How to Optimize for Agentic Search on E-commerce Sites

Agentic web traffic grew 1,300% in the first eight months of 2025. Google’s research found AI agents take an average of 4.9 steps per query, searching, comparing, and evaluating before delivering a result. OpenAI has shifted from keeping transactions inside ChatGPT to redirecting users to merchant sites. The shift from AI answering questions to AI completing tasks is already underway. Most content still isn’t built for it.

TL;DR

• Agentic traffic grew 1,300% in eight months. Agents take 4.9 steps per query on average.
• Agents don’t just cite content. They act on it. Comparing products, checking prices, booking reservations.
• Most agent protocols are developer infrastructure. UCP is the one that matters for content teams.
• The basics matter more than the protocols. Structured data, clear product information, semantic HTML.
• OpenAI now redirects users to merchant sites. The fear that agents would keep all traffic on-platform hasn’t materialized.

What is agentic search?

AI that doesn’t just answer questions. It completes tasks.

Traditional AI search retrieves information and summarizes it. Agentic search goes further. It compares products across multiple sources, cross-references claims, checks availability, and makes recommendations backed by evidence. Google’s SAGE research found agents take an average of 4.9 steps per query: searching, opening pages, evaluating information, and refining their approach before delivering an answer.

Semrush documented four tiers of increasing complexity:

Tier	Example	What the agent does
Simple query	“How much does a Tesla Model 3 cost?”	Fetches a price and returns it
Comparison	“Compare the Model 3, Ioniq 6, and i4 on range and price”	Pulls specs from multiple sources, builds a comparison
Research brief	“Best electric sedan under $50K with 300+ mile range and tax credit eligible”	Searches across manufacturers, tax databases, and review sites, synthesizes a ranked recommendation
Delegated action	“Book a test drive for the top two at the nearest dealer Saturday”	Navigates booking forms and completes the transaction

Most content today is optimized for tier 1, but tier 4 is where the traffic and revenue are heading.

Will AI agents send traffic to my site or keep users on their platform?

They send traffic. OpenAI shifted in March 2026 from native checkout within ChatGPT to redirecting users to merchant sites for purchases. The fear that agents would absorb all transactions on-platform didn’t play out.

Google’s agent-friendly websites guide describes browser agents that navigate sites on behalf of users, analyzing visual renderings, raw HTML, and the accessibility tree. These aren’t abstract future agents. They’re crawling the web today.

What this means for your site: when an agent lands on your product page, it needs to find structured, machine-readable information immediately. If your pricing is behind a “See Price in Cart” button, the agent can’t report it. If your product specs are in images without alt text, they don’t exist. If your availability requires selecting a location first, the agent sees “out of stock” and moves on.

“Browser agents may access your website to gather the data they need to complete tasks, such as analyzing visual renderings, inspecting the DOM structure, and interpreting the accessibility tree.”

The sites that perform best for agentic traffic have product information, pricing, availability, and specifications present in the initial HTML as structured data, rather than hidden behind clicks, loaded through JavaScript, or buried in images.

Does my e-commerce site need UCP?

Eventually, yes. But structured data matters more today.

Universal Commerce Protocol (UCP) is an open standard backed by Google that lets agents discover and transact on e-commerce sites without custom integrations. It’s the protocol most likely to become the standard for agentic commerce. Google’s own documentation points to it as the path forward.

What UCP does: it gives agents a machine-readable surface for your product catalog, pricing, availability, shipping, and return policies. Instead of an agent scraping your HTML and guessing which text is the price, UCP tells it directly.

UCP is a longer-term play. Structured data on your pages works today and reaches far more systems. If you can only do one, do structured data first and add UCP when your catalog infrastructure supports it. Google Merchant Center feeds are an intermediate step that works now and aligns with the UCP direction.

How do I structure my product pages for AI agents?

The same GEO fundamentals help agents, plus a few additions.

What carries over:

• Structured data: Product, Offer, AggregateRating, Organization schemas
• Clear, descriptive headings: not “Features,” but “Battery Life and Charging Speed”
• Answer-first paragraphs under each heading
• Server-side rendering or static generation for critical content

What agents need beyond standard GEO:

• Accessible interactive elements. Google’s agent guide emphasizes semantic HTML: <button> for buttons, <a> for links, <label> for form fields. Agents navigate the accessibility tree. A <div onclick> button is invisible to them.
• Stable element IDs. Agents remember which button did what. If your “Add to Cart” button has a different ID every page load, the agent can’t reliably interact with it.
• No information behind clicks. Agents don’t click to reveal content the way humans do. If your pricing, specs, or availability require expanding an accordion or selecting a variant, consider whether that information could live in the initial HTML.
• Real alt text on product images. Agents can’t see images. They read alt text. alt="product image" tells the agent nothing. alt="Sony WH-1000XM6 wireless noise-canceling headphones in silver" tells the agent exactly what it’s looking at.

What you don’t need to do:

• Rewrite your product descriptions for agents. Good human descriptions work for agents too. They’re looking for facts, not SEO copy.
• Implement every protocol. UCP is the one to watch. The rest are developer infrastructure.
• Build separate “agent pages.” Agents read your existing pages. Make those pages agent-readable.

What’s different about optimizing for agents vs. AI search?

AI search is about getting cited. AI agents are about being used to make a decision.

Slate HQ’s analysis of 300,000+ citations showed that AI engines draw from fundamentally different source pools depending on the platform. Agentic search adds another layer: the information has to survive extraction and be usable for decision-making.

For AI search (citation)	For AI agents (action)
Clear headings and answer-first paragraphs	Structured data with explicit values
Unique perspective and original data	Product specs in consistent, comparable formats
Extractable passages	Machine-readable pricing, availability, shipping
Authoritative positioning	Verified claims the agent can cross-reference

Most GEO fundamentals serve both purposes. The gap is structured data and making sure the information an agent needs to act on isn’t hiding behind JavaScript or interactive elements.

What should I do now vs. wait on?

Do now:

1. Add structured data to your product pages. Product schema, pricing, availability, ratings. This serves AI search and AI agents simultaneously.
2. Check if your core information survives without JavaScript. Open your product pages in a text-mode browser. Can you see the price, specs, and availability? If not, agents can’t either.
3. Use semantic HTML for interactive elements. <button>, <a>, <label>. This is the foundation agents navigate on.
4. Add descriptive alt text to product images. This is how agents “see” your products.

Do soon:

5. Submit your product feed to Google Merchant Center. This is the bridge between today’s structured data and tomorrow’s UCP.
6. Audit your product information for comparability. If an agent pulls specs from your site and a competitor’s, can it build a meaningful comparison? Consistent units, explicit values, and comparable formats matter.

Watch and assess:

7. UCP adoption. When major e-commerce platforms build UCP support, the implementation path will be clearer. For now, structured data and Merchant Center cover the same ground.
8. Agent analytics. Track referral traffic from AI agents. GA4 doesn’t identify agent traffic cleanly yet, but dedicated tools are emerging. When measurement catches up, you’ll want a baseline.