Template Deconstruction Framework
What You Get
A 4-step system to take any template (proposal, email brief, contract scope, design feedback) and turn it into something AI can fill accurately and consistently. You identify every blank, map the data needed for each blank, design targeted prompts, and verify the output.
This is from Hazel’s AI coaching curriculum. It’s the difference between “AI that fills blanks randomly” and “AI that understands context.”
The Prompt
I have a template that I'm filling manually. I want to turn it into
an AI-powered workflow. Help me deconstruct it and create targeted
prompts for filling it.
Here's the template:
[Paste the template. Include all blanks, placeholders, sections,
variables—everything that needs to be filled.]
---
BUSINESS CONTEXT:
- What is this template for? [e.g., client proposals, intake forms, etc.]
- Who uses it? [e.g., you, your sales team, etc.]
- How often is it used? [e.g., 5x per week, monthly, etc.]
- What goes wrong when filled manually? [e.g., inconsistency, missing info, etc.]
DATA YOU HAVE ACCESS TO:
[List the sources: CRM, spreadsheet, client questionnaire, past docs, etc.]
DATA YOU DON'T HAVE:
[What info is missing? What do you need to ask for or research?]
---
OUTPUT: Create a TEMPLATE DECONSTRUCTION with 4 sections:
SECTION 1: BLANK INVENTORY
List EVERY blank and variable in the template. Categorize them:
- Static (same every time): [blanks]
- Conditional (depends on X): [blanks]
- Data-driven (pulls from database): [blanks]
- Research-required (needs you to look it up): [blanks]
- Open-ended (requires thinking/writing): [blanks]
SECTION 2: DATA MAPPING
For each blank, answer:
- What data does this need?
- Where does that data come from?
- Is it available right now, or do we need to gather it?
- Is this data consistent (A or B), or does it need customization?
SECTION 3: PROMPT ENGINEERING
Write a TARGETED PROMPT for filling each section of the template.
These should be specific. Not: "Fill this template"
Instead: "Based on [data], write a [type of content] that [specific outcome]"
SECTION 4: VERIFICATION CHECKLIST
What should you check in the AI output before using it?
- Accuracy tests (did it use the right data?)
- Quality gates (does it match your brand voice?)
- Completeness tests (did it hit all the required elements?)
- Context checks (does it make sense for this specific situation?)How to Use It
Step 1: Find a template you fill regularly (at least 3-5 times per month). Paste it into Claude with the prompt above. Be honest about what goes wrong.
Step 2: Claude breaks down the template into sections and identifies every blank. It shows you which blanks are simple (pull from CRM) and which are complex (require thinking).
Step 3: Claude writes targeted prompts for each section. You test these prompts with Claude or your preferred AI tool. Refine them until the output is good enough to use.
Step 4: Build the workflow. You gather the required data (client info, context, etc.), feed it into Claude with the targeted prompts, and get a filled template. You verify it against the checklist.
Step 5: Iterate. After 5-10 runs, you’ll see where the AI consistently misses or hallucinate. Update the prompts. Tighten the data requirements.
Example Output
# Template Deconstruction — [Your Template Name]
---
## SECTION 1: BLANK INVENTORY
**Static Blanks** (same every time):
- Your company name
- Your contact information
- Your company's service description
**Conditional Blanks** (depends on project type):
- Project scope (varies by scope)
- Timeline (varies by complexity)
- Pricing structure (varies by service tier)
**Data-Driven Blanks** (pulls from database):
- Client name
- Client email
- Client industry
- Project start date
- Budget amount
**Research-Required Blanks** (you have to find it):
- Competitor analysis for this client
- Industry trends relevant to their business
- Market opportunity size
**Open-Ended Blanks** (requires thinking/writing):
- Why this approach is best for them (customized reasoning)
- Success metrics specific to their goals
- Risk mitigation strategies
---
## SECTION 2: DATA MAPPING
| Blank | Data Needed | Source | Available? | Notes |
|-------|------------|--------|-----------|-------|
| Client name | Client's legal business name | Intake form | Yes | Always filled |
| Project scope | What they're asking us to build | Sales call notes + written brief | Sometimes | Need to standardize |
| Budget | What they're willing to spend | Sales conversation or proposal request | Sometimes | Sometimes implied, need to confirm |
| Timeline | When they need it done | Project brief or sales notes | Sometimes | Often vague—need to clarify |
| Competitors | Who they compete with | Industry research + their own mention | Sometimes | We need to research if not mentioned |
| Success metrics | How they'll measure success | Discovery call or their stated goals | Sometimes | Need a discovery call to define |
| Why this approach | Why we recommend what we recommend | Our methodology + their specific needs | Always available | We know this, just need to contextualize |
**Data Gaps Identified:**
- Timeline is often vague. Need to add a "Timeline Clarification" step before filling the proposal.
- Competitors sometimes aren't mentioned. Need a research step.
- Success metrics need a discovery call to define. Can't assume.
---
## SECTION 3: PROMPT ENGINEERING
### Prompt 1: Executive Summary
You are a proposal writer. Based on the following client info and our analysis, write a 2-paragraph Executive Summary.
CLIENT INFO:
- Company: [name]
- Industry: [industry]
- Challenge: [what they told us]
- Their goal: [what they want to achieve]
OUR APPROACH: [High-level description of how we’d solve this]
INSTRUCTIONS:
- Paragraph 1: Restate the challenge in our words, showing we understand them deeply
- Paragraph 2: Introduce our approach and why it’s right for them
- Use their industry language where appropriate
- Make them feel understood and confident
- Avoid jargon; be clear and direct
- 150-200 words total
TONE: Professional but warm. Confident but not arrogant.
