Personal Agent Protocol (PAP): Big Tech gives small businesses a way to sell to AI agents

What is the Personal Agent Protocol?

PAP is a shared rulebook for AI assistants that act for a person. It sets how an assistant like Meta's Muse finds what a business offers, signs in for the customer and gets the job done, whether that is a return, a booking or a purchase. Anyone can implement it.

Today most assistants use websites the way people do. They load pages, click buttons and fill in forms. When that fails they call the support line or open web chat. PAP gives them a direct, signed-in connection instead, so the same task takes seconds.

What PAP means in one minute

• Your next customer may be software. A person asks their assistant to buy, book or return, and the assistant does the visiting.
• It is a standard, not a product. No one owns it and any store, platform or assistant can use it, the way any site can offer "Log in with Google".
• It replaces guessing with asking. Instead of reading your pages like a person, the assistant asks your store what it offers and what it may do.
• It does not replace your website. The website is still where agents start, so a clear site matters more, not less.

PAP at a glance

The short version, before the detail.

The rest of this page explains each part in plain words, and what a store can do this month.

Why it is happening now

Meta launched Muse, its personal AI assistant, in September 2026. Within weeks it had millions of US users and sat above ChatGPT in the App Store. People use it to buy gifts, sign their kids up for classes and book appointments.

That created a problem for stores. A store cannot tell an assistant acting for a real customer from a random bot, so many block it. Amazon already blocks Meta's agents. As Bret Taylor put it when he announced PAP, it is kind of chaos until a standard exists.

What the numbers tell you

• Adoption came first, rules second. Muse reached millions of US users within weeks of its September 2026 launch, before any shared standard existed.
• Blocking is already happening. Amazon blocks Meta's agents. Stores that block every bot may also block real customers who use an assistant.
• Big retail moved fast. Walmart and Shopify signed on at launch, so a large share of US online stores will have a path to support PAP early.
• Support costs are the hidden driver. Assistants that fail on the website fall back to phone and chat, the most expensive channels a business runs.

What breaks today, and what PAP fixes

Take one task: a customer asks their AI assistant to return a pair of shoes and order the next size.

Where stores lose AI shoppers today

• CAPTCHAs and bot walls. They stop the assistant at the exact moment the customer wants to buy.
• Prices and stock that only show after clicks. If the assistant cannot read them, it compares you on incomplete data or skips you.
• Policies hidden in PDFs or pop-ups. Return windows and shipping costs decide many purchases, and agents need them in plain text.
• Account-only actions. Returns, order changes and bookings behind a login are where PAP helps most, because the agent can finally sign in for the customer.
• Layout changes. A redesigned checkout button can break every assistant that learned the old one. A protocol does not break when the page changes.

How PAP works

What changes for a store

• Guest tasks stay open. Stock checks, prices and policy questions work before sign-in, so make those answers public and accurate.
• Account tasks get a trusted path. Signed-in sessions run on OAuth, so the store knows which customer the agent acts for.
• One visit across channels. A question asked before sign-in and an order change after it count as one session, so context is not lost.
• You pick the route. Small stores can let agents use the website, larger ones can expose APIs, and support-heavy businesses can route agents to their own AI agent.

Who controls what

PAP gives both sides a say. The customer decides how far their assistant can go. The business decides what it allows.

Decisions a business should make before PAP ships

• Which tasks are open to agents. For example: answer questions and check orders, but not cancel subscriptions.
• Which tasks need a human. Refunds above a set amount, account deletion or anything regulated.
• Which door each task uses. Simple lookups through the website, orders through an API, complaints through your support agent.
• How you will spot agent traffic. Separate agent sessions in analytics so you can see what they ask and where they fail.

Who is in, and who is not

Decagon open-sourced a related consent protocol, PACT, and joined the PAP working group on launch day. OpenAI and Anthropic were not part of the launch. Taylor said he expects them to join and would be really disappointed if competitors did not use it.

What the line-up means

• Meta brings demand. Muse users are the shoppers who will send agents to stores first.
• Shopify and Stripe bring the stores and the payments. Merchants on those platforms are likely to get support without building it themselves.
• Sierra, Genesys and Decagon bring customer service. Expect returns, order changes and support tasks to be the first use cases, before full shopping.
• Missing players matter. Without OpenAI, Anthropic and Amazon, stores may need to support more than one standard for a while, such as OpenAI's Agentic Commerce Protocol as well as PAP.

What comes next

• The v0.1 spec is due later in October 2026, with design workshops and a reference implementation for developers.
• Detailed permissions could let customers and companies limit specific actions.
• Push notifications could let a company tell an assistant the moment an order ships or a flight is delayed.
• Payments extensions could let an assistant complete a purchase without seeing the customer's card. Stripe is a founding member.

How to get your store ready now

The spec is weeks away, but agents already read your store today. These steps help with PAP and with AI search at the same time.

For the store side in detail, see preparing your Shopify store for AI shopping agents and the Agentic Commerce Protocol explained.

Quick checklist for small businesses

• Every product page shows price, stock, sizes and shipping cost without a click.
• Product structured data (schema.org Product and Offer) is complete and matches the page.
• Returns, shipping and warranty each have their own plain-text page linked from the footer.
• Your bot protection lets through known assistants for guest tasks instead of blocking all automated traffic.
• Your top 20 customer questions are answered on the site in one or two sentences each.
• You have asked ChatGPT, Gemini, Perplexity and Muse what they know about your store this month.
• Someone owns "AI shoppers" on your team, even if it is one hour a week.