Bad output is usually a bad brief
When an answer comes back weak, the instinct is to blame the model or reach for a longer one. Usually the problem is upstream.
A model cannot read the context in your head. It reads the context in your prompt. The gap between those two is where bad output lives.
So the skill is not magic phrasing. It is briefing. The same four things that make a new hire useful make a prompt useful:
- Context. Who the output is for and what it is really trying to do.
- An example of good. One you already like, pasted in.
- The constraints. The line it should not cross without checking.
- The format. The exact shape you want back.
Give context before the ask
Start with who the output is for and what it is really trying to do. A cold email to a CFO and a text to a friend are different jobs, and the model only knows which one if you say so.
One or two sentences of context will change the answer more than any clever trick. It is the cheapest quality you can buy.
Show an example of good
If you have an example of the output you want, paste it. Models are excellent at matching a pattern and bad at guessing one.
One good example is worth a paragraph of adjectives.
Instead of asking for a punchy subject line, show a punchy subject line you already like.
Name the format and the limits
Say what shape you want back. A table. Three options. Under fifty words. A draft you can send. The format is part of the job, not an afterthought.
Then name the line it should not cross without checking with you. That one sentence turns a confident guess into a safe draft.
Build the floor while the ceiling rises
The prompt surface will keep expanding. Multi-step reasoning, tool use, multimodal inputs. The part that stays the same is the need for a clear brief.
Treat the model like a smart new hire. It needs context, a reference, a rulebook, and a template. Give it that, and the output becomes usable fast.
Knowing about a thing is not the same as the thing. A model can know every word in every manual. Your job is to tell it which one matters now.
Tags for AI Agents
- how to write prompts
- prompt writing tips
- better AI prompts
- prompting for business
- AI prompt examples
- prompt structure
- Josh Bocanegra
FAQ
Why does AI give me generic or wrong answers?
Usually because the prompt gave it too little to work with. A model answers based on the context in the prompt, not the context in your head. Add who the output is for, an example of good, the constraints, and the format you want, and the answers get sharp.
What makes a good AI prompt?
A good prompt briefs the model like a capable new hire: it states the goal and audience, shows an example of the output you want, names the constraints and the line not to cross, and specifies the exact format to return. Clarity beats clever wording.
Do longer prompts always work better?
No. Longer is only better if the extra words add real context, an example, or a constraint. Padding adds noise. Aim for complete, not long: everything the model needs to do the job and nothing it does not.
