home/blog/agentic-commerce-readiness-levels
agentic commerce readiness · agentic commerce

Agentic Commerce Readiness: The 0–4 Levels Explained

Lire en français

Agentic Commerce Readiness: The 0–4 Levels Explained

By Cosolo, Founder of LoudPixel. Authored: 2026-09-13. The levels described here are the ones the free report grades on any domain; what they mean is public, how they are measured is not.

TL;DR: Agentic commerce readiness is whether an AI agent — shopping on a person's behalf — could find what you sell, read the price and the terms, and complete a purchase from your site without a human clicking through. It is a different question from AI visibility, which only asks whether the AI names you. LoudPixel grades it as a level from 0 to 4: invisible, listed, discoverable, buyable, operated. Level 3, buyable, means an agent can complete a purchase on your own rails. Most sites we check are Level 0 or 1, and most of them are named by an AI somewhere — which is the problem: being named and then unbuyable is a sale lost at the last step.

Key Facts:

  • Agentic commerce is an AI agent discovering products, comparing offers, and completing checkout on a shopper's behalf. It is distinct from AI-assisted shopping, where the human stays in the loop at every step.
  • Shopify now publishes developer documentation for agents that negotiate, discover products, complete checkout and monitor orders on its stores (shopify.dev/docs/agents). The protocols are real and dated; LoudPixel keeps a public record of what each one says and when we last read it.
  • Readiness is graded, not guessed: five levels, each defined by what an agent can actually do on the site, published for free on any domain.
  • The honest limit, stated on every report: buying inside ChatGPT itself is a separate, US-only programme. A Level 3 site is buyable by an agent on the site's own rails; we do not claim more than that for anyone.

What agentic commerce actually means

For twenty years a sale on the web had the same shape: a person searched, clicked, read, compared, added to cart, typed a card number. Every part of a website was designed for that person — the hero image, the reassurance copy, the "Buy now" button placed where a thumb lands.

An AI agent shopping on someone's behalf does none of that. It fetches your pages the way a program does: no images rendered, no JavaScript run, no button pressed. It reads what is declared — a product, a price, a currency, a policy, a way to pay — and if a thing is not declared in a form it can read, the thing does not exist. The person on the other end asked their assistant for "a standing desk under $600 that ships to Montréal this week" and will be shown three shops. Whether yours is one of them depends on whether the agent could read your price, your shipping and your returns, and whether it could get from your product page to a confirmed order.

Agentic commerce readiness is the name for how far along that path an agent can get on your site today.

How it differs from AI visibility — and why you need both

AI visibility answers: when a buyer asks an AI what to use, does the answer name you? Every tool in that market measures it, and it matters — a buyer who never hears your name never buys.

Agentic commerce readiness answers the next question: once the AI has named you, could its agent actually buy? It is the stage after visibility, and it fails silently. Nothing in your analytics shows an agent that read your product page, could not find a currency on the price, and moved on to the shop where it could. There is no bounce, no abandoned cart, no form. The deal was decided in a program's parsing step.

If you are shopping for a tool to measure the first funnel, twelve of them are compared here — every one measures visibility, and the comparison says which.

The two funnels are measured separately on purpose, because a site can be strong in one and empty in the other:

AI visibilityAgentic commerce readiness
The questionDoes the AI name you?Could its agent buy from you?
Who fails itSites the engines cannot read or do not trustSites that read fine and declare nothing
Where the loss showsNowhere in analyticsNowhere in analytics
Who measures itEvery AI visibility toolLoudPixel, as a level from 0 to 4

The free report grades both on any domain, in two sections, because a business that reads only the first half of its result will fix the wrong thing.

What makes a store ready for an agent

Strip away the protocols and it comes down to four things an agent needs, in the order it needs them:

  1. It can find what you sell. Your products exist somewhere a program can enumerate them, not only inside a page a human scrolls.
  2. It can read the terms. A price with a currency. Availability. Shipping and returns it can locate and quote back to its principal before spending their money.
  3. It can identify you and your policies. Who is selling, under what terms of service, with what refund promise — the documents a buying agent is required to check before it commits.
  4. It can complete and confirm. A path from "this one" to a confirmed order that does not depend on a human clicking a button the agent cannot see.

Each of those is a stage, and the levels below are those stages in order. A site that has done the fourth without the second is not Level 4; the agent never gets that far.

The five levels, and what an agent can do at each

Level 0 — invisible

The agent cannot tell what you sell. It may be able to read your homepage as prose, but nothing on the site declares a product, a price or an offer in a form a program can rely on. To a human the shop looks complete. To the agent it is a brochure.

What changes at the next level: the agent can at least tell you are a seller and what of.

Level 1 — listed

An agent can read your offer and find your terms — it cannot yet act on either. It knows what you sell and under what conditions, and can say so to its principal, but there is no path from that knowledge to a product it could put in a cart. This is where most sites that have "done their SEO" land: readable, quotable, unbuyable.

What changes at the next level: the agent can find your catalogue and knows how to deal with you.

Level 2 — discoverable

An agent can find your catalogue and knows how to deal with you: it can locate a specific product, read what it needs to compare it, and has established the terms of doing business. It can now put you in a shortlist with a number beside your name. It still cannot complete a purchase; a human has to take over at the last step.

