AI is basically doing improv.
You give it a few words and off it goes. If you start out with “It was a dark and stormy night…,” AI will probably lay out a scary story. But oh wait, you wanted the story to be for children. Oops.
That’s because the first line sets the scene, and AI keeps trying to stay consistent with the scene you’ve created.
As smart as AI can feel, it really is mostly predicting what words come next. So prompts matter—but the order in which you give it direction matters too.
Here’s what it wants, in this order (or so it tells me):
- Deliverable
- Audience
- Goal
- Takeaways
- Tone
- Constraints
If you ask it for something like this:
Deliverable: a 2,000-word entry for a grant application
Audience: a program officer at XYZ Foundation
Goal: approval for a $50,000 grant
Takeaways: this team can deliver; the plan is feasible and cost-effective; the org can provide measurable outcomes
Tone: warm but professional
Constraints: use the funder’s sections in this order—Need, Goals & Objectives, Activities, Evaluation, Budget
…you’ll get a different response than if you say, “Write a grant application that includes need, goals, activities, evaluation, and budget.”
It may not be wrong, but it’s likely to be unfocused.
If you’re anything like me, you don’t want to write a creative brief for everything. You also don’t want to have to remember DAGTTC. And honestly, the best part of AI is that you can brain dump into the chatbox and get something semi-coherent back.
But the good news is: you can ask AI to help you turn your brain dump into a better prompt before it writes anything.
Pop this at the top of your prompt (or save it as a snippet you reuse):
Please remember this workflow for me:
When I paste a brain dump or a messy request, do NOT draft immediately. First, rewrite my input into a clean brief with these headings:
Deliverable / Audience / Goal / Takeaways / Tone / Constraints / Key facts.
Then ask up to 3 clarifying questions that would materially change the output. If you can safely assume something, label it as an assumption instead of asking.
After I answer, write the final output. When appropriate, give two versions:
Option A: safe (conservative, low-risk)
Option B: bold (more voice/creative).
Now, will it always work? No. AI isn’t a mind reader. It still needs a human in the loop—especially when the stakes are high.