A wave of “AI readiness” checkers appeared this year. Point one at your site and it scores your llms.txt, your markdown fallbacks, your schema markup, your MCP server, your presence in AI search indexes. Useful work — and almost all of it measures one thing: whether an AI system can read and cite you. Call that AEO — answer-engine optimization, the successor to SEO.

E-commerce has a harder question. When a shopping agent arrives with intent and a payment method, can it find the product, resolve the price and stock, add to cart, and complete a purchase? That’s a different property of a site. We call the work of optimizing for it AOO — agentic ordering optimization — and the gap between the two is where merchants are currently being misled.

The same store, two verdicts

We’ve now watched both kinds of audit run against the same domains, including our own, and the divergence is structural, not noise:

  • A content site can score high on readability and be unbuyable. Clean markdown, perfect llms.txt, generous docs — and no machine-readable Offer, no cart an agent can reach, checkout behind a bot wall. A general-purpose “AI readiness” rubric can hand it a B while a purchase attempt dies at step one.
  • A mature Shopify store can score low on readability and transact fine. One audit we reconciled scored a working skincare shop 41/100 — docking it ~20 points for missing a developer portal, an npm SDK and a CLI. It’s a soap merchant. Its product JSON-LD, feed, price and stock were flawless on the pages an agent actually buys from.

Neither verdict is dishonest. Each rubric measures what its authors care about. The problem is that both get marketed under the same phrase — “agent-ready” — and a merchant reading a single number can’t tell whether it predicts citations or transactions.

Where the two audits actually diverge

The readability layer and the transaction layer overlap maybe a third. The rest splits cleanly:

AEO checks (can it read you): llms.txt and markdown twins of your pages · schema breadth (FAQ, Article, Organization) · crawlability for AI user-agents · presence in AI search and app indexes · docs for humans and models.

AOO checks (can it buy from you): Product/Offer JSON-LD with price, currency, availability on the product page itself · a feed or API exposing the full catalog, not the homepage · cart and checkout reachable without JavaScript gymnastics or bot-blocking · shipping and returns stated machine-readably · payment rails an agent can complete · and identity — an agent that confuses you with a name-twin never reaches checkout at all.

The failure modes are different too. Readability fails soft: the agent cites a competitor instead of you, and you never know. Transactability fails hard: the order doesn’t happen, revenue walks. One is marketing loss; the other is a closed door on a sale that was already decided.

Why the confusion persists

Because readability is cheaper to audit. Fetching llms.txt is one request; attempting a purchase path takes a real crawler with a rubric per platform, and checking price/stock truthfulness means reading the catalog, not the homepage. So the market filled up with readability scores first, and “agent-ready” quietly came to mean “AI-legible” — a much weaker claim than merchants think they’re buying.

The data points the other way. The agent traffic that matters to a store isn’t reading blog posts — it’s trying to buy, and it gives up at the first hard failure. Our own scans across the Agent-Ready Index keep finding the same shape: stores whose content layer is fine and whose transaction layer loses double-digit points on fixable, boring things — an Offer missing from the pricing page, stock absent from the feed, checkout fenced off from non-browser clients.

What to do with any score you’re shown

Ask one question of the rubric before you act on the number: which checks fail a purchase, and which only fail a citation? Fix in that order:

  1. Identity — one unambiguous entity (Organization JSON-LD, sameAs, a disambiguation page if your name is shared). An agent that can’t resolve who you are never gets further.
  2. The Offer where the question lands — price, currency, availability in JSON-LD on the product and pricing pages themselves, not just the homepage.
  3. The path to checkout — cart reachable headlessly, no bot-wall on the buy flow, shipping and returns machine-readable.
  4. Then the readability layer — llms.txt, markdown twins, schema breadth. It compounds the first three; it doesn’t replace them.

Our free scan is built around that ordering — it weighs access and the transaction path above documentation polish, because that’s the order in which revenue is lost.

Sources

A store an agent can quote but not buy from is a brochure. The audits that matter next measure the sale.