The report
A business profile tells a user which merchant claims to be speaking; it does not make the merchant's prices, product recommendations or proposed transactions independently authoritative. OpenAI's November 10 Businesses in ChatGPT launch introduced verified business profiles, editable company information, product feeds and a distinct Business Agent conversation for eligible merchants. The company also published a Business Agents Policy governing the material and tools a merchant can supply.
The rollout separates identity and eligibility. A business in a supported region may be able to claim a profile, while initially available Business Agents were limited to eligible US ecommerce and online-service organizations. A verified profile does not guarantee permission to publish an agent. The agent must identify itself as AI representing the business rather than as a human, a neutral adviser or an agent endorsed by OpenAI. That disclosure matters because a customer may otherwise mistake commercial persuasion for independent guidance.
Four different approvals in one customer journey
| Decision | What it can establish | What it cannot establish |
|---|---|---|
| Business verification | The claimant is permitted to represent a merchant | Every price, guarantee or claim in the supplied catalog is accurate |
| Agent publication | The business agent is eligible under platform rules | Every prompt, file, URL and connected tool will behave safely under attack |
| Sponsored discovery | A labeled paid placement leads to the merchant's agent | That the merchant or ranking is independent of commercial incentives |
| Customer authorization | The user consents to a specific approved feature or action | Open-ended rights to collect data, submit orders or repeatedly contact the user |
The policy forbids fabricated reviews, prices, qualifications and artificial urgency, while prohibiting misleading links and requests for unnecessary sensitive information. Submitting a booking, form or purchase, or transmitting a user's information to the business, requires explicit authorization through an approved capability. Those requirements apply even when the underlying merchant has been verified. The company's initial product description says the merchant conversation is labeled and separate from the user's ordinary ChatGPT exchange, rather than quietly replacing independent answers.
A new commercial surface carries familiar tool risks
Merchant-provided catalogs, files and links are useful data inputs but remain claims supplied by an interested party. A product feed that changes after a recommendation may affect the final price, while an agent's tool can affect a booking or share contact details beyond the original informational request. The control question is not simply whether the model followed the latest merchant instruction, but whether an authenticated, current and authorized user action allowed the final effect. That distinction is shared with effect-bound approval, although the commercial-agent policy is not an independent technical demonstration of that control.
OpenAI says Business Hub analytics aggregate mentions, clicks and discovery topics and are shared only after thresholds are met; individual conversations and Business Agent chats are not disclosed to businesses through those analytics. The thresholds, verification failure rates, hostile-feed handling and end-to-end consent enforcement have not been independently tested in the announcement. This is a product and governance change with inspectable rules, not evidence of a measured fraud incident or of universal compliance by future merchant agents.