Definition
The practice of writing instructions to AI models to get desired outputs. Important for power users; increasingly invisible to end users as AI systems get better at handling ambiguous requests. It covers the standing instructions behind a tool as much as anything typed into a chat box: the role, the rules, and the edge cases.
What it looks like in practice
Two dog groomers use the same AI text assistant. The first types 'reply to this customer' and gets a polite, generic message offering to help. The second has written the instructions once: you are the front desk at a grooming salon, doubles are priced per dog, we do not take aggressive dogs without a meet-and-greet, never quote a price for a matted coat without a photo, and if the owner sounds upset, hand it to Dana. Same customer message, same underlying model. The second salon's reply names the price, asks for the photo, and offers Thursday. The instructions did that, not the model.
Why it matters
This is why the same tool can produce polished results for one business and generic filler for another; the gap is usually in the instructions, not the model. For an owner, the practical takeaway is a buying signal. Writing prompts that hold up means encoding your voice, your pricing rules, and the edge cases you have learned the hard way, then testing them against real messages and fixing what breaks. That is ongoing work, closer to training a new hire than configuring software, and it is the part most owners would rather hand off than learn. Worth asking any vendor who maintains those instructions once you are live.