AI Can Shop for You. Can It Spend Your Gift Cards?

The gift card is in a kitchen drawer. The shopping assistant is on your phone. Somewhere between the two sits $80 that you have already paid for, received as a present or earned as a reward.

Now imagine asking an AI assistant to buy something with a final checkout total of $100. A useful assistant would notice the eligible $80 balance, ask permission to apply it and show you the remaining $20 before placing the order. An assistant that overlooks the card could complete the same purchase with $100 of fresh money. The product would arrive either way. Your wallet would tell a different story.

That is the question behind this investigation: when shopping moves into a conversation, does the money you already have at a merchant move with it?

The timing is unusually good. On September 16, 2026, PayPal described an internal comparison in which an AI assistant bought a gift card using its WebMCP approach. Buying the card was the task. Spending a card that somebody already owned was not demonstrated. That distinction runs through much of the industry's current story. [S01]

Our review of platform documentation, payment specifications and gift-card APIs finds a more interesting answer than a simple yes or no. Ordinary merchant checkouts already handle stored value. Developers have tools for applying it. One commerce protocol explicitly describes gift-card split payments; another has a gift-card proposal that leaves that particular combination outside its scope. What those building blocks do not establish is a universal assistant that finds your existing cards and reliably spends them across merchants and shopping channels. [S06] [S13] [S14] [S25]

A shopping habit is emerging; a redemption habit has not been measured

There is a real audience for the question. Statista Consumer Insights found that 24% of U.S. respondents aged 18–39 had used AI platforms to research products in the previous 12 months, compared with 18% of those aged 40–64. Enjoyment of AI shopping was reported by 23% and 15%, respectively. The underlying questions were fielded between June and September 2025. [S02]

These figures describe discovery and attitudes. They do not count orders placed by agents, gift cards purchased through an assistant or old balances redeemed at checkout. The distinction matters because a recommendation can be helpful long before the assistant is authorized—or technically equipped—to move money.

A shopper can use AI to compare a jacket's size, delivery date and reviews, then complete the transaction personally. Another can let an assistant prepare a basket but still approve the final payment. A third might delegate more of the task under a spending limit. Treating all three as the same behavior makes the market look easier to measure than it is.

The statistical evidence reviewed here does not supply a U.S. adoption rate for using AI to redeem existing gift-card balances. It supports the relevance of investigating that capability, rather than a claim that the capability is already commonplace.

The unused-balance problem is older than the chatbot

Gift cards were getting forgotten long before conversational shopping arrived. In Accenture's 2024 research, presented by Statista, 34% of respondents with an unused balance said they had forgotten the card. The same proportion cited a limited choice of merchants. A quarter considered the amount too small to bother using; 23% cited a restricted redemption channel, and another 23% reported technical difficulty redeeming it. Respondents could select several reasons. [S03]

A reminder could help with forgetting. It cannot, by itself, make a merchant accept somebody else's gift card, make an unavailable item appear or remove a restriction on where a balance can be used. The findings therefore suggest different jobs for an assistant, with very different integration requirements.

There is also a difference between remembering that a card exists and knowing its current value. An old email may contain the original denomination. A receipt may show the balance after a previous purchase. Neither necessarily proves what is available now. An assistant that proposes a payment needs a current answer from the merchant or the service authorized to maintain that balance.

Another Accenture question asked what would improve the gift-card experience. Easier balance checks led at 36%, followed by reminders at 35%. Easier redemption online and in stores, and more versatile usage options, each received 32%. [S04]

The consumer request is remarkably practical: make the value visible, remind me about it and make it easier to use. AI could be a useful interface for those jobs. The survey does not tell us whether any particular AI product performs them, or how much additional redemption it would cause.

For a sense of scale, Bankrate's September 2024 release, based on a YouGov survey, estimated approximately $27 billion in unused U.S. gift cards, vouchers and store credits. Forty-three percent of adults held at least one, and holders reported an average of $244. Those are a dated measure of outstanding value and a holder average—not annual losses, not an accounting breakage total and not a pot of money that AI can automatically recover. The online survey covered 2,373 adults, including 1,010 holders, on August 19–21, 2024. [S05]

For the accounting distinction, see our investigation into gift-card breakage and the money left behind. A balance still waiting to be spent is not proof that it will never be redeemed.

Buying, delivering and redeeming are three separate jobs

An announcement that an AI agent can buy gift cards sounds close to an announcement that it can spend them. The verbs conceal different transactions.

Buying a card exchanges new money for a prepaid product. Delivering it gets a code, link or other credential to the recipient. Redeeming it applies existing value to a purchase from an accepting merchant. The buyer of the card and the person redeeming it may not even be the same person.

Google's Gemini documentation provides a clear purchase example: its Google Play connected app can help users find and buy digital gift cards, with the purchase taking place in Google Play. The documented route requires, among other conditions, an adult user, a personal account and the Android Gemini app; availability varies. Google Play separately documents sending gift cards and distinguishes its own cards from specialty retailer cards. Neither document establishes a general Gemini service for spending a customer's existing retailer balances. [S10] [S11]

finperks supplies another concrete example. Its August 13 article pitches a gift-card API to AI-agent developers. The API documentation describes creating orders for gift-card products and presenting the resulting codes, emails or links. Those are purchase and delivery functions. The balance endpoint in the documentation concerns a client's funding account; it should not be mistaken for a lookup of whatever retailer card a consumer happens to own. [S20] [S21] [S22] [S23]

Paymentology's August 3 discussion approaches the opportunity through issuing infrastructure: controlled virtual credentials, spending limits and related controls. That is relevant to delegated payment. It is not evidence that an assistant can discover and redeem an unrelated merchant's existing gift card. A network payment credential and a closed-loop retailer code can both represent prepaid value while following different acceptance rules. [S19]

PayPal's recent example makes the distinction particularly visible. The company reported three structured tool calls versus more than 30 interface steps in its small internal gift-card-purchase benchmark, with lower inference cost and faster median checkout. Those are supplier-reported sandbox results, not an independent measurement of real-world redemption. Its described flow also brings the buyer into PayPal to review and approve payment; it was an optional, limited Store Sync rollout. [S01]

The important question after any such announcement is what happened to the value. Was new value issued? Was an existing credential delivered? Or was a merchant's outstanding balance actually reduced to pay for goods?

The $80 card and the $100 order

Consider the same item, at the same merchant, with a final total of $100 including applicable charges. The shopper has a valid $80 gift card accepted for that order. The following is an illustration of payment allocation, not an observed transaction or a provider benchmark.

Table 1. The product can be identical while the funding differs

The first route preserves $80 of fresh spending compared with the second. It does not create an $80 discount: the shopper has consumed an asset that already belonged to them. The third route introduces a real choice. A lower sticker price can coexist with a larger immediate cash outlay.

That choice should belong to the shopper. Someone expecting to spend the gift card next week might prefer the $90 offer. Someone trying to keep this month's bank-account spending down might prefer to use the existing balance now. An assistant cannot infer those priorities from price alone.