Agentic Readiness Grader.
When ChatGPT Shopping or Google AI Mode tries to draft your product into a cart, six signals decide whether the transaction can even start. Four of them are readable from a public URL, and the grader scores those live: the UCP profile, the ACP discovery file, Product JSON-LD, and robots.txt admittance. The other two sit inside Google Merchant Center — the grader names them and tells you where to look instead of guessing.
Rate limit: 1 run per 30 seconds per IP. The grader reads only public endpoints — the URL you enter, its robots.txt, and its /.well-known/ files. It has no access to your Merchant Center account, so two of the six checks below are reported rather than scored.
Six checks gate agent checkout. Four can be read from a public URL.
All six belong on the list — each one can keep a product out of an agent's cart on its own. Only four of them leave a trace a public crawl can see. The grader scores those four and states plainly that the other two are outside its reach, because a number invented for a signal nobody measured is worse than a blank.
| Check | Why it gates agentic commerce | What this tool can do | Fix guide |
|---|---|---|---|
/.well-known/ucp profile | Required for Google's Universal Commerce Protocol, in pilot as of mid-2026. No profile, no participation in AI Mode checkout. | Scored. Fetched live from your domain. | UCP guide |
/.well-known/acp.json discovery | Proposed discovery file for the OpenAI/Stripe Agentic Commerce Protocol. RFC-stage as of mid-2026 — absence is informational, not a failure. | Scored. Fetched live, reported as informational. | ACP guide |
native_commerce opt-in | Per-SKU UCP eligibility flag, shipped through a supplemental Merchant Center feed. No flag, no agent checkout for that product. | Not scored. Nothing on your storefront reveals it. Needs Merchant Center access. | UCP guide |
| Merchant Center feed health | Disapproved or misclassified items are dropped from agent surfaces before any protocol logic runs. | Not scored. Item-level status lives in your Merchant Center account, not on the page. | Agentic commerce |
| Product JSON-LD completeness | Agents resolve price, availability, and identity from schema. A missing gtin or availability field breaks cart drafting. | Scored. Parsed from the URL you enter. | Schema for AI search |
| robots.txt AI-crawler admittance | A blocked crawler never reads the PDP, which removes the product from research and checkout paths alike. | Scored. Evaluated bot by bot against the AI crawler roster. | Crawl surface guide |
How to check the two the grader cannot
Both blind spots are Google Merchant Center data, and both are short checks for whoever holds standard or admin access to the store's Merchant Center account. Ask that person for the following, or get access yourself.
- Feed health. Open the Products view in Merchant Center and read the item-level issues. Disapprovals and category misclassifications remove products from agent surfaces before any protocol check runs, so clear those first.
native_commerceopt-in. Find the supplemental feed that carries the attribute and confirm it setsnative_commerceon the products that are actually eligible — subscriptions, services, rentals, and age-restricted items generally are not. The primary feed stays untouched. The UCP guide walks the feed setup.
If both come back clean and the grader's four public checks pass, the store's agentic-checkout groundwork is in place as far as anything outside Google can tell. That is the honest ceiling on a free tool, and it is worth knowing where the ceiling sits before a vendor tells you otherwise.
How to read the score
The composite is the share of available points earned across the four checks the grader can verify from public URLs. Nothing else is in the arithmetic: native_commerce and Merchant Center feed health carry no points in either direction, so they cannot inflate a score or drag one down. Each finding prints its own points earned and points available, so the denominator is visible rather than implied.
Bands: 80 and above, ready on the public layer; 60–79 partial, usually robots.txt drift or thin Product JSON-LD on top sellers; 40–59 work to do, treat the score as a planning input for next quarter; below 40 foundational gaps — read make my store agent-ready before re-running. Read a high score for exactly what it is: a store can pass all four public checks and still be shut out of UCP checkout because native_commerce was never set on its products. The score measures the layer the grader can see.
Two results are not bands. Partial scan means one of the scorable checks did not run — a storefront that serves no reachable product page, for instance — so the percentage covers fewer checks than usual and should not be read as a readiness verdict. Not measured means nothing scorable completed at all, and no number is reported rather than a misleading zero.
Why this matters now: agent-completed checkout is arriving in stages. ChatGPT Shopping and Perplexity Buy with Pro already draft carts; ACP and UCP are in pilot as of mid-2026 for full agent-to-store purchases. Stripe's ACP documentation describes the checkout flow the discovery file feeds. The protocols are young and the grader hedges accordingly — RFC-stage items are reported as informational. What is not speculative is the dependency chain: no crawler access means no research visibility, no schema means no cart drafting, no UCP profile means no AI Mode checkout. For the strategic picture, see the agentic commerce optimization guide.
Protocols change monthly. The public layer should be re-checked weekly.
eCommerce Insights runs these same public-endpoint checks on a schedule across the whole catalog, product by product, and pairs them with citation tracking across ChatGPT, Perplexity, Google AI Overviews, Gemini, Claude, and Copilot — so you see both whether an agent can read the PDP and whether the engines recommend it in the first place. The Merchant Center layer stays your team's to verify. See the agentic commerce solution.
Ask AI about the Agentic Readiness Grader
Have your favorite AI engine summarize this tool for your storefront.
Frequently asked questions
What does the Agentic Readiness Grader actually check?
/.well-known/ucp profile Google's Universal Commerce Protocol requires, the proposed /.well-known/acp.json discovery file for the OpenAI/Stripe Agentic Commerce Protocol (RFC-stage as of mid-2026, so absence is informational), Product JSON-LD completeness on the URL you enter, and robots.txt admittance bot by bot across the major AI crawlers. Two further checks are named but not scored: the native_commerce opt-in and Merchant Center feed health, neither of which is visible from outside your Merchant Center account. The first findings render free; the rest come with an email.Why can't the grader check native_commerce or my feed health?
native_commerce opt-in is a per-SKU attribute shipped through a supplemental Merchant Center feed, and item-level feed health is Merchant Center data — both sit behind Google authentication that a free public-URL tool has no access to. The grader reports them as not determinable, says what they gate, and leaves them out of the composite rather than inventing a number. To check them yourself you need standard or admin access to the store's Merchant Center account: the Products page for item-level disapprovals, and the supplemental feed that carries native_commerce for per-product eligibility.What is agentic commerce readiness and why does it matter in 2026?
Is this the same as the AEO Grader?
What does the composite score mean?
native_commerce and feed-health checks carry no points in either direction, so they neither inflate nor deflate the number. Bands: 80 and above is ready on the public layer; 60–79 is partial, usually robots.txt drift or thin Product JSON-LD; 40–59 is a planned fix cycle; below 40 means protocol-layer work has not started. A high score is not proof of eligibility — a store can pass all four public checks and still be shut out of UCP checkout if native_commerce is never set on its products.Do I need /.well-known/acp.json today?
Four checks scored. Two named. Nothing guessed.
The grader turns "are we ready for agentic commerce?" into an answerable question, and marks the part of the answer no public crawl can reach.