Jobs to be done · Diagnose · SEO · Ecom

How to check if AI agents can read your PDPs.

If a shopping agent landed on your product page right now, could it actually find the price, the stock status, and the returns policy? Pages built for humans hide those facts in JavaScript, images, and accordions that agents parse badly — and an agent that can't parse the page won't draft it into a cart.

Quick answer

Paste any PDP into the free AI Crawler Simulator to see the page exactly as an AI crawler does — 10-bot robots.txt verdicts, the plain-text view, Product JSON-LD and JS-shell checks. Then run the Agentic Readiness Grader for the agent-readability score with a fix attached to every failed check. The paid product scores every SKU on every refresh.

The slow way: read your PDP like a bot

The manual audit is doable for a handful of pages. Fetch the PDP with curl or a browser's "view source" — not the rendered DOM — and ask four questions. Is there a complete Product JSON-LD block with price, availability, sku, and brand in the raw HTML? Does /robots.txt admit GPTBot, ClaudeBot, PerplexityBot, and Google-Extended? Could a parser extract the price without executing JavaScript? Can you reach the returns policy through a plain link an agent would follow?

Most teams find at least one surprise per page: the price exists only after a JavaScript hydration pass, the size chart is a JPEG, the returns policy lives in a modal with no crawlable URL. The method breaks down at catalog scale — checking four signal groups across 1,500 SKUs by hand is not a job anyone finishes — and it goes stale the next time the theme updates or an app injects markup.


The eCommerce Insights way

  1. Grade one PDP free. Run the Agentic Readiness Grader on a hero SKU. The score lands in about 30 seconds with per-bot crawler access shown as an allow/deny ledger.
  2. Read the four signal groups. Product JSON-LD completeness; robots.txt admittance for the AI crawlers; machine-readable price and availability; discoverable returns and shipping policies. Where applicable, the grader also flags agentic-checkout wiring (Shopify Agentic Storefronts, ACP discovery files — hedged, since the protocols are in pilot as of mid-2026).
  3. Scan the catalog. On a paid plan every SKU gets the agent-readability score alongside its citation score on each refresh — weekly on Starter, daily on Growth — so a theme update that breaks JSON-LD shows up within a cycle, not a quarter.
  4. Ship fixes as diffs. Missing schema fields, robots.txt lines, and metafield values arrive as reviewable diffs. Approve and push to Shopify via the admin API (Growth, early access) or export the CSV.
  5. Re-test. Rerun the grader after shipping, and spot-check with the AI Crawler Simulator — it renders the plain-text view a non-JS crawler extracts. The pass condition is simple: everything an agent needs to draft a cart — price, availability, variant, policy — present in machine-readable form on first fetch.

The wider context — what agents do with readable PDPs, and what ACP and UCP change — is covered in the agentic commerce solution and the ACP guide. Schema.org's Product type reference is the canonical field list.

What "good" looks like

Product JSON-LD with price, availability, sku, brand on every PDP100%
AI crawlers admitted in robots.txt (GPTBot, ClaudeBot, PerplexityBot, Google-Extended…)allow
Price/availability parseable without JavaScript executionyes
Returns and shipping policies reachable by crawlable URLyes

An agent-readability score above 80 generally means an agent can complete the research-and-draft loop on the page. Scores below 50 almost always trace to one structural cause — blocked crawlers or client-side-only rendering — rather than many small ones.

Ask AI about this job

Have your favorite AI engine apply this walkthrough to your store.

Frequently asked questions

What does the agent-readability score actually check?
Four signal groups per SKU: Product JSON-LD completeness (price, availability, sku, gtin, brand, reviews), robots.txt admittance for the AI crawlers that matter, machine-readable price and availability (not rendered only by JavaScript or baked into images), and discoverable returns and shipping policies. Where applicable it also checks agentic-checkout wiring. Full definition in the glossary.
Why would an agent fail to read a page that looks fine to humans?
Because pages built for humans hide critical facts in places agents parse poorly: prices rendered client-side by JavaScript, size charts shipped as images, returns policies buried in accordion components, availability shown only as a styled button state. An agent drafting a cart needs the fact in markup, not in pixels.
Does this matter if agent checkout isn't mainstream yet?
Yes, because the same signals feed today's surfaces. ChatGPT Shopping and Perplexity Buy with Pro already draft carts from PDP data, and the research engines use the same structured signals to decide what to recommend. Full agent-completed checkout via ACP and UCP is in pilot as of mid-2026; the PDP groundwork is identical either way.
Is agent-readability the same as my citation score?
No — they answer different questions. The citation score asks whether AI engines recommend the SKU in shopping answers. The agent-readability score asks whether an agent landing on the PDP can parse it cleanly enough to recommend it or draft it into a cart. They improve with overlapping fixes, which is why eCommerce Insights reports both per SKU.
Can I check a headless or custom storefront?
Yes. The grader is URL-based and works on any storefront. Headless stacks fail most often on client-side rendering — schema that exists in the React tree but not in the served HTML. See audit a headless storefront for the rendering-specific checks.

See your PDPs the way an agent does.

Agent-readability scored per SKU, fixes attached. 14-day trial, no credit card.