You’ve seen what a good prompt contains. This is the flip side: the habits that tend to produce bad ones.
The good news about prompting mistakes is that they’re not exotic. It’s the same handful over and over, and once you can spot them in your own drafts, you start catching yourself mid-sentence. Here are the ones that come up most, each with the fix.
Assuming 3B knows your context
The task is so clear in your own head that you describe it the way you’d describe it to a teammate who already knows the setup. 3B doesn’t have that shared history, so it fills the gaps itself, reasonably, and not always the way you’d have chosen.
- Before: “Handle the requests like we usually do.”
- After: “When a request comes in, check the item against our approved-equipment list and flag anything that isn’t on it.”
Piling everything into one sentence
One long run-on with five asks bolted together gives 3B no clear order to follow and gives you no clean place to check the result.
- Before: “Get the requests and check them and make a list and tell the team and also save it somewhere.”
- After: Break it into ordered steps, one action each. (That’s Breaking complex tasks into steps in practice.)
Describing the how instead of the what
When you tell 3B which mechanics to use, but what you actually care about is the result, you box it in, and you often end up with something clunkier than if you’d just named the goal and let it work.
- Before: “Make a loop that goes through each row and checks a field and…”
- After: “For each request, flag the ones still waiting after five days.” Say what you want; leave the how to 3B.
Leaving out the limits
Unstated boundaries have a way of coming back as the bug you find later. If there’s a cap, a filter, or an “only when,” it’s worth saying up front.
- Before: “Summarize the requests and post them.”
- After: “Summarize requests from the last 24 hours, and post nothing if there aren’t any.”
Vague success criteria
If you never say what “done right” looks like, you can’t really judge whether the result got there, and neither can 3B.
- Before: “Make it look good.”
- After: “The summary should fit in a few lines and lead with the requester and the item.”
The thread through all five is the same one: say what you mean, in order, with the limits included. Miss one and you tend to get a build that’s close, and close is the slowest kind of wrong to fix.
You can plan and you can prompt. Next comes making it right when the first go isn’t. How to give effective feedback: steering a build back on course without starting from scratch.