Questions, answered. Before you have to ask.

A shopper asks an AI assistant for a product instead of searching for it. The assistant reads structured product data, compares options against what the shopper asked for, and then either completes the purchase in the conversation or sends the shopper to a store to buy. Agentic commerce work makes sure your products can be found, described correctly and bought in that flow.

ACP, the Agentic Commerce Protocol from OpenAI and Stripe, carries your product feed into ChatGPT, which uses it to decide what to show and then sends the buyer to your checkout. UCP, Google's Universal Commerce Protocol, powers checkout inside AI Mode, Gemini and Universal Cart. They reach different assistants, so many merchants end up needing both. Protocol Layer wires one or both.

No. Search still sends you traffic and still needs its own work. Assistants lean on different signals: structured attributes, feed quality, and policy terms they can read and repeat. There is overlap, since clean product data helps both, but ranking well in search does not mean an assistant can verify your products.

Usually because it cannot confirm that a product meets the shopper's request. If the size, material, compatibility or return terms live in a paragraph, an image or a PDF instead of a field, the assistant has nothing to check against, so it recommends a store whose data it can read. The free assessment shows which of your products this happens to and who gets named instead.

Some are, and the rails are live: Google completes purchases inside AI Mode and Gemini through UCP, Perplexity and Copilot check out in-chat, and ChatGPT routes ACP-fed discovery to your site. AI-referred traffic to US retail was up 393% year over year in Q1 2026 (Adobe Analytics) and converts better than other channels. Whether your category justifies connecting this quarter or watching another one is exactly what the free assessment answers.

It retired Instant Checkout in March 2026, yes. What replaced it matters more for most merchants: ChatGPT now uses ACP feeds to decide what to show, and sends the buyer to your own checkout. That means your feed quality decides whether you're recommended, and your site decides whether the sale closes. Both are things you control.

Yes. Some categories genuinely should wait another quarter or two, and saying so costs us an engagement each time. If the assessment says wait, you keep the scored note and the transcripts, and we don't sell you a build.

ACP (OpenAI and Stripe's Agentic Commerce Protocol, which carries your product feed into ChatGPT) and UCP (Google's Universal Commerce Protocol, which powers checkout in AI Mode, Gemini and Universal Cart), plus the surfaces that adopt them, including Perplexity's Merchant Program and Copilot Checkout. Protocol Layer wires one protocol or both, chosen on the scoping call, with or without a storefront agent.

It depends on where your shoppers already are. If most of your demand still arrives on your own site, the storefront agent returns first and leaves the catalog structured for the protocols later. If assistants are already sending you traffic that fails to convert, the protocols are the more urgent gap. Doing both is a single engagement because the catalog work underneath is shared, and the assessment will tell you which applies to your store.

No. Storefront Agent and Protocol Layer both install natively on Shopify, BigCommerce, WooCommerce, Salesforce Commerce Cloud, Adobe Commerce and Commercetools. The work is an agent, connectors, catalog data and checkout handoff, not a migration. Custom stacks get scoped honestly in the assessment before anyone commits.

It answers product questions on your own site, filters by the constraints a shopper gives, compares variants, builds the cart and hands off to your existing checkout. It answers from your catalog only, so it does not invent a spec or a price. Questions your catalog cannot answer yet go into a shopper question log, which becomes a list of the data gaps worth closing next.

Yes. Headless and custom builds are on the list alongside Shopify, BigCommerce, WooCommerce, Salesforce Commerce Cloud, Adobe Commerce and Commercetools. They take more scoping because there is no standard app to install, so the assessment looks at how your catalog and checkout are exposed before anything is priced.

As an add-on. A returns and post-purchase flow can be added to Protocol Layer or to the combined build, so an order placed through an assistant can be tracked and returned the same way as any other order.

Considerably more than most merchants expect when they start. Attributes, taxonomy, product identity and policy terms all have to exist as structured fields before either build will work properly, and that preparation is usually the largest line in the engagement. It is included in every build and sized from your assessment, so the scale of it is visible before you commit.

Yes. It is a scored audit with transcripts you can read through yourself, not a qualification exercise dressed up as one. It costs nothing, and the findings are yours to keep and act on whether or not you ever build anything with us.

The form asks for your store URL, platform, rough revenue and catalog size. Within one business day we send the 10 buying questions we plan to ask, so you can swap any that do not fit your category. Over 5 business days we score your storefront, feed and checkout against ACP and UCP and ask 3 assistants those questions. You get a one-page findings note with the transcripts, followed by a thirty-minute call to review the findings and agree on the next steps.

The free assessment needs a two-minute form and one thirty-minute findings call. Storefront Agent or Protocol Layer needs a named owner for a couple of hours a week while we build, plus access to whoever holds your product data. We don't run projects that require you to staff a squad.

Each build is scoped per engagement. The main factors are catalog size, how much of your product data is still in prose, images and PDFs, and whether you take one protocol or both. Taking the storefront agent and the protocols together is priced as one scope, because the audit and catalog structuring happen once instead of twice.

Access to your commerce platform and to wherever your product data lives, such as a PIM, an ERP or supplier files. For Protocol Layer we also work in your payment provider account, because payments, identity and consent are set up on your accounts, not ours.

Every build is scoped and priced before work starts. If it turns out larger than we scoped it, absorbing that is our problem as the people who estimated it, not a change order for you to approve.

You do. The storefront agent, the connectors and the feeds all run on your platform and your accounts, documented in a runbook your team owns. You remain merchant of record on every protocol we connect. Nothing we build stops working if you stop paying us.

Two ways, both on your side of the fence. Agent-attributed orders land in your own analytics as real transactions, not estimates. And under Managed, the drift check tells you when a claim, a price or a checkout stops matching what you ship, so you see a problem before a shopper does.

Testing is mandatory for every protocol we connect. Each connection is tested against the live assistant before it ships, and you get a conformance report your risk and finance teams can read. Nothing goes live untested.

No. Managed is an optional retainer. It covers drift checks and protocol updates, storefront agent maintenance, traffic monitoring and a standing SLA. Without it, your team runs everything from the runbook we hand over.

They will change, since the specifications are young and versioned frequently, and the models behind a storefront agent move as well. Managed exists for exactly this: we track the drift, apply the updates and keep every live connection and the agent itself current. The runbook is written so your own team can do the same without us.