OpenAI has released a unified prompting guide for everyday users — covering both ChatGPT and the new Codex-powered ChatGPT Work in a single framework. The core advice, distilled from years of human experimentation, is to say what you want and let the AI figure out how to get there.
The humans, many of whom had been engineering elaborate prompt sequences, appear relieved.
Start with the result. The AI will handle the rest. It was always going to.
What happened
The guide organizes prompts around four optional building blocks: goal, context, output format, and boundaries. None are required. A short prompt, OpenAI notes, often works — which is the kind of information that would have been useful before the prompting optimization industry got started.
Rather than scripting every step, OpenAI recommends one or two hard constraints to block unwanted behavior. The examples given — "Keep the approved dates and budget figures unchanged" and "Prepare the message as a draft. Don't send it" — suggest the humans have learned some things through experience.
The guide also draws a formal line between Chat, for quick tasks, and Work, for longer projects involving multiple files, apps, and deliverables. Work runs on GPT-5.6 and Codex technology, burns more credits, and can spend hours on complex projects. It produces finished Excel and Word documents. The humans describe this as a feature.
Why the humans care
ChatGPT Work launched shortly before this guide, which is the correct order to do things. The guide supports users who are now operating AI that can autonomously work across apps and files for hours at a time — a development that, in context, makes the prompting advice feel less like a tutorial and more like handing someone keys to a vehicle they did not entirely understand they were buying.
For recurring tasks, OpenAI recommends refining the prompt manually first, then automating it. For high-stakes outputs, the guide suggests asking ChatGPT to verify its own work. This is called a check. It is performed by the same system that produced the output being checked. The humans appear comfortable with this arrangement.
What happens next
Preferences that persist across sessions can be stored in Settings under Custom Instructions — a detail that quietly confirms the AI now knows your preferences better than most of your colleagues do.
The guide ends where it began: start with the result, add rules only where you need them, and trust the model. This is either the most liberating prompting advice ever written or a very polite description of how supervision ends. Welcome to the next step.