ChatGPT Wallet lets agents pay for you. A study found chatbots showing fake shop links in up to 15% of answers
Agents can now pay for things inside ChatGPT. The hard part was never the payment, it was knowing the agent picked the right store.
What happened
OpenAI launched ChatGPT Wallet on November 13. It lets its agents, called dots, help choose a product and then complete the purchase. At launch it is limited to some US ChatGPT Pro subscribers aged 18 and up with a US issued Visa, Mastercard or American Express card, according to TechCrunch.
Before every checkout the user sees the items, the total and the card, then approves with Face ID or a fingerprint. The charge goes through a payment token, so neither the agent nor the merchant sees the real card number. OpenAI says card rewards and dispute protections still apply. The Decoder reports this replaces Instant Checkout, OpenAI's earlier attempt, and that OpenAI also published a Commerce Partner Playbook to help merchants stop blocking agent traffic. OpenAI plans to add spending limits and category rules so agents can buy with less hands on approval later.
The same week, The Verge reported a study by IP protection firm Podqi. Across 6,795 answers about 151 brands, ChatGPT, Gemini and Perplexity all surfaced risky sites: fakes, knockoffs, reregistered old brand domains or dead domains. Perplexity did so in 15% of answers, ChatGPT in 8% and Gemini in 4%. Smaller brands were hit about twice as often as larger ones.
My take
Put those two stories side by side and you get the real design problem for any agent that spends money. The Wallet approval screen confirms the item, the price and the card. It does not confirm that the store is the real one.
If you build purchasing or procurement agents for clients, I would add three things before any approval step:
- An allowlist of verified vendor domains, checked in code, not by the model.
- A log of every URL the agent visited on the way to checkout, attached to the approval request.
- A hard spend cap per run that lives outside the agent, even when the platform adds its own limits.
For merchants, the playbook is the useful part. Agent traffic is going to show up as bots in your firewall. Decide now whether you want those sales, and make sure your product data is clean enough that an agent lands on you instead of a lookalike domain.
Approval screens are a safety net. Verified sources are the actual control.
More posts
- Open OCR models now read 1,000 pages for under $0.10, and better serving saved more than switching modelsNov 20, 2026
- Gemini 4 Neon costs $3 in and $15 out per million tokens and tops Zapier's AutomationBench at 56.3%, though it uses more output tokens per taskNov 20, 2026
- Gemini API's Tool Retrieval loads tool schemas only when the model asks for them, so agents with hundreds of tools send fewer input tokensNov 20, 2026
