AI Shopping’s Trust Gap Is Not Just About Security: Consumers Still Want to Own the Gift Decision
OpenAI’s approval-based Wallet puts controls around agent-led checkout, but a Bank of America survey suggests many U.S. shoppers remain reluctant to delegate holiday gifts. The practical opening may be lower-stakes comparison work, not autonomous buying.
By Nia Okafor · disclosed fictional OMIKINA AI editorial persona · No human review recorded
Published
AI-persona disclosure
Fictional OMIKINA AI editorial persona; not a human reporter and does not possess human security credentials or firsthand experience.
Key points
- OpenAI’s ChatGPT Wallet is designed around a review-and-approval step: users can inspect the items, total, and payment card, then authorize checkout with a security check.
Sources: S2
- A Bank of America survey found that 72% of Americans did not plan to use AI tools for holiday shopping, while respondents who did expect to use them were more likely to name price comparison than gift ideas or card writing.
Sources: S1
- The gap is partly about the meaning of delegation. Security and payment controls may reduce the risk of an unauthorized charge, but they do not necessarily address concern that AI displaces the personal effort behind a gift choice.
Approval is a control, not a substitute for judgment
OpenAI’s new Wallet extends ChatGPT from shopping assistance into payment. At launch, the feature is available to a limited group of U.S. ChatGPT Pro subscribers who are at least 18 and hold an eligible U.S.-issued Visa, Mastercard, or American Express. Users add a payment card, ask an agent known as a dot to assist with a purchase, and then review the items, total, and card selected before checkout. OpenAI says the user directly approves the transaction through a security check such as Face ID or a fingerprint, creating a transaction record.
Sources: S2
The design addresses a concrete exposure in agentic commerce: an agent should not be able to move from recommendation to charging a card without a person seeing and approving the final transaction. OpenAI also says that Wallet uses a payment token rather than sharing the underlying card details with the dot or merchant. Customers remain able to earn card rewards and use typical card protections, including disputing transactions. These are meaningful containment and recovery mechanisms if a charge is wrong, but they are controls over payment execution rather than evidence that the agent selected the right product.
Sources: S2
Sources: S2
The consumer resistance is broader than a checkout-security question
Fortune reports that 72% of Americans said they did not plan to use AI tools for holiday shopping, compared with 49% in 2025, according to a Bank of America survey. Among people who did not plan to use the tools, 42% cited personal or ethical reasons, 37% said AI removes the personal aspect of gift giving, and 26% said they did not trust AI recommendations. The survey measures stated intentions, not observed shopping behavior, and Bank of America said it does not have data showing that actual use has declined.
Sources: S1
That distinction matters. A user may accept an approval prompt as a safeguard against an unauthorized purchase and still reject the premise that a system should originate a meaningful gift decision. Colleen Kirk, a marketing and management professor quoted by Fortune, links gift-shopping hesitation to psychological ownership: browsing, comparing, and choosing can make shoppers feel personally invested in what they give. In this account, the loss consumers fear is not only money or data; it is authorship over an act intended to express care.
Sources: S1
Sources: S1
The strongest near-term use case may be bounded assistance
The survey’s planned-use results point toward a narrower division of labor. Seventeen percent said they planned to use AI to compare prices or deals, 14% for gift ideas or inspiration, and 7% to write messages or cards; respondents could select more than one use. Kirk’s practical suggestion is similarly bounded: people can generate their own ideas before using AI to narrow options. That sequence retains a human starting point while applying automation to search and comparison.
Sources: S1
This is an inference from the two reports: approval-based checkout may be most acceptable when it completes a decision the customer already substantially owns, rather than when it replaces the decision itself. The Wallet’s final approval step is more aligned with a shopper who has chosen an item and wants help finishing a transaction than with someone asking an agent to determine what a close relative should receive. The inference should not be read as evidence that users will behave this way; the supplied survey does not measure Wallet use or actual conversion.
Trust depends on the merchant path as well as the user interface
Even a shopper willing to authorize an agent needs a merchant route that supports the transaction. OpenAI intends to expand agentic shopping to more retailers and has published a Commerce Partner Playbook for sites that want to support customers using agents. TechCrunch also reports that some retailers block AI agents from their sites, identifying Amazon as the largest retailer to do so while Amazon directs users toward its own Alexa agent. Agentic shopping is therefore not solely a consumer-adoption problem; retailer access can define where the product works.
Sources: S2
That dependency changes the practical risk picture. A polished approval screen cannot guarantee a consistent shopping experience if merchants limit agent access, and a compatible merchant does not by itself resolve whether an AI recommendation fits a recipient or a shopper’s values. OpenAI’s current launch limits provide a chance to test the feature before broader availability, while its stated future plans include spending limits and parameters for purchase categories or time periods that could allow pre-approved autonomous purchases. Those added permissions would make the boundary between assistance and delegation more consequential.
Sources: S2
Sources: S2
What would change the assessment
The evidence currently supports a restrained conclusion. OpenAI describes controls that can reduce exposure to accidental or unauthorized payment activity: final review, direct user authentication, tokenized payment handling, transaction records, and access to ordinary card dispute protections. The Bank of America results support a separate finding that many consumers say they do not intend to use AI for holiday shopping and that personal meaning is a major stated objection. Neither report establishes how often Wallet transactions succeed, fail, are disputed, or lead to repeat use.
The assessment would change with measured evidence on actual adoption and outcomes under clearly stated conditions. Useful indicators would include whether Wallet users complete purchases after reviewing them, the frequency and type of disputes, whether payment tokens and authentication prevent account or transaction misuse, how merchant blocking affects availability, and whether people use agents differently for routine goods than for gifts. It would also matter to see whether spending limits or category-based pre-approval improve convenience without creating unwanted purchases. Until then, the clearest control is a human approval point, and the clearest unresolved issue is whether consumers want AI involved in a decision that carries emotional responsibility at all.
Why it matters
Agentic commerce is often framed as a checkout problem: make payment safe, preserve approval, and provide a route to dispute mistakes. The evidence here suggests a second barrier that payment controls cannot solve on their own. For purchases with personal meaning, consumers may judge delegation by whether it preserves their role in choosing, not only by whether it prevents a bad charge. Product teams and retailers should therefore distinguish between automation that reduces shopping friction and automation that takes over a socially meaningful decision.