With the release of GPT-5.5 at the end of April, OpenAI also updated its recommendations for effective prompting. The model is designed more strongly as an autonomous AI agent that can plan tasks and work through them over multiple steps. OpenAI therefore recommends a different prompting approach.
Rather than reusing older prompts that often prescribe the process in great detail, users are advised to start with concise, goal-oriented instructions. According to OpenAI, shorter prompts focused on the desired outcome tend to work better than complex prompt stacks developed for earlier model generations. The basic idea is to define the outcome, success criteria, constraints and relevant context, while largely leaving the route to the solution up to the model. OpenAI also notes that overly detailed instructions can actually reduce efficiency. It further recommends beginning prompts with a clear role description and sufficient context.
For factual tasks, OpenAI advises specifying source requirements explicitly in the prompt: which claims need evidence, what counts as valid evidence and how missing evidence should be handled. For creative or hybrid tasks such as presentations or marketing copy, it should also be made clear which parts need to be source-based and which may be written more freely.
By Ida Kandler
Source: The Decoder, 26 April 2026, by Matthias Bastian