coding agent to "deploy a Next.js app" and the context is relatively stable. Ask a shopping agent for "the best birthday gift for my mom" and the response depends on budget, location, stated preferences, past conversation history, and whether the agent has access to real-time inventory. The same brand can appear or vanish depending on how the question is framed.

3. The transaction layer is live

Coding agents generate code. Shopping agents generate purchases. Google's Buy for Me, OpenAI's Agentic Commerce Protocol, and Perplexity Buy are closing the loop between recommendation and transaction. If a checkout flow can't accept an authenticated agent, the agent picks a merchant it can complete the cycle with. Your checkout infrastructure is now a ranking signal.

What This Means for Your Brand

The shift from "consumer searches, consumer clicks, consumer buys" to "consumer asks, agent recommends, agent buys" is already underway. Here's what smart brands are doing right now:

Track your AI Share of Voice

You can't optimize what you can't measure. Run your category-defining queries across ChatGPT, Gemini, Claude, and Perplexity. Document which brands appear, in what order, with what sentiment. Do this weekly — AI model updates change recommendations constantly.

Optimize your structured data

Shopping agents rely heavily on structured product data — Schema.org markup, clean pricing information, availability signals, review aggregates. If your product pages are optimized for humans but unreadable to agents, you're invisible.

Build agent-ready checkout

Google, Stripe, Mastercard, and Visa are aligning around the idea that agents must be first-class transaction actors. If your checkout flow can't be completed by an AI agent, you will lose sales to competitors whose checkout can.

Monitor the competitive landscape

When an AI shopping agent recommends your competitor instead of you, understand why. Is it a content gap? A review sentiment problem? A structured data issue? A pricing disadvantage? Each of these is fixable — but only if you can see it.

The Bottom Line

Vercel built AEO tracking because they understood that if coding agents don't recommend you, developers don't use you. The same logic applies to commerce — except the consequences are measured in billions, not millions.

73% of consumers are already using AI in their shopping journey. 44% prefer AI search over traditional search. AI recommendations convert at 4.4x the rate of traditional search. And the global agentic commerce opportunity is projected at $3–5 trillion by 2030.

The brands that build visibility tracking for shopping agents today will own the discovery layer of tomorrow's commerce. The brands that wait will wonder where their customers went.

Vercel showed us the architecture. The question is: who builds it for commerce?