Skip to content
Vouch

All posts

Agentic commerce: what happens when AI agents do the buying

By Vouch7 min read

Vouch blog cover with the title "Agentic Commerce: When AI Does the Buying" beside a line drawing of a shopping trolley with two wheels and a small red dot at its top right.

Agentic commerce is what happens when an AI assistant stops only recommending products and starts completing the purchase. The buyer states what they want, the assistant researches, compares and, within limits the buyer sets, checks out on their behalf.

This is no longer a slide in a keynote. OpenAI, Perplexity and Google have each shipped pieces of it, with dates and documentation. This guide lays out what they have actually launched, what decides whether an agent picks your product, and what a brand can do now. The platform facts carry links and dates; the advice is labelled as editorial recommendation.

What the platforms have shipped

OpenAI: Instant Checkout and the Agentic Commerce Protocol

On September 29, 2025, OpenAI launched Instant Checkout in ChatGPT. US users can buy from US Etsy sellers inside the chat, with over a million Shopify merchants announced as coming soon at launch. It supports single-item purchases first, with multi-item carts and more regions to follow, according to the announcement.

Three details from OpenAI's own post matter for brands:

  • Product results are organic and unsponsored, ranked on relevance to the user.
  • When several merchants sell the same product, ChatGPT says it weighs availability, price, quality, whether the merchant is the primary seller, and whether Instant Checkout is enabled.
  • Merchants pay a fee on completed purchases, remain the merchant of record, and keep fulfilment, returns and support in their existing systems.

Underneath it sits the Agentic Commerce Protocol (ACP), an open standard co-developed with Stripe and open-sourced the same day. Stripe's announcement describes the payment side: a Shared Payment Token scoped to a specific merchant and cart total, so the buyer's card details are not exposed to the assistant.

Perplexity: Buy with Pro

Perplexity moved earlier. In its shopping launch (November 18, 2024), it introduced Buy with Pro: US Pro subscribers can buy select products from select merchants without leaving Perplexity, with free shipping on those orders, and a redirect to the merchant's site when Buy with Pro is unavailable. Its product cards are described as unsponsored recommendations.

The same post launched the Perplexity Merchant Program, free for merchants, which lets retailers share product specifications directly. Perplexity's stated benefit: being in its index with robust details increases a product's chances of being recommended, because it can better judge quality and relevance.

Google: the Agent Payments Protocol

Google's piece is infrastructure rather than a storefront. In September 2025, it announced the Agent Payments Protocol (AP2), an open protocol developed with more than 60 payments and technology organisations, including Mastercard, PayPal, American Express and Coinbase. AP2 uses cryptographically signed "mandates" to prove what a buyer authorised: an Intent Mandate records what the buyer asked for, and a Cart Mandate records the exact items and price they approved. It is payment-agnostic and covers both purchases where the buyer is present and delegated tasks such as "buy concert tickets the moment they go on sale".

Can an agent buy without asking?

Both live designs keep a human in charge, though at different depths. OpenAI says ChatGPT users explicitly confirm each step before any action is taken. Google's AP2 goes further into delegation: a buyer can sign an Intent Mandate with rules such as price limits and timing, which then lets the agent complete the purchase when those exact conditions are met, with the Cart Mandate as a tamper-proof record of what was approved.

For a brand, the design detail worth noticing is that the approval record fixes the exact items and price. A quote your site gives an agent is not a loose estimate in these flows; it is the number the buyer's signature locks in. Accurate pricing data is not only a ranking input, it is what the buyer ends up paying.

What this changes for brands

In classic ecommerce, your site had to persuade a person who was already on it. In agentic commerce, an intermediary reads the web on the buyer's behalf, forms a shortlist and can act on it. The practical consequences:

  • Being recommended matters more, because the agent may complete the purchase inside the conversation. There is no later visit where your retargeting or merchandising gets a second chance.
  • Product data becomes a sales surface. Both OpenAI's ranking factors and Perplexity's merchant programme point the same way: agents favour products whose price, availability and specifications they can read clearly.
  • Trust signals travel with the product. Reviews, returns terms and primary-seller status are the kind of facts an agent weighs when a buyer asks "can I trust this one?"

These inferences are editorial, drawn from the platform documentation above, not from published conversion data.

What to do now

The recommendations below are suggestions to test, grounded in the documented platform behaviour rather than measured outcomes.

Get your product facts machine-readable and current

Price, stock, sizes, delivery times, warranty and returns should be in your pages' HTML text, in your feeds and consistent between the two. An agent that finds a stale price has a reason to pick the rival whose data is clean.

Decide which agent channels you will support

If you sell physical products in the US, review Instant Checkout (via the Agentic Commerce Protocol) and the Perplexity Merchant Program, and weigh the fees and data-sharing terms against a new sales channel. Both are documented at the links above.

Keep the content fundamentals in place

Agents read the same web your other AI visibility work targets. The access, content and measurement basics in our guide to AI search visibility, the product-selection dynamics in how ChatGPT Shopping chooses products and the ranking factors in how to rank in ChatGPT search all still apply. Agentic checkout sits on top of them; it does not replace them.

Watch agent traffic separately

Purchases and visits that arrive via assistants are visible in your analytics. Our guide to seeing AI referral traffic in GA4 covers the setup. If agentic checkout grows for your category, that channel is where the early signal will show.

What we do not know yet

Honesty about the gaps matters here. As of November 2026, none of the launch posts above publishes conversion or volume data for agent-completed purchases, so any claim about how much revenue agents drive is guesswork. The two live buying experiences above are US-only at launch. And the protocols themselves are young: ACP and AP2 both launched in September 2025, and their terms can change. Treat agentic commerce as a channel to prepare for and measure, not a proven revenue line.

How Vouch helps

Agentic commerce raises the stakes on a question Vouch already measures: when a buyer's conversation with an assistant reaches "which one should I buy", is your brand the one it names? Vouch runs Threads, six-question modelled buyer conversations from discovery to decision, on 4 of the 7 assistants (ChatGPT, Claude, Gemini and Perplexity) three times a week, and records where your brand drops out and which rival replaces it. Single prompts run daily across all seven tracked engines. The weekly Vouch Rate (0 to 100) combines stage survival with the final recommendation.

Vouch is invite-only through a waitlist, every invitation starts with a 14-day free trial, and plans start at $39 a month (as of November 2026). Join the Vouch waitlist, or check whether AI crawlers can reach your product pages with the free AI crawler checker.