Agentic Commerce in 2026: How to Prepare Your Store for AI Shopping Agents

For twenty-five years, e-commerce has assumed a human on the other end of the screen — someone who searches, clicks, compares tabs and taps "buy." In 2026 that assumption is quietly breaking. A new layer of software, the AI shopping agent, is starting to do the browsing, the comparing and, increasingly, the buying on a person's behalf. This is agentic commerce, and it is reshaping who — or what — you are actually selling to.

The shift is no longer theoretical. OpenAI and Stripe are maintaining an open Agentic Commerce Protocol (ACP), with its latest stable spec dated 2026-04-17. Google has rallied 60-plus partners behind a rival Agent Payments Protocol (AP2). Perplexity's Comet browser can compare products and complete basic transactions, and Amazon has already sued Perplexity over an agent buying on its site. When the buyer is an algorithm, the old playbook of persuasion, urgency banners and A/B-tested checkout flows no longer applies. This guide explains what agentic commerce is, how the plumbing works, and the concrete steps to make your store ready for it.

What is agentic commerce?

Agentic commerce is any transaction where an AI agent — not the shopper directly — performs the discovery, evaluation and purchase steps. Instead of a person visiting three sites and weighing the options, they tell an assistant "find me a quiet mechanical keyboard under $120 that ships to me by Friday," and the agent does the legwork: it queries catalogues, filters by the stated constraints, checks stock and delivery, and either presents a shortlist or completes the order outright.

It helps to separate two things that often get blurred together:

• Agentic discovery — the agent researches and recommends, but the human still clicks through and pays. This is the buy-side mirror of optimizing for AI search, and most brands are already exposed to it whether they realise it or not.
• Agentic checkout — the agent completes the transaction itself, inside the assistant, using a delegated payment credential. The shopper never lands on your product page or your checkout. This is the harder, newer frontier, and it is where the protocol wars are being fought.

The distinction matters because your preparation differs. Discovery rewards clean data and a trustworthy brand. Checkout rewards machine-readable pricing, stable APIs and a decision about whether you want an agent transacting on your behalf at all.

Why this is happening now

Three forces converged in the last year. Assistants like ChatGPT and Perplexity crossed from answering questions into acting on them. Agentic browsers — Comet on desktop, Android and iOS — put a purchasing agent one click from any web page. And the payment rails caught up: tokenized, scoped credentials now let an agent pay a merchant without ever handling a raw card number.

The commercial logic is stark. If an agent makes the choice, the discipline that used to win — ranking, ad creative, landing-page conversion — gives way to a new one: being the option the agent selects. Discovery is shifting from a human scanning a page of results to a machine filtering against explicit constraints, which is the same structural change we traced in how the AI buying journey differs from classic search.

The protocol stack: ACP vs AP2

Agentic checkout needs two questions answered: how does money move, and who authorized this purchase. Two open standards have emerged to handle them, and — importantly — they are designed to stack, not to replace each other.

Protocol	Backers	What it does	Question it answers
ACP (Agentic Commerce Protocol)	OpenAI, Stripe, Meta	Agentic checkout, cart & product feed, delegated payment and authentication	How the transaction and payment happen
AP2 (Agent Payments Protocol)	Google + 60 partners	Cryptographically signed "mandates" proving a user authorized an agent	Who authorized the purchase

Under ACP, the agent collects the buyer's payment selection, hands the merchant a narrowly scoped token, and the merchant charges that token through any compliant payment provider — while remaining the merchant of record. The whole transaction can complete inside ChatGPT or another compatible agent; the consumer never leaves the chat. Stripe shipped its Agentic Commerce Suite in December 2025, and PayPal joined as a payment provider in October 2025.

AP2 works one layer up, on trust. It produces a cryptographic chain of proof — a signed mandate — so a merchant, bank or card network can verify that an agent acted with genuine user consent rather than going rogue. In a world where software can spend your money, that authorization layer is not optional; it is what makes agentic checkout auditable and disputable. Expect a compliant store to eventually speak both: ACP for the mechanics, AP2 for the consent.

The friction: not everyone wants agents on their site

Agentic commerce is not a frictionless utopia, and it would be dishonest to pretend otherwise. Late in 2025, Amazon sued Perplexity and won a preliminary injunction blocking its Comet agent from accessing password-protected parts of Amazon's site to buy on users' behalf. The dispute captures a real tension: a purchasing agent can disintermediate a marketplace, strip away its merchandising surface, and turn a curated storefront into a raw price-and-availability API.

For you, the merchant, that means a strategic decision, not just a technical one. Do you want to be a destination shoppers experience, or an option an agent quietly picks and pays? For commodity products where price and delivery decide the sale, agent-readiness is pure upside. For brands that sell on story, fit or trust, letting an agent flatten your storefront into a data feed carries a cost. Neither answer is wrong — but you should choose deliberately rather than have the choice made for you.

How to prepare your store: a practical checklist

Whether you lean into full agentic checkout or just want to be discoverable to agents, the groundwork overlaps. Work through these in roughly this order — the early items are quick, the later ones compound.

1. Let the right agents crawl you. In robots.txt, allow the discovery and shopping bots you want — OAI-SearchBot and ChatGPT-User for ChatGPT, PerplexityBot for Perplexity, Google-Extended for Google's AI surfaces. Blocking them, even by accident, removes you from consideration entirely.
2. Make your product data machine-first. Rich Product schema with brand, GTIN, aggregateRating and a full offers block — currency, price and priceValidUntil — is what lets an agent qualify and compare your item. A 2026 analysis found only about 9% of Shopify product pages carried the structured data agents need; the bar to stand out is still low.
3. Publish a clean, current feed. Get your catalogue into the merchant feeds assistants read (Bing Merchant Center and Google Shopping). Agents lean on feeds far more than on scraping individual pages, and a stale feed is worse than none.
4. Expose accurate stock and delivery. Agents filter hard on constraints like "in stock" and "arrives by Friday." Inaccurate availability or shipping data gets you silently filtered out — and, once checkout is automated, generates costly cancellations.
5. Keep prices unambiguous. One canonical price per item, in structured markup, with any conditions (subscription, region, tax) explicit. Agents cannot infer a hidden discount the way a human reads a banner.
6. Decide your checkout posture. Evaluate ACP-based Instant Checkout and the Perplexity path against your model. Weigh the per-transaction fee and loss of storefront control against reaching buyers at the exact moment of decision. This is a business call, not just an integration.
7. Strengthen verifiable trust signals. Fresh reviews, visible ratings and editorial mentions are the tiebreaker between comparable products — for an agent just as for a human. Build the experience, expertise, authority and trust signals that AI systems already use to decide who to cite.
8. Instrument for an invisible channel. Because the recommendation — and sometimes the sale — happens inside the assistant, classic analytics under-report it. Prompt-test the buying questions you want to win, watch for referral and checkout activity tied to agentic sources, and track your share of recommendation versus named competitors.

What to expect, and when

The technical work moves fast: opening robots.txt, fixing a feed and enriching schema can improve agent eligibility within days. Adopting a checkout protocol is a larger project measured in weeks, gated more by internal decisions and payment-provider onboarding than by the code itself — the integration work of wiring ACP tokens through your payment provider and exposing a clean, machine-readable product API is the kind of build the YuSMP Group team handles. And the softer, compounding work — trust signals, buying guides, keeping data accurate — is a 60-to-90-day effort that pays off as steadier agent recommendations.

Set expectations honestly. Agentic commerce today is early: transaction volume is small, the standards are still settling, and the legal boundaries are being drawn in court. But the direction is not in doubt. The merchants who make their data machine-first and decide their checkout posture now will be ready when the volume arrives — and far better positioned than those who wait for agents to define the terms for them.

The takeaway

Agentic commerce reframes the oldest question in retail. It is no longer only "how do I convince a person to buy?" but also "how do I make it effortless for a machine acting on their behalf to choose me?" The answer rewards the same fundamentals that win in AI search — clean data, unambiguous structure, verifiable trust — applied to your catalogue and your checkout. Get discoverable to agents first, decide deliberately whether you want them transacting for you, and instrument a channel you cannot fully see. Do that while most stores are still built purely for human eyes, and you claim ground that will be far more contested a year from now. For the recommendation side of this shift, see our guide to getting your products into AI shopping.