AI Shopping Agent Optimization Checklist (2026)

Agentic shopping has started shipping. ChatGPT Search and Operator, Perplexity Shopping, Google AI Overviews, Microsoft Copilot — they're all transacting on shoppers' behalf today. The catalogs that win are the ones an agent can parse cleanly. This is what to ship against, in priority order. Treat it as a backlog, not a checklist to clear in one sprint — the first eight items cover most of the gain.

Foundation — make sure the agent can read the page

1. Render-parity with a real browser. Whatever a human sees, an agent crawler with a headless browser should also see. No critical content that only loads after a five-second JS delay or behind a click.
2. robots.txt allows agent crawlers. Don't accidentally block ChatGPT-User, PerplexityBot, Google-Extended, or ClaudeBot. Check against the published user-agent strings.
3. Canonical and hreflang are clean. Agents prefer a single canonical URL per product. Variant URLs should rel="canonical" to the parent, not the reverse.

Structured data — give the agent something machine-readable

4. Schema.org Product JSON-LD on every PDP. name, sku, brand, image, description, offers, and as many category-specific properties as apply. Validate with Google's Rich Results Test and Schema.org's own validator.
5. Offer node per variant, not aggregate-only. Use AggregateOffer at the parent if needed, but always emit per-variant Offer nodes so the agent can compare colors, sizes, and per-variant prices.
6. GTIN, MPN, brand populated. Empty strings count as missing. These identifiers are how agents cross-reference your product against reviews and prices from other sources.
7. BreadcrumbList for category context. Helps the agent place the product in your taxonomy.

Spec table — make the facts unambiguously extractable

8. Specs in a <table> or <dl> — not prose. Definition lists or HTML tables. Avoid spec attributes embedded inside hero images or marketing copy.
9. Consistent units across the catalog. Pick one unit system per attribute and stick to it. Mixing oz and ml in the same catalog makes you look unreliable.
10. Enumerations from a defined vocabulary. Use a fixed palette of color names, a fixed size taxonomy, a fixed set of material names. Stop the synonym sprawl.

Accuracy — don't contradict yourself

11. Title matches spec. "12oz steel bottle" in the title and "350ml aluminum" in the spec is a disqualifier — conservative agents skip products with contradictions.
12. Image alt text describes the product. Not "Untitled-1.jpg". Alt text is one of the few cues an agent has for image content.
13. Pricing matches between schema and visible page. JSON-LD price should match what the user sees. Agents catch the discrepancy and lose trust.

Integrity — don't look stale

14. No broken image URLs. Every image returns 200. CDN-served preferred over origin-hosted.
15. Unique SKUs and identifiers. No duplicates across the catalog. Duplicate GTINs are a quality red flag.
16. Category and related-product links resolve. No 404s in the navigation an agent might follow.

Trust signals — make the agent willing to recommend you

17. Reviews exposed via Schema.org AggregateRating. Inline ratings in JSON-LD, not just in a JavaScript widget the agent can't reach.
18. Return policy and shipping marked up. Schema.org MerchantReturnPolicy and shippingDetails inside the Offer node.
19. Availability is current and correct. InStock / OutOfStock / PreOrder in JSON-LD. Stale availability is worse than no availability.

Don't ship in checklist order. Ship in score order.

Run a free Bhoroli audit on your live storefront. The result tells you which of these items your catalog already passes, which it fails, and which fixes will move your PDQ score the most. Ship those first; come back to the rest next sprint.

Related: the PDQ framework · agentic commerce readiness audit · Shopify product data audit