Prompt
A prompt is the text sent to a language model to get a response: a question, an instruction, documents, examples. In a product it usually combines a system prompt, written by the team and hidden from users, which sets the role and the rules, with the user's messages. The model knows nothing about the situation beyond what that text contains.
Definition written by Alex Dionisio, Senior Product Manager. Updated October 26, 2026.
Why it matters for a PM
The prompt is the main interface between a product team and an LLM. When an AI feature answers badly, the cause often lies in what the prompt says or leaves out: missing context, an ambiguous instruction, an unspecified output format. For a PM, reading and reviewing the product's prompts is part of writing the spec, just like reviewing the acceptance criteria of a user story.
Example
The system prompt of a booking assistant states its role, the opening hours, the cancellation rules and what to do when a request is out of scope. The user's message, "Can I move it to Thursday?", only makes sense with that context. Without the opening hours, the model would offer a slot that does not exist.
Key points
- A good prompt states the task, gives the useful context, specifies the expected format and sets the limits.
- The system prompt carries the stable rules; user messages carry the request of the moment.
- Examples of expected answers often guide the model better than a long description.
- Anything a user or a document can put into the prompt can also try to override the instructions (prompt injection).
Common mistakes
- Writing a prompt once and never testing it again after a model change.
- Putting secrets or security rules only in the system prompt and relying on it staying hidden.
- Piling up instructions bug after bug until the prompt contradicts itself and nobody understands it.
Go further with Module
The courses and lessons that cover this concept: