Loyalty Programs and Promo Codes in Agentic Checkout: Making Your Offers Visible to AI Agents

The Offer the Agent Never Saw

Most of the conversation about AI shopping agents is about getting chosen: product data, reviews, schema, and feeds. Less attention goes to what happens after the agent picks your product and starts checkout. That is where your incentives live: member pricing, loyalty points, free shipping for a tier, and the promo code from last week's email.

If an agent cannot see or apply those incentives, two things go wrong. The shopper pays a price they would not have paid on your site, or the agent compares you to a competitor on list price alone and you lose the sale. Merchants on the Shopify community forums have reported that loyalty rules often do not fire on orders that come through AI agents, because the order never passes through the storefront flows those apps hook into.

The fix is not a new loyalty app. It is publishing your incentives in a form agents can read, through the protocol the agent uses to check out.

How Google's Universal Commerce Protocol Models Incentives

Google's Universal Commerce Protocol (UCP), co-developed with Shopify and launched in January 2026, is the most fully specified example today. Shopify's UCP page says discounts, loyalty programs, subscriptions, fulfillment options, and terms are all supported by the protocol. In the current spec version (2026-04-08), incentives are split into two extensions that sit on top of checkout.

The discount extension

The discount capability is named dev.ucp.shopping.discount. A business can advertise it for cart, checkout, or both. Key behaviors from the spec:

  • Codes go in, applied discounts come out. The agent sends codes in discounts.codes. Your response lists every discount that actually applied in discounts.applied, each with a title and an amount.
  • Automatic discounts are first-class. Sales that apply without a code are returned with automatic: true, and the agent cannot remove them.
  • Rejections are explicit. If a code fails, you return a message with a specific reason, such as discount_code_expired, discount_code_invalid, discount_code_combination_disallowed, or discount_code_user_not_logged_in.
  • Cart discounts carry forward. Codes applied to a cart must be carried into checkout.
  • Stacking is ordered. A priority field controls which discount applies first.

The loyalty extension

Loyalty is a separate capability, dev.ucp.common.loyalty. Google's implementation guide is direct: you must advertise loyalty in your UCP profile, and because loyalty benefits often have a monetary impact, you must also enable discounts. In practice, member pricing shows up as an applied discount with clear display text, such as a "Club member benefit" line in the order totals.

The loyalty response can describe:

  • Program and membership. A program name plus a masked member ID, never the raw member number.
  • Tiers and benefits. Tier names with a description of each benefit, when your program has tiers.
  • Earning forecast. The points or rewards the shopper will earn on this order.
  • Multiple memberships. A shopper can hold more than one membership, for example a store club plus a partner program, and the response can apply benefits from each.

Claims vs. Verified Members

The most important design choice in UCP's loyalty spec is that the agent's statement "this shopper is a member" is treated as a claim, not proof. Your response has to reflect which state you are in:

  1. Verified. You confirmed membership, so you return provisional: false and may show the masked member ID.
  2. Accepted but unverified. You return provisional: true and must not show the member ID yet. The agent has to show the discount to the buyer as provisional.
  3. Cannot verify. You return a recoverable error so the agent can ask the shopper to sign in.

Google's guide recommends identity linking as the main way to verify membership. UCP's identity linking capability uses standard OAuth 2.0 with PKCE, so the shopper signs in to your account system once and the agent can then act with their personalized pricing. The spec states that identity linking upgrades the experience without gating it: guests can still buy, they just do not get member benefits.

Why This Is a Visibility Issue, Not Just a Checkout Issue

It is easy to file this under payments engineering. But incentives shape which product an agent recommends. An agent working for a shopper who belongs to your program can only weigh your member price if you expose it. If you do not, your offer looks the same as a competitor's list price, and the agent has no reason to prefer you.

The same applies to promo codes. Shoppers increasingly paste codes into an AI conversation. If your checkout returns a vague failure, the agent cannot explain why, and an unexplained failure at the last step is a reason to switch stores.

This connects to the broader data work covered in price and availability accuracy: agents compare prices across sources, and any mismatch between the price you advertise and the price at checkout erodes trust in your store.

A Practical Checklist

You do not need to implement UCP by hand to act on this. Most merchants will reach these protocols through their platform (Shopify, BigCommerce, Salesforce Commerce Cloud, and others) or a loyalty vendor. What you control is the data and the rules underneath.

Clean up your discount logic

  • Make rejection reasons specific. Audit what your platform returns for an expired code, a minimum-spend failure, or a non-stackable combination. Agents can only explain the reasons you give them.
  • Document stacking rules. If you cannot state in one sentence which discounts combine, an agent cannot either.
  • Retire zombie codes. Old influencer and affiliate codes that still technically work cause unpredictable agent behavior. Expire what you no longer intend to honor.
  • Prefer automatic discounts for broad sales. A sitewide sale that needs no code is the easiest incentive for an agent to apply correctly.

Make your loyalty program machine-readable

  • Write benefits as plain statements. "Gold members get free standard shipping" is easy to model. "Exclusive perks" is not.
  • Ask your loyalty vendor about agent support. Several loyalty and promotions vendors are building UCP and ACP extensions. Talon.One, for example, has announced loyalty and discount extensions for UCP under what it calls the Unified Incentives Protocol. Ask your vendor for a roadmap and dates, not marketing claims.
  • Check your account system for OAuth support. Identity linking depends on a standards-based sign-in. Legacy or heavily customized login flows may need work.

Test agent orders end to end

  • Place test orders through agent channels you are live on and confirm member pricing, codes, and points behave as they do on your storefront.
  • Reconcile points on agent orders. Confirm agent-placed orders earn rewards. If they do not, you are quietly penalizing your best customers for using AI.
  • Keep incentive data consistent between your feed, your site, and checkout. A sale price in Merchant Center that checkout does not honor is the kind of mismatch agents notice.

Where the Standards Stand

This is early. UCP checkout is available only to select merchants in the United States, and the spec is versioned and still changing. OpenAI's Agentic Commerce Protocol and other agent checkout flows handle incentives differently, and vendors such as Talon.One have publicly said the current UCP spec does not yet cover every promotion type. Stored-value redemption, such as spending points as payment, is noted in the spec as something for the payment instrument or a future capability.

The direction is still clear. Agents will apply the incentives you expose in structured form and ignore the rest. Merchants who get their discount rules and loyalty benefits into clean, explicit data now will be ready as each platform turns these capabilities on. For a refresher on the protocols themselves, see ACP vs AP2. For help auditing how your product and offer data reaches AI agents, see our technical foundation work.

Want to see how AI engines perceive your brand?

Get an Audit for $87