Glossary entry

Agent-readability score

The second half of the two-score model: can an AI shopping agent parse the product page well enough to recommend it — or draft it into a cart.

Last updated June 2026

The five checks

Product JSON-LD completeness: the structured fields an agent extracts first — name, brand, GTIN, offers, ratings — present and valid per schema.org/Product. Robots.txt admittance: whether GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and OAI-SearchBot are allowed to read the page at all. Machine-readable price and availability: structured offer data matching live store state, not stale or prose-only. Discoverable policies: returns and shipping terms an agent can find and quote, since agents factor them into recommendations. Agentic-checkout wiring: where applicable, whether the store participates in a checkout channel — Shopify Agentic Storefronts, or ACP/UCP endpoints for custom backends.

Why it matters for ecommerce

Agent-driven selection is stricter than human browsing. A human forgives a missing bullet; an agent scoring structured fields treats it as a gap and eliminates the SKU before price or reviews are weighed. A catalog that reads fine to shoppers can still underperform in agent flows — and those flows are expanding: ChatGPT Shopping and Perplexity's Buy with Pro already draft carts, and agent-completed checkout protocols are in pilot as of mid-2026.

The strategic point is that the score measures readiness without requiring a protocol bet. Whichever of ACP or UCP wins, the PDP groundwork is shared — clean schema, admitted crawlers, machine-readable price, availability, and policies. The same groundwork also lifts citations in today's research answers, so nothing is stranded if protocol timelines slip.

Reading the score: an example

A home-goods brand's best-selling fan scores 64/100: Product JSON-LD found but missing GTIN and review fields, GPTBot and ClaudeBot admitted but Google-Extended blocked by an old robots.txt rule, price structured but availability stale since the last theme update, returns policy present only inside a PDF (illustrative example). Each gap maps to a specific fix — two schema fields, one robots line, one template variable, one HTML policy page — and the score recomputes on the next refresh because every check reads the brand's own pages.

How it relates to the citation score

The citation score measures today's outcome: do engines recommend the SKU in research answers. The agent-readability score measures structural readiness for the cart-drafting and checkout stages of agentic commerce. They are reported side by side because the fix lists overlap heavily, and a SKU strong on one but weak on the other tells a precise story: cited but unparseable means fragile wins; parseable but uncited means the content and review work has not landed yet.

How eCommerce Insights computes it

All five checks run per SKU on every refresh, and each failing check ships with the concrete fix — a schema diff, a robots.txt line, a metafield value. The free Agentic Readiness Grader runs the same checks on any single PDP, no signup.

Related terms

Citation score — the first half of the two-score model.
Agentic commerce — the three-stage shift the score prepares a catalog for.
ACO (Agentic Commerce Optimization) — the discipline this score measures progress in.
Product schema — the largest single input to the score.
llms.txt — a related crawler-guidance signal, checked alongside robots.txt.

Ask AI about agent-readability scores

Have your preferred AI engine summarize this definition for your catalog.

Frequently asked questions

What does the agent-readability score actually check?

Five things per SKU: Product JSON-LD completeness, robots.txt admittance for AI crawlers, machine-readable price and availability, discoverable returns and shipping policies, and agentic-checkout wiring where applicable. Every check reads your own pages, so the score updates as soon as fixes ship — no waiting on engine re-crawls.

Do I need this if agent checkout is still in pilot?

Yes, because the earlier stages are already live. Research answers and draft carts (ChatGPT Shopping, Perplexity Buy with Pro) read the same structured data the score checks. Checkout protocols being in pilot as of mid-2026 just means the third payoff arrives later; the first two are current.

Which AI crawlers should my robots.txt admit?

At minimum the bots behind the engines you want citations from: GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, and Google-Extended. Blocking them removes your PDPs from retrieval entirely — the most common single cause of a low agent-readability score in eCommerce Insights's scans to date.

Is the agent-readability score a standard metric?

No — it is an eCommerce Insights-defined term, published as a canonical definition. The checks behind it reference public standards and specs (schema.org, robots.txt conventions, published protocol documentation), so the score is reproducible, but the composite and its name are this glossary's.

Can a SKU score high here and still not get cited?

Yes. Agent-readability measures whether the page can be parsed, not whether engines choose it. A parseable PDP with thin answer coverage or weak review signal loses citations on content grounds — which the citation score surfaces. Reading the two scores together is the point of the model.

Go deeper

Agentic Readiness Grader — free single-PDP version of these checks.
Agentic commerce solutions — the readiness-without-protocol-betting strategy.
Schema for AI search — how to fix the JSON-LD input.
AI Agent Lens docs — per-bot crawler access evaluation in the product.

See where every product in your catalog stands on this. Start a 14-day free trial — no credit card — or grade one PDP free in 30 seconds.