# Merchant readiness when the buyer may be an external AI agent

Companion to **Your next customer may be an AI agent**.

This is **channel readiness**, not the 13-check copilot score in [`shopping-agent-readiness-checklist.md`](shopping-agent-readiness-checklist.md). Score **your** store as an input to someone else's shopping agent. A passing copilot checklist does not imply a passing channel score (and the reverse).

Not a client engagement. Not a conversion KPI. Adobe July 2026 AI-referral figures belong in the measurement artifact, not here as a target you will "hit."

## Buyer path (what you do not control)

```text
Human shopper
    → AI shopping agent (not yours)
    → Discovery
    → Comparison
    → Merchant selection
    → Purchase (your existing checkout / handoff)
```

You control whether each hop has **facts**. You do not control the agent's ranking prompt.

## Channel scorecard (one category)

Check a box only if an external system can get the fact **without rendering HTML**. Count checked items. Below 8: you are not a reliable merchant for agentic discovery. 8–12: one category can be a feed. 13+: you can evaluate agentic commerce without lying about stock and price.

### Product data

- [ ] Unique SKU / GTIN where the channel requires it
- [ ] Title and description match the SKU the warehouse picks
- [ ] Structured attributes (size, material, color, compatibility, width) as fields
- [ ] Variant parent/child IDs are explicit
- [ ] Restricted categories flagged (age, hazardous, Rx)

### Offer facts

- [ ] Price + currency + tax/shipping disclosure rules checkout already uses
- [ ] Inventory current, or honest `unknown` (not a stale in-stock badge)
- [ ] Shipping promise / delivery window as data, not only PDP prose
- [ ] Return / warranty copy versioned and fetchable
- [ ] Discount eligibility is a rule in OMS — agents must not invent coupons

### Evidence and access

- [ ] Reviews or ratings available as structured records (or an honest "none")
- [ ] Catalog / inventory / price APIs (or a documented feed) with auth and rate limits
- [ ] Machine-readable product pages (JSON-LD or equivalent) — HTML is presentation
- [ ] robots / bot policy does not block the discovery surfaces you actually want
- [ ] No requirement that the only catalog is a JavaScript-only storefront

## When NOT to treat this as a launch

- Inventory is a nightly dump and in-stock is a CSS class
- Price lives only in HTML after a client-side render
- You plan to "let the agent scrape the PDP" instead of publishing a contract
- Leadership wants an autonomous checkout tool in week one

## Score

| Score | Action |
| ----- | ------ |
| < 8 | Fix data and APIs. Do not staff an agentic-commerce sprint. |
| 8–12 | Pilot **one** category as a feed + golden questions. |
| 13+ | You can evaluate merchant selection without inventing stock and price. |

Reuse the copilot list only after this score is honest: [`shopping-agent-readiness-checklist.md`](shopping-agent-readiness-checklist.md).