Think Before You Prompt

The Pre-Prompt Worksheet

Every good prompt starts before the prompt. It starts with four questions that force you to think about what you actually need, what success looks like, what could go wrong, and what the AI needs to know. This takes three minutes. It will save you thirty.

These are not brainstorming questions. They are forcing functions. Each one eliminates an entire category of bad output before it happens.

Question 1: What is the specific outcome I need?

"Content" is not an outcome. "Blog post ideas" is not an outcome. "Help with my campaign" is definitely not an outcome. A specific outcome has a format, a purpose, and a destination.

Bad: "I need social media content." Good: "I need 5 LinkedIn post drafts, each under 200 words, promoting our Q2 customer research report to VP-level marketing leaders."

The difference is that the second version tells you exactly what done looks like. You could hand it to someone and they would know whether they succeeded. The first version is a wish. The second version is a brief.

When you cannot articulate the specific outcome, it usually means you have not done the strategic work yet. That is not a prompt problem -- that is a planning problem. Go solve that first.

Question 2: What does success look like?

This is different from the outcome. The outcome is what you get. Success criteria is how you judge it.

Think about it this way: if two people each wrote you 5 LinkedIn posts matching your outcome description, how would you decide which set is better? The answer to that question is your success criteria.

Example: "Success means each post leads with a surprising data point from the report, uses a conversational tone that sounds like our CEO's voice (direct, slightly contrarian, no jargon), and ends with a question that invites discussion rather than a generic CTA."

Now you have a rubric. And here is the practical benefit: these success criteria become instructions in your prompt. You are literally writing your prompt by answering this question.

Question 3: What are the failure modes?

This is the question most people skip. It is also the most valuable one.

Failure modes are not the opposite of success criteria. They are the specific, predictable ways AI will mess up this particular task. You can often predict these from experience.

Example failure modes for the LinkedIn post task:

• Posts sound like generic corporate marketing instead of a real person
• AI invents statistics that are not in our actual report
• Posts are too long for LinkedIn and will get cut off
• Every post starts with "Did you know..." or some other cliche hook
• The tone is too promotional -- reads like an ad instead of thought leadership

Each of these becomes a "Do NOT" instruction in your prompt. You are building your prompt's guardrails by thinking about what goes wrong.

Question 4: What constraints does AI need to know?

Constraints are the facts about your situation that AI cannot infer. Your brand voice. Your compliance requirements. Your audience's sophistication level. Your budget. Your timeline. The platform limitations. The political dynamics at your company that mean certain topics are off-limits.

Example constraints for the LinkedIn post task:

• Our CEO never uses emojis on LinkedIn
• The report is gated behind a landing page at example.com/q2-research
• We cannot mention competitor names (legal approved this policy last quarter)
• Our audience is senior -- they have seen every marketing tactic, so nothing should feel like a "growth hack"
• Posts will go out Tuesday through Thursday, 8am EST

These are things you know that AI does not. Every constraint you fail to mention is a place where AI will guess. And AI's guesses are based on internet averages, not your specific situation.

A Completed Worksheet

Here is what the worksheet looks like filled out for a real marketing scenario -- launching a new feature for a B2B SaaS product:

PRE-PROMPT WORKSHEET
====================

Task: Product launch announcement for new AI-powered reporting dashboard

1. SPECIFIC OUTCOME
Three-part email sequence (announcement, feature deep-dive, customer story)
targeting existing customers on our Growth plan ($299/mo). Goal: drive 40%
of Growth customers to activate the new dashboard within 14 days of launch.

2. SUCCESS CRITERIA
- Each email is under 200 words (our analytics show Growth users skim)
- Feature benefits are tied to specific pain points we hear in support tickets:
"I spend 2 hours building weekly reports" and "My CEO wants data I can't pull"
- CTAs go directly to the dashboard (not a landing page, not a blog post)
- Sequence has logical progression: awareness → understanding → social proof

3. FAILURE MODES
- Emails sound like a product changelog instead of a benefit-driven narrative
- AI invents customer quotes or case study details we haven't provided
- Generic SaaS language: "streamline," "leverage," "unlock the power of"
- Emails are too long -- our Growth users are operators, not readers
- No urgency mechanism (we need activation in 14 days, not "whenever")

4. CONSTRAINTS
- Brand voice: confident, concise, slightly informal (we use contractions,
never use "Dear [Name]")
- Must mention that the dashboard is included in their current plan at no
extra cost (this is the #1 question support expects)
- Customer story in email 3 must use the anonymized case study from Acme Corp
(real metrics: 67% reduction in reporting time)
- Cannot reference competitor products by name
- Subject lines must be under 45 characters (our inbox testing data)

Notice what happened. By answering four questions, you have essentially written your prompt. The worksheet IS the prompt -- you just need to format it. Every answer becomes an instruction, a constraint, or a success criterion that the AI can act on.

This is the shift: from "hoping AI figures it out" to "telling AI exactly what you need." The worksheet is not extra work before prompting. It IS the work of prompting.