What is the Agentic Commerce Protocol (ACP) and where does it stand?

The Agentic Commerce Protocol (ACP) is an open-source standard governed by OpenAI and Stripe. With the ACP, AI agents can assist shoppers with making purchases within conversations.

ACP interactions are a part of agentic commerce, in which people can discover products, make decisions, and complete purchases without leaving the conversation they’re having with an AI agent.

Like the Universal Commerce Protocol (UCP), another open standard protocol for agentic commerce co-developed by Shopify and Google, the ACP is the underlying protocol that determines how an AI agent, merchant, and payment system work together to create the agentic shopping experience.

Without protocols like the ACP, every AI agent would need a custom integration with each merchant and payment provider to complete a purchase. The ACP replaces those integrations with a shared language for validating payment credentials, confirming orders, and approving or rejecting purchases.

What is the current status of the ACP?

The ACP powered Instant Checkout in ChatGPT, which launched in September 2025 and allowed shoppers to complete purchases without ever leaving the chat.

But OpenAI announced in March 2026 that Instant Checkout would move purchases to third-party apps. Shoppers can still discover products inside ChatGPT, but they’ll pay for purchases through a retailer’s own app or website.

That means the ACP still runs the infrastructure connecting ChatGPT to merchants, but it no longer handles the full transaction.

How does the ACP work?

When a shopper describes what they’re looking for, ChatGPT ranks results based on previous queries, availability, price, and whether a merchant is the primary seller. The ACP pulls product results from merchants whose feeds meet ACP requirements.

During the pilot, a buy button appeared directly in ChatGPT if merchants had enabled instant check-out. The shopper confirmed delivery details, selected a payment method, and completed the order without leaving the interface.

The transaction ran through the merchant's commerce platform and payment infrastructure via an API. Stripe handled payment processing by default, though merchants could use their own ACP-compatible payment provider. The merchant accepted or declined orders, and they handled fulfilment and support as they normally would.

The appeal for shoppers and merchants was immediacy. Shoppers could buy products within ChatGPT without tapping out to a storefront to complete the purchase.

Now, shoppers need to complete the purchase via the merchant’s website or app.

What agentic commerce means for B2C marketers right now

More consumers are using AI agents to discover, research, and buy products. Klaviyo’s 2026 AI Consumer Trends Report found that 39% of consumers have purchased a product based on an AI recommendation in the past 6 months. Another 27% have learned about a new product from an AI recommendation and then gone on to research it further.

The report also found that 30% of AI searches contain 8+ words, and 52% contain 3–7 words, with multiple descriptors. 78% include emotional or personal context at least some of the time. These personal, long-tail queries matter because AI agents use them as signals for personal preferences, context, and intent. Every answer to every future query is informed by those personal signals.

For example, someone may have previously asked an AI agent for advice on dealing with shin splints. When they later search for running shoes, they may get recommendations for surface stability shoes over lightweight racers, because the AI agent knows those may be better for shin splints.

Long-tail, contextual search queries call for more detailed, contextual product descriptions and reviews. That means more detail that speaks to specific use cases, personal context, and product positioning compared to others on the market.

In other words, agentic commerce is more than a channel shift. It’s the collapse of discovery, research, and sometimes purchase phases to one channel. Before AI, shoppers discovered products on one or more channels and moved on to a brand’s owned channels for comparison and purchase. Now, agentic commerce is consolidating more of this online journey to one place.

Owned channels still matter, as does traditional SEO. But agentic commerce is influencing how brands need to present their product data, so they’re showing up on AI channels to compete in these new spaces.

How to prepare your brand for agentic commerce

If you have a solid SEO foundation already, you’re well set up to extend discoverability to answer engine optimisation (AEO). Consider the following steps as complementary to your SEO efforts, so you’re prepared to serve up the context and detail required for an AI search query:

1. Audit and enrich your product data and metadata

LLMs use bots to crawl and index your pages. If data is missing or inaccessible, it won't be picked up.

Review your product data for detail and accuracy. Vague or incomplete descriptions, missing attributes, and outdated inventory reduce how often AI agents recommend your products.

To prepare your catalogue for AI recommendations, start with the content itself:

• Standardise attributes such as materials, colours, units of measurement, etc., so AI agents can interpret them consistently.
• Write product titles and descriptions in clear, natural language. Strive to answer questions with context and situational awareness. Instead of ‘whey protein powder’, think, ‘whey protein powder with 25g of protein per serving and no artificial sweeteners, ideal for post-workout recovery’.

Then, make sure LLM crawlers can read your content. Here’s what to remember:

• LLMs use crawlers like GPTBot and PerplexityBot that follow your robots.txt rules by default. Make sure key product pages aren't accidentally blocked. A complete XML sitemap helps, too, giving crawlers a full map of your site so they don't miss newer or more obscure product pages.
• Add schema.org markup to your product pages as JSON-LD, so AI crawlers can read key product information, such as names, descriptions, images, prices, and availability.
• Audit the presence of Global Trade Item Numbers (GTINs) and Stock Keeping Units (SKUs). AI agents use these to verify that products exist and are accurately represented. Missing identifiers can cause products to drop out of recommendations entirely.

2. Track and measure LLM attribution

Start identifying site traffic and conversions from LLMs. For example, when visitors arrive via ChatGPT, track them with a UTM parameter (e.g., utm_source=chatgpt.com). This is your first signal that agentic commerce could become a meaningful revenue channel for your brand.

You can even use a conversational AI assistant, like Klaviyo’s app for ChatGPT or Claude connector, to analyse LLM-driven traffic and conversions with simple prompts.

3. Turn agentic commerce signals into marketing content

Use information from AEO queries to guide new product page content and omnichannel campaigns. If you don’t have the resources to beef up all your product descriptions at once, let attribution data be your guide instead.

For example, if you’re noticing more AI-attributed traffic to your summer dress product category, that could be your signal to add more context to descriptions, add more product shots, and launch a new campaign that addresses contextual signals.

Klaviyo Composer, Klaviyo’s built-in AI marketing agent, can:

• Build cross-channel campaigns. Based on a prompt, your product catalogue, and complete customer context from Klaviyo B2C CRM, the AI marketing agent builds ready-to-send campaigns and flows with audience suggestions, personalised messaging, and timing recommendations.
• Honour brand guardrails. The AI marketing agent generates within your specific voice and rules from the first output. You maintain full strategic control, and nothing launches without your sign-off.
• Refine your vision. In a multi-turn conversation, the AI marketing agent helps you get it just right, with natural conversation and file uploads. Composer also learns as it goes, so every decision is smarter than the last.