Instructions, questions, or conversational input submitted to an AI system.
A prompt is input supplied to an AI system to request or guide a response. It can be a question, instructions, examples or contextual information. In a conversation, earlier exchanges may also affect what the system receives, so the visible latest sentence is not always the entire basis of the answer.
A clear prompt explains the task and the information needed to complete it. For example, asking for a customer explanation based on an approved service list is more specific than asking for “better marketing.” Useful constraints can describe the audience, format, sources and what should not be invented. They do not guarantee that the system will follow every instruction correctly.
The input should also respect data boundaries. Do not include confidential customer material merely because it would make the task easier. Use the business’s approved tools and handling rules. A prompt is a working instruction, not a reason to bypass source checking or treat a generated answer as automatically authorized for publication.
Takeaway: State the task, relevant context and limits clearly. Review the result against those requirements rather than assuming a detailed prompt guarantees a correct answer.
Example
A repair business asks an AI tool to turn an approved inspection checklist into a plain-language preparation email. It specifies the audience and says not to add services, prices or guarantees. An editor then compares the draft with the checklist before sending it.
What to check
Keep the important facts separate from optional style preferences. If the answer must rely on supplied material, identify that material explicitly and check whether unsupported details appeared. Preserve the prompt when testing changes so the comparison has a meaningful basis.
For more detail, see OpenAI prompt engineering guidance.