Imagine opening a support ticket because you were charged twice for a game. Before resolving the problem, customer support needs to work out where to send the complaint. A friendly four-paragraph explanation might accompany the conversation. To organize the queue, however, a single word already helps: “billing.”
That is the kind of task OpenAI is targeting with the Decisions API, which became generally available on October 22, 2026, following a public beta, according to the company’s changelog. The tool is aimed at developers. The announcement therefore concerns a way to build services, rather than presenting a new button for people chatting with ChatGPT.
An API is a connection that lets an application use another system’s service. In this case, the program sends information and questions to the AI and receives answers in constrained formats: one option from a set, an estimated probability, or a score based on defined criteria. The official documentation says the available model is gpt-6-luna.
Returning to our hypothetical example, the support system’s developer could set up the categories “billing,” “account” and “technical issue.” The AI would read the request and select one. The application would use that classification to route it. The company could also request an urgency assessment, provided it explained what it considered urgent.
This helps define the tool’s scope. Selecting a category does not authorize the AI to refund a purchase, block an account or invent procedures. Those actions depend on how the service was built and the permissions it was given. The options and criteria remain choices made by whoever designed the support system.
For the person using the service, the interesting benefit would be spending less time in the wrong queue. Developers gain an answer the program already knows how to use, without having to search for a conclusion buried in a text. It is a discreet application, but a potentially useful one for repetitive tasks.
The caution begins when a classification looks more certain than it is. Choosing quickly does not mean choosing correctly. A customer might mention a charge while trying to recover a hacked account. A category that is perfectly valid for the system can still send the person to the wrong team.
That is why the most important work takes place before and after the answer: defining good options, testing ambiguous cases and establishing when someone needs to review the decision. An option such as “refer for human review” may be more valuable than forcing every request into three boxes.
The Decisions API stands out by highlighting this small task that is often hidden behind a conversation with AI. The real benefit will appear if the complaint reaches someone who can solve it sooner. Until then, a neatly filled-in label is still only the start of customer support.