What changes at the next level: the agent can finish the job.

Level 3 — buyable

An agent can complete a purchase from you today, on the path buyers in your category actually use — from finding the item to a confirmed order, with the policies it is required to check all locatable along the way. This is the level that turns a recommendation into a sale without a human clicking through.

One sentence appears on every report at this level, and it belongs here too: Level 3 means an agent can complete a purchase on your own rails. Buying inside ChatGPT is a separate, US-only programme we do not claim for you.

What changes at the next level: the store keeps working for agents after the first order.

Level 4 — operated

An agent can buy from you, and you are running the channel — feed fresh, promotions dated, policies live. Nothing left to close: the store also answers the questions that come after a purchase — order status, the invoice, how to cancel — in a form its principal's agent can read. Re-audit when your catalogue or your policies change; otherwise the work is done.

Why measure it now

Three dated facts, all from the platforms rather than from us:

  • Shopify publishes agent documentation covering discovery, checkout and order monitoring on its stores, with a command-line tool for developers building agents (shopify.dev/docs/agents).
  • The payment rails are moving the same way: Stripe's documentation now describes agent-initiated commerce flows, and the protocol work around agent checkout is public and versioned. LoudPixel's protocol watch records what each specification said and the day a person last read it.
  • Our own audit of our own site, on 1 September 2026, scored Level 0 — no offer an agent could read, no currency on any price, no refund policy it could find. We published the failure and the fix the same day, because a readiness grade from a business that had not passed it would be worth nothing.

The engines that recommend you are already reading in this mode. Our own server-side counter recorded 41 reads of loudpixel.ai's machine surfaces by AI assistants in 30 days — Anthropic 25, OpenAI 13 — none of which appeared in any analytics view, because those readers run no JavaScript. The agents are here; the question is what they find.

What a store owner should do this week

Not code. Three decisions, in order:

  1. Get your level. Run the free report on your domain. It states the level in plain language, says what an agent could and could not do, and dates the reading. It costs nothing and asks for no login.
  2. Read the second section before the first. The visibility section tells you whether you are named; the readiness section tells you whether the name would have converted. If you are named and Level 0 or 1, the second section is where the money is.
  3. Decide who does the work. The report says what is missing, not how to build it; that is deliberate, because the how depends on your platform and your developer. If you build the site yourself, the free extension grades any page as an agent reads it, one page at a time. If you want it done and re-measured, that is the audit and the fix pack, delivered by a person.

If you are the developer, or you are sending this to one, the technical version is a separate page: how an AI agent buys from your site walks the six steps in the order an agent meets them, says what has to be in front of it at each one, and names the published specification that governs each step with the day we last read it.

How to check your level for free

The free report grades any domain: the twenty-two questions a buyer would ask about your AI visibility and your agentic commerce readiness, answered in plain language, with the evidence underneath and the date on top. The levels themselves are defined once, on the methodology page, which also says plainly where a score stops being reliable — this post describes what each level means for you; that page is where the definition lives and is maintained. It is published as a page, so you can send it to your developer or quote it to a client. Run it on a competitor's domain too; the level beside their name is the fastest way to understand your own.

Frequently asked questions

What is agentic commerce?

Agentic commerce is an AI agent discovering products, comparing offers, and completing a purchase on a shopper's behalf, without the human clicking through each page. It is distinct from AI-assisted shopping, where a person stays in the loop at every step, and from a merchant's own internal AI tools.

How is agentic commerce readiness different from AI visibility?

AI visibility is whether an AI names you when a buyer asks what to use. Agentic commerce readiness is whether an AI agent could then find your product, read the price and terms, and complete the purchase. The first is measured by every AI visibility tool; the second is graded by LoudPixel as a level from 0 to 4.

What do the levels mean?

Level 0, invisible: an agent cannot establish what you sell or on what terms. Level 1, listed: it can read your offer and find your terms but cannot act on either. Level 2, discoverable: it can find your catalogue and knows how to deal with you, but still cannot complete a purchase. Level 3, buyable: it can complete a purchase from you today on your own rails. Level 4, operated: it can buy, and you are running the channel — feed fresh, promotions dated, policies live. The definitions are maintained on the methodology page.

Does Level 3 mean people can buy from me inside ChatGPT?

No. Level 3 means an agent can complete a purchase on your own site's rails. Buying inside ChatGPT is a separate programme, US-only at the time of writing, and we do not claim it for anyone.

I sell on a marketplace. Does this apply to me?

Not to your marketplace listings — the marketplace's own agent buys from its own catalogue, and nothing on your site changes that. It applies to anything you sell on a site you control.

How do I check my store's readiness level?

Run the free report on your domain. It is published as a dated page, states the level in plain words, and lists what an agent could and could not do. No login, no cost.


The bottom line. Being named by an AI is the first half of a sale that now happens in two halves. The second half is whether the agent that named you could buy — and that half fails in silence. Get your level, read the second section first, and then decide who closes the gap.

Does AI name your site? Free 60-second check.

See who ChatGPT, Perplexity, and Google’s AI name for your buyers’ questions.

Check free →

Written with AI assistance and reviewed by LoudPixel for accuracy.

Check your AI search visibility.

See which AI engines name your site — in about 60 seconds. Free, no signup.

Scan free