How AI shopping agents find and read your products

The discovery gap the payment rails don’t close

The standards comparison on this site answers “how does an agent complete a purchase?” But before any checkout, an agent has to discover your product, understand it, and decide it is the right answer to a shopper’s prompt. That decision happens inside a generative engine — ChatGPT Shopping, Google AI Mode, Perplexity, Gemini — that reads structured data, not the rendered page a human sees. If the machine-readable layer is thin or contradictory, you are invisible at the exact moment the agent is choosing what to recommend.

The traffic is no longer hypothetical. Over the 2025 holiday season, traffic to US retail sites from generative-AI tools rose 693% year over year — the largest jump of any industry (Adobe Analytics). Salesforce put a number on the downstream value: shoppers arriving from AI search channels converted nine times more often than those coming from social referrals (Salesforce). And the share of purchases touched by AI is compounding: Adobe’s mid-2026 data put AI-influenced shopping journeys at 18.3% of tracked online transactions in Q2 2026, up from 4.1% two years earlier.

How an agent actually reads a product page

A generative engine does not “look” at your storefront. It fetches a representation of the page — often without running your JavaScript — and extracts a structured record: title, description, price, availability, identifiers (GTIN/SKU), images, ratings. It then cross-checks that record against your feed and any agentic endpoints, and only a coherent, complete record earns a place in the answer. The pipeline below is the same one the agentic protocols formalise at the discovery stage.

Five signals do most of the work at the discovery and comparison stages:

• schema.org Product & Offer — the canonical, structured description of the item. Incomplete Offer markup (missing price, availability, priceCurrency) is one of the most common reasons an agent cannot quote your product confidently.
• Server-rendered price and availability — if these only appear after client-side JavaScript runs, a crawler that does not execute scripts sees an empty or stale value.
• Crawler access — robots.txt and the emerging llms.txt convention decide whether AI crawlers (GPTBot, PerplexityBot, Google-Extended and others) are even allowed to read you.
• Agentic endpoints under /.well-known/ — discovery files and agent cards that ACP and UCP use to advertise that your catalogue is transactable by agents.
• Claims coherence — the price, stock and ratings must agree across HTML, schema.org and feed. Disagreement reads as untrustworthy data and gets filtered out.

A neutral, no-email way to see what an agent reads on your own store is a third-party GEO audit. Verity Score, for example, publishes a page-by-page GEO readiness framework for Shopify catalogues that scores exactly these signals — agent access, server-side price/availability, schema completeness and claims coherence — and you can run a free GEO audit of a storefront without handing over an email. It is the clearest illustration of the discovery gap described above, on your own URLs.

Watch: what agentic shopping looks like end to end

Shopify’s own walkthrough shows the merchant side of the pipeline — how its Catalog, Universal Cart and Checkout Kit let an agent search hundreds of millions of products and complete checkout inside an AI conversation. It is the clearest two-minute picture of why machine-readable product data, not page design, is what decides whether you appear.

What the research proves about getting cited

The single most-cited study on this question is GEO: Generative Engine Optimization, presented at ACM SIGKDD (KDD) 2024 by researchers from Princeton, IIT Delhi and the Allen Institute. Testing across a benchmark of diverse queries — and validating on the live Perplexity.ai engine — the authors found that targeted content changes could boost a source’s visibility in generative answers by up to 40%. Crucially, the methods that worked are not keyword tricks; they are credibility signals.

A merchant checklist for agent-readiness

Working through the discovery and comparison stages, the concrete things to verify on a storefront:

• Validate schema.org Product/Offer on every product URL, with price, priceCurrency, availability and a GTIN/SKU present.
• Confirm price and availability are in the server-rendered HTML, not injected by client-side JavaScript only.
• Check robots.txt and llms.txt allow the AI crawlers you want (and block the ones you do not).
• Make sure your /.well-known/ discovery and agent-card endpoints respond, so ACP/UCP-aware agents can transact — see how to accept payments from AI agents once they arrive.
• Reconcile claims across HTML, schema and feed so price, stock and ratings never disagree.
• Apply the GEO findings: cite sources, add credible quotes and concrete statistics to product copy.

You can run these checks by hand, or with a dedicated GEO auditing tool that grades a Shopify catalogue page by page against the same signals; Verity Score’s free GEO audit is one no-email option for a first pass.

Where this fits with the payment standards

Discovery and payment are two halves of the same loop. Get found and correctly read at steps 1–2, then get paid cleanly at step 3. To go deeper on the payment half, start with the standards comparison, understand how AI agents pay, and see how to accept agent payments. For the latest moves across the ecosystem, read the June 2026 roundup and the European launches. The dated timeline lives on the changelog.