The framework for getting what you actually want from AI
This prompt guide is based on Stop settling for "this isn't what I wanted" from AI, featuring Pam Didner, published on the AI Lab by ActiveCampaign.
Get the quick-start guide
The big idea
“Good prompting is re-prompting.” That’s from Pam Didner, who has led 16+ AI trainings this year for companies like Southwest Airlines, Intel, and the U.S. Army. The first output is never the answer — it’s the start of a conversation. Treat AI like a junior assistant you’re briefing: set the objective, give context, define success, share references, then review and iterate.
The framework
Four mistakes derail most prompts. Each has a fix.
| Mistake | What it looks like | The fix |
|---|---|---|
| Being too general | “Write a 1,000-word blog about cybersecurity” | Add audience, brand voice, format, and examples. Treat the AI like a junior hire who needs context. |
| Treating prompts as one-and-done | Accepting the first draft, or abandoning the tool when it’s off | Re-prompt. Say the same thing differently. Add missing context. The first output is v1, not the answer. |
| Over-relying on copy-paste | The same prompt used for LinkedIn, TikTok, and email | Adapt for the platform. A prompt that produces good LinkedIn copy will flop on TikTok. Rewrite for the format. |
| Forgetting to fact-check | Publishing without review because the output sounded confident | Review every output. AI hallucinates. Build a review step into every workflow that touches real customers. |
The re-prompting triage
When the output is off, run this three-step loop before giving up on the task:
1. Evaluate your prompt
Was it specific enough? Did you share audience, examples, or constraints? Generic in, generic out.
2. Re-prompt
Say it differently. Add the missing context. Swap in examples. Rename variables. Small changes often shift the output dramatically.
3. Go back and forth
Talk to AI like a teammate. Explain what you don’t like. Tell it to ask you clarifying questions. For complex tasks, break the problem into sub-questions and stitch the answers back together.
When to use this
- A campaign output sounds generic or off-brand and you’re about to abandon the tool
- You’re briefing a team on how to prompt AI consistently (especially new hires)
- You’re rolling out a new AI workflow and need a shared standard for what “good enough” looks like before publishing
Common mistakes
- Accepting the first output → re-prompt at least once before you act on what the model returns
- Vague audience briefs → name the persona, format, and brand voice up front
- Reusing the same prompt across platforms → adapt for each channel’s tone
- Skipping the fact-check → review every claim before anything ships
Quick-start
Pick one AI workflow you already run (campaign ideation, persona drafting, homepage audits). Rewrite your current prompt with audience, context, and format explicitly stated. Run it, then ask the model two follow-ups: one to refine, one to push for “novice” ideas. That’s the loop — specific brief, review, re-prompt.
Related
More data from the AI Lab.