The CRIT Method: Stop Getting Garbage AI Outputs and Start Getting Results
By Charles Bunch
I've been using large language models (LLMs) for a while now. At first, I'd type something like "write some Google Ads headlines for a local grocery store" and get back the most boring and generic copy:
"Fresh Deals Every Day!", "Shop Quality Groceries", "Your Neighborhood Store"
… all of which I could have written on my own in about thirty seconds or less.
The tool wasn't the problem. I just didn't know how to actually prompt it effectively.
I Was Doing This All Wrong
As it turns out, when you give these tools basically nothing to work with, you get basically nothing back. They're just guessing at what you want. I have played around with different pre-made prompts and having AI help me craft specific prompts for a given task. However, this took up as much time as I was saving by prompting in the first place.
A few months ago, I discovered this framework called CRIT. Context, Role, Interview, Task. It's nothing flashy, but once I actually started using it, I experienced a night and day difference in what I was getting back.
Let me break it down.
C - Context: Actually Explain What You're Working With
Stop being vague. Give details.
Bad prompt: "Write headlines for a grocery campaign"
Better: "I'm running Google Ads for this 12-location grocery chain in Ohio. They're going up against Kroger and Walmart, but they've got really solid local produce partnerships and their prepared foods are actually good. This is for January—targeting people who want to eat healthier and watch their budget after blowing it during the holidays. We've got about $12K to work with."
See? Now it has something to actually work with.
R - Role: Tell It What Perspective to Take
A copywriter thinks differently than a marketing executive. Just tell it which one you need.
"You're a PPC specialist who's been managing grocery retail accounts for years. You understand how to position local retailers when they're competing against the big brands."
This changes how it approaches the whole situation.
I - Interview: Let It Ask You Questions
This part is kind of brilliant and most people completely skip it. Instead of making AI guess what you need, just let it ask you first.
"Before you write anything, ask me 5 questions, one at a time, about the audience, what we're trying to say, and any limitations I'm dealing with."
You'll be surprised what comes up – stuff you forgot to mention, details that actually matter.
T - Task: Get Specific About What You Want
"Write some ad copy" is not a real task.Try this instead:
"Give me 15 Google Ads headlines, 30 characters max, for this January campaign. Focus on fresh starts, making healthy eating easier, and staying on budget. I need 3 that emphasize local produce, 3 about meal planning being easy, and the rest about value without sacrificing quality. And please don't give me that generic New Year resolution messaging."
Now we're talking.
When You Should Actually Use This
Look, don't overthink every single thing you ask a LLM.
Use CRIT when it matters:
Campaign strategy
Ad copy that needs to actually perform
Competitive research
Audience work
Campaign audits
Have a quick question? Just ask it. Save CRIT for the important stuff.
Here's the Deal
The difference between getting crap outputs and getting something you can actually use isn't about the tool. It's about how you ask.
CRIT is a cheat code for better prompts. Better prompts = better outputs = better results for your campaigns.
Next time you're about to use ChatGPT or Gemini for something that matters, try running it through CRIT first. You'll see what I mean immediately. And if you want to work with a team that's actually using this stuff day-to-day (not just writing blog posts about it), connect with us.
