AI agents are starting to shop on people’s behalf — ChatGPT can complete a purchase inside a conversation, Perplexity surfaces and compares products directly in its answers, and Google’s AI Mode increasingly skips the results page entirely and recommends. If your product data isn’t structured for a machine to read, compare, and act on, you can rank well in classic search and still be invisible in the exact moment someone is ready to buy. This is agentic commerce, and optimizing for it is now a distinct discipline from both SEO and traditional GEO.
We’ve been tracking how our clients’ product and service pages perform when an AI assistant is doing the comparing instead of a human scrolling a results page. The pattern is consistent enough to write down: pages that read cleanly to an agent — clear pricing, explicit availability, structured attributes, no marketing fluff blocking the facts — get pulled into agent answers far more often than pages that require a human to piece the details together visually.
What is agentic commerce, exactly?
Agentic commerce is any transaction where an AI system — not a human clicking through a results page — does some or all of the searching, comparing, and purchasing on a person’s behalf. In 2026 this shows up in a few concrete forms: conversational shopping inside ChatGPT and similar assistants, comparison answers inside Perplexity and AI Overviews that name specific products and prices, and browser-based agents that can navigate to a site and complete a checkout with minimal human input. The common thread is that a piece of software is reading your page, not a person — and software reads differently than people do.
Why does this matter more than regular GEO?
Regular GEO is about getting cited or summarized in an AI answer — showing up when someone asks a question. Agentic commerce optimization is about surviving the next step: being selected, compared favorably, and transacted with, without a human ever loading your page in a browser. That means the things that matter shift from “is this content authoritative and well-structured” to “can a machine extract the price, the availability, the specifications, and the terms with zero ambiguity.” A blog post can win at GEO with a strong narrative. A product page has to win at agentic commerce with clean, unambiguous data.
The five things an AI shopping agent actually checks
1. Structured, machine-readable pricing
Price needs to exist as actual structured data (schema.org Product/Offer markup), not just as text inside a styled div that a script generates after the page loads. If your price only renders client-side via JavaScript after several seconds, some agents will never see it at all.
2. Explicit availability and fulfillment details
“In stock,” lead time, shipping regions, and return policy all need to be stated plainly and, ideally, marked up in schema. Agents comparing your product against a competitor’s will often default to whichever page states these facts with more certainty — vague copy loses by default.
3. Attributes broken into fields, not paragraphs
A sentence like “comes in three sizes and two colors, perfect for any home” is far harder for a system to parse and compare than a table or list: Size: S/M/L. Color: Black/Sand. Specifications belong in structured lists or tables, not woven into persuasive copy.
4. A page that loads and renders fast without heavy client-side dependencies
Many AI crawlers and agents have limited patience for JavaScript-heavy rendering. If your core product facts only appear after multiple render cycles, treat that as a reliability problem, not just a performance one.
5. Consistent facts across every place your product appears
Agents cross-reference. If your own site says €49 and your Google Shopping feed says €45 because a sale wasn’t synced, that inconsistency can knock you out of consideration entirely rather than just costing you a click.
A simple audit you can run this week
| Check | How to test it | Fix if it fails |
|---|---|---|
| Price visible without JavaScript | View page source (Ctrl+U) and search for your price — is it in the raw HTML? | Server-render price and Product schema instead of loading it client-side |
| Product schema present | Run the page through Google’s Rich Results Test | Add Product, Offer, and AggregateRating schema where accurate |
| Availability stated in plain text | Read the page as if you can’t infer anything — is “in stock” written anywhere? | Add explicit availability text near the price, not just an icon |
| Specs in a table or list | Scan the page for a structured spec block | Extract key attributes out of paragraph copy into a labeled list |
| Price consistency across channels | Compare your site, Google Shopping feed, and any marketplace listing | Sync pricing and stock from a single source of truth |
Does this replace SEO or GEO work?
No — it sits alongside it. Your blog content and topical authority still earn you the citations and trust signals that get you considered in the first place. Agentic commerce optimization is what happens once you’re in the running and a system is deciding whether to recommend you over the next result. Think of it as the difference between getting invited to the shortlist and actually winning the comparison once you’re on it. Skipping either half leaves money on the table: all the authority in the world doesn’t help if your price is buried in a JavaScript widget an agent can’t read.
What we’d prioritize first if you sell online
- Audit your five highest-revenue product or service pages for the checks above before touching anything else — fix the pages that matter most first.
- Make sure your Product/Offer schema is accurate and updates automatically with price and stock changes, not manually.
- Pull your most important specifications out of marketing paragraphs and into a plain, labeled list near the top of the page.
- Check that your price and availability render in the raw HTML, not only after JavaScript executes.
- Reconcile pricing across your website, ad feeds, and any marketplace listings so an agent never finds two different answers.
How do you even know if it’s working?
This is the part most small businesses get stuck on: when an AI agent recommends your product and someone buys, that visit often doesn’t show up as a normal referral in your analytics. The click may arrive as direct traffic, or not register a session at all if the transaction happens inside the assistant itself. Don’t take a flat line in your “AI referral” report as proof nothing is happening — it’s frequently a measurement gap, not a performance gap. Watch for softer signals instead: branded search volume ticking up without a matching ad spend increase, direct traffic growing on product pages you haven’t otherwise promoted, or customers mentioning in support conversations that “ChatGPT recommended you.” None of these are definitive on their own, but together they’re a more honest picture than an analytics dashboard that was never built to track this kind of visit.
The honest caveat
Agentic commerce is still early, and the assistants doing this comparison work today will change how they crawl, cite, and transact over the next year. Nobody — including the platforms themselves — has this fully settled. But the underlying principle won’t age out: machines reward pages that state facts plainly and unambiguously, and that’s also simply good practice for accessibility and conversion. Fixing this now is a low-regret bet even if the specific platforms shift.
This builds directly on the fundamentals we covered in The GEO Audit and in how to optimise product pages for AI shopping answers — if you haven’t run either of those checks yet, start there before layering on agentic-commerce specifics. And since measurement is the part everyone struggles with, it’s worth pairing this with how to track AI-driven traffic in GA4 when referral data is missing.