5 Ways to Actually Get AI to Do What You Want
5 Ways to Actually Get AI to Do What You Want
by Mohit Panchal
You ask AI to write something. It comes back okay-ish. You clean it up, use it, move on. Then the next time you try the same thing, the output is noticeably worse and you're not sure why.
That inconsistency is frustrating, and it's the most common complaint from people who use AI regularly but haven't built a real system around it. One day it feels like a superpower. Next it feels like a time sink.
Here's what's actually happening: AI doesn't have good days and bad days. The results you get are almost entirely a function of how clearly you set up the task. Which means you have more control over the output than you probably think.
These are 5 prompting techniques that consistently improve AI output quality, regardless of which tool you're using. No technical background required. Each one takes about 10 seconds to apply and makes a measurable difference in what comes back.
First drafts that need less cleanup
Outputs that match the right tone, format, and audience from the start
Less back-and-forth with the tool to get something usable
More consistent results across different tasks and team members
Tip 1: Tell it who it's talking to
AI doesn't know who it's writing for unless you say so. Without that context, it guesses, and the guess shows up in the output as the wrong tone, the wrong reading level, or assumptions that don't fit your audience.
Before you describe the task, describe who's on the receiving end.
Before:
"Write a follow-up email to a client."
After:
"Write a follow-up email to a skeptical operations manager at a mid-size construction company. He's practical, doesn't like fluff, and makes decisions based on ROI."
The second prompt comes back with a noticeably different email. Shorter, more direct, less salesy. That's the version you actually send.
Tip 2: Give it a role, not just a request
Assigning AI a role or perspective changes how it approaches the task. It's the difference between asking a search engine and briefing a colleague.
You'd naturally tell a colleague who they are in a given context: "You're a project manager reviewing this for a client handoff." That framing shapes what they pay attention to and how they write it up.
Same principle applies to AI.
Before:
"Summarize this meeting."
After:
"Act as an experienced operations consultant. Summarize this meeting for a VP who wasn't there, focusing on decisions made and next steps."
The second prompt comes back sharper, more structured, and written for the right reader.
Tip 3: Specify the format before you ask for the content
AI will default to whatever format feels most natural given your prompt. Sometimes that works. Often it doesn't. If you care how the output looks (and most of the time you do), say so upfront.
Before:
"Give me a project update."
After:
"Give me a project update as 4 bullet points, each under 25 words. Lead with status, then flag anything at risk."
Specifying format isn't micromanaging. It's one of the fastest ways to reduce back-and-forth. When the structure is right on the first pass, you spend less time editing and more time using the output.
Tip 4: Show it an example of what good looks like
If you have a previous output you liked, paste it in. Tell the model: "This is the style and format I want. Write it like this." You can also describe what you don't want. Both work.
Before:
"Write a bio for our website."
After:
"Write a bio for our website. Here's one we've used before that got the tone right: [example]. Match this style. Avoid corporate language and anything that sounds like a LinkedIn summary."
What you're doing is narrowing the range of possible outputs. The more reference points you give, the less the model has to guess, and the closer the first draft is to what you actually needed.
Tip 5: Use the 10-80-10 approach for anything that matters
Think of AI-assisted work as three stages, not one.
The first 10% is yours: define the goal, the audience, the constraints, and what success looks like. The middle 80% is where AI does the heavy lifting. Draft, generate, summarize, structure. Let it go. The final 10% is yours again: review with judgment, catch what's off, and make it yours.
Before:
Paste in a vague request, use whatever comes back, skip the review.
After:
Set up the task clearly (10%), let AI draft it (80%), then review it before it goes anywhere (10%).
People who skip the first 10% get inconsistent outputs and blame the tool. People who skip the last 10% send out work they haven't actually reviewed. Both are avoidable, and both are more common than they should be.
Summary
The variable that changes AI output quality is almost always the input. These 5 techniques work in ChatGPT, Claude, Copilot, or whatever tool you're already using. None of them require technical knowledge. They just require a bit more intention before you hit send.
Specify the audience. Assign a role. Set the format. Give an example. Stay in the loop at both ends.
Do those things consistently and you'll get noticeably better results with a lot less frustration.
Want your whole team getting results like this, not just the one person who figured it out on their own?
That's exactly what we build at Champion AI: the internal capability to use AI well, consistently, and in ways that actually move your business forward.