### Prompt 2: Scope of Work
You are a scope writer for service proposals. Based on the project brief, write a detailed Scope of Work that’s specific and unambiguous.
PROJECT DETAILS:
- Service: [what we’re doing]
- Timeline: [when it needs to be done]
- Key deliverables: [what they get]
- Client’s constraints: [any limitations we know about]
DELIVERABLES TO INCLUDE:
- [Deliverable 1 with specifics]
- [Deliverable 2 with specifics]
- [Deliverable 3 with specifics]
INSTRUCTIONS:
- Use numbered or bulleted format
- Be specific: “3 rounds of revisions” not “revisions as needed”
- Include what’s NOT included (scope boundaries)
- Make it clear what triggers phase 2 (if applicable)
- Use plain language, not jargon
- Each deliverable should take 1-2 sentences
TONE: Clear and professional. This is a contract document.
### Prompt 3: Pricing Rationale
You are a proposal writer explaining pricing decisions to a client. Based on their project and our pricing model, write a Pricing Rationale section that makes them feel like the investment is justified.
PROJECT SCOPE:
- What we’re building: [description]
- Timeline: [length]
- Complexity: [simple/moderate/complex]
- Their budget: [what they said they’d spend]
OUR PRICING:
- Total project cost: [amount]
- Why it costs this: [brief explanation of what drives cost]
CLIENT CONTEXT:
- Industry: [industry]
- Size: [company size]
- Budget constraints: [any limitations]
INSTRUCTIONS:
- Explain the pricing without apology
- Connect the cost to the value and outcomes
- If we’re at or above budget, explain why it’s worth it
- If we’re below budget, you can hint at why without sounding cheap
- Use outcomes (what they’ll get) not just time/effort
- 150-200 words
TONE: Confident and educational. You’re teaching them why this is a good investment, not pleading.
### Prompt 4: Next Steps & Timeline
You are a proposal writer. Based on the project, write a clear Next Steps section and timeline.
WHAT WE’LL DO: [Project phases and timeline]
SUCCESS CRITERIA: [What good looks like]
INSTRUCTIONS:
- Use a timeline format (Month 1: X, Month 2: Y)
- Include key decision points for the client
- Include deliverable dates
- Be realistic. Don’t over-promise on speed.
- Close with a call-to-action: what do we need from them to move forward?
- Keep it scannable (short, clear, bold for key dates)
TONE: Clear and organized. This builds confidence that we know what we’re doing.
---
## SECTION 4: VERIFICATION CHECKLIST
Before you send the proposal to the client, check:
**Data Accuracy:**
- [ ] Client name spelled correctly
- [ ] Company name correct
- [ ] Industry correctly identified
- [ ] Timeline matches what they said
- [ ] Budget aligns with their stated amount (or we explained the difference)
**Quality Gates:**
- [ ] Executive Summary shows deep understanding of their challenge
- [ ] Scope is specific (not vague like "as needed")
- [ ] Pricing is presented confidently (not apologetically)
- [ ] Language matches our brand voice (not corporate-sounding)
- [ ] It reads like *one* document, not four separate templates stitched together
**Completeness:**
- [ ] All sections are filled (no [BLANK] placeholders)
- [ ] All required elements are present
- [ ] No placeholder text remains
- [ ] Numbers and dates are formatted consistently
**Context & Logic:**
- [ ] Timeline is realistic for the scope
- [ ] Pricing aligns with scope (not over/under)
- [ ] Scope matches what they asked for (or we explained why we're recommending differently)
- [ ] Next steps are clear and actionable
- [ ] Call-to-action is obvious (what do we need from them?)
**Brand & Voice:**
- [ ] Tone is consistent throughout
- [ ] No jargon or buzzwords
- [ ] No typos or grammatical errors
- [ ] Formatting is clean and scannable
**Red Flags to Catch:**
- [ ] AI apologizing ("I understand this is expensive, but...")
- [ ] AI being vague ("revisions as needed" instead of "3 rounds")
- [ ] AI contradicting itself (timeline says 4 weeks but scope says 8)
- [ ] AI using wrong client name or company details
- [ ] AI tone shifting mid-document
If any of these fail, go back and refine the specific prompt, not the whole thing.
---
## HOW THIS SCALES
**For you solo:** You run client info through the 4 prompts, verify against
the checklist, and send. Time savings: from 2 hours to 20 minutes.
**For your team:** You document the 4 prompts as standard procedure. New
team members follow the prompts and checklist. Everyone produces consistent
output.
**For different template types:** You build these deconstructions for
every recurring template: intake forms, scope documents, contracts,
emails, briefs, etc. Each one gets its own set of targeted prompts.Tips
One: The best prompts are specific and contextual. “Write a summary” is useless. “Based on [their challenge], write a 2-paragraph summary that shows you understand their specific situation” is useful.
Two: Your checklist is more important than the prompts. A great prompt followed by no verification will still produce garbage. Verification catches the AI’s mistakes.
Three: Start with one template. Don’t try to deconstruct everything at once. Master one, then move to the next.
Four: The “Data Mapping” section will show you exactly where your process is broken. If you don’t have timeline info, no prompt will make up for it. Fix the data flow first.
Five: Iterate based on what the AI gets wrong. After 5 runs, you’ll see patterns. “The AI always forgets to mention ROI.” Add that to the prompt. “The AI tone is too formal.” Add tone guidance.
Six: Share the prompts with anyone who fills the template. They don’t need to understand the theory—they just follow the checklist.
Want more methods? Browse the full library at hazelq.com/methods.
Built with Claude. Every prompt in this library has been tested in the latest Claude model.