You asked AI a question and got garbage back. The problem probably wasn't the AI—it was the prompt.
The difference between useful AI output and useless AI output is almost always how you asked.
Why Prompts Matter
AI doesn't read minds. It responds to exactly what you give it.
Vague prompt: "Help with this spec" AI response: Generic, unhelpful information
Clear prompt: "Extract all submittal requirements from this HVAC specification section, organized by equipment type, with spec paragraph references" AI response: Exactly what you need
The quality of your output is directly proportional to the quality of your input.
The Basic Prompt Structure
Good prompts have three parts:
1. Context
Tell the AI what it's working with and who you are.
I'm an MEP estimator reviewing bid documents for a hospital project.
2. Task
Tell the AI exactly what you want it to do.
Review the attached specification and identify all requirements
that exceed standard commercial construction practice.
3. Format
Tell the AI how you want the output structured.
List each requirement with:
- Spec section reference
- Requirement description
- Why it exceeds standard practice
Complete Example
Context: I'm an MEP estimator reviewing bid documents for a hospital project.
Task: Review the attached specification Section 23 73 00 (Air Handling Units)
and identify all requirements that exceed standard commercial construction practice.
Format: List each requirement with:
- Spec section reference
- Requirement description
- Why it exceeds standard practice
- Estimated cost impact (high/medium/low)
Focus on: equipment features, testing requirements, documentation requirements,
and installation conditions that would add cost beyond a typical office building.
Prompt Patterns That Work
Pattern 1: Role Assignment
Tell AI to act as a specific role.
Act as an experienced MEP project manager reviewing this RFI response.
Identify any ambiguities that could lead to disputes later.
Why it works: Focuses the AI on relevant considerations for that role.
Pattern 2: Step-by-Step Instructions
Break complex tasks into steps.
Analyze this contract clause by:
1. First, explain what the clause means in plain English
2. Then, identify the obligations it creates for the subcontractor
3. Finally, flag any unusual terms compared to standard AIA subcontracts
Why it works: Prevents AI from skipping parts of the task.
Pattern 3: Examples
Show the AI what you want.
Extract equipment from this schedule in this format:
Example:
- Equipment: AHU-1
- Capacity: 10,000 CFM
- Location: Mechanical Room 101
- Special requirements: Seismic certification required
Now extract all equipment from the attached schedule using this format.
Why it works: Demonstrations are clearer than descriptions.
Pattern 4: Constraints
Tell the AI what NOT to do.
Summarize this meeting transcript.
Constraints:
- Do not include general discussion, only decisions and action items
- Do not exceed 500 words
- Do not add information not in the transcript
Why it works: Prevents common AI tendencies to over-include or embellish.
Pattern 5: Verification Request
Ask AI to check its own work.
After extracting the requirements, review your list and verify
that each item includes the spec paragraph reference.
Why it works: Catches errors before you do.
Construction-Specific Prompts
For Bid Review
Review this ITB document and extract:
1. Bid due date and time
2. Pre-bid meeting date (if any)
3. RFI deadline
4. Bonding requirements
5. Insurance requirements
6. Any unusual bid conditions
Flag anything that deviates from standard practices.
For Specification Analysis
I'm reviewing Division 23 specifications for an HVAC bid.
For each specification section, identify:
- Submittal requirements
- Testing requirements
- Commissioning requirements
- Warranty requirements
- Any owner-furnished equipment
Organize by spec section number.
For Contract Review
Review this subcontract and identify clauses related to:
- Payment terms and retainage
- Change order procedures
- Delay claims and liquidated damages
- Indemnification
- Insurance requirements
For each clause, note:
- Location in document
- Key terms
- Anything that deviates from standard AIA A401
For RFI Drafting
Help me draft an RFI about a conflict between the architectural
and mechanical drawings.
Issue: The ceiling height shown on A-201 (9'-0") conflicts with
the duct routing shown on M-301 (requires 10'-6" minimum).
Draft an RFI that:
- Clearly describes the conflict
- References both drawings
- Proposes a solution
- Requests clarification
Keep it professional and concise.
For Change Order Narrative
Help me write a change order narrative for additional work.
Facts:
- Original scope: Install 20 VAV boxes per drawing M-401
- Changed scope: RFI-023 response added 5 additional VAV boxes
- Additional cost: $12,500
Write a narrative that:
- Establishes entitlement (why we're owed additional compensation)
- Describes the changed condition
- Justifies the cost
- References the relevant documents
Keep it factual and professional.
Common Prompt Mistakes
Mistake 1: Too Vague
Bad: "Review this document" Good: "Review this specification and extract all testing requirements"
Mistake 2: No Context
Bad: "What does this clause mean?" Good: "I'm an electrical subcontractor. What obligations does this indemnification clause create for me?"
Mistake 3: No Format Specified
Bad: "List the submittal requirements" Good: "List the submittal requirements in a table with columns for: Item, Spec Section, Type (product data/shop drawing/sample), and Lead Time"
Mistake 4: Asking for Opinions
Bad: "Should I bid this job?" Good: "List the risk factors in this bid package that I should consider in my bid/no-bid decision"
Mistake 5: Single Giant Prompt
Bad: One 500-word prompt asking for everything at once Good: Break into smaller, focused prompts and build on each response
Iterating on Prompts
Your first prompt rarely gets perfect results. Iterate:
Step 1: Start Simple
Begin with a basic prompt and see what you get.
Step 2: Identify Gaps
What's missing? What's wrong? What's unclear?
Step 3: Add Specificity
Refine your prompt to address the gaps.
Step 4: Test Again
Run the improved prompt.
Step 5: Save What Works
Keep a library of prompts that work for common tasks.
Building a Prompt Library
Create templates for tasks you do repeatedly:
# Submittal Register Extraction
Context: Extracting submittal requirements from project specifications.
Prompt:
Review the attached specification sections and extract all
submittal requirements. For each submittal, provide:
- Submittal number (assign sequentially)
- Description
- Spec section reference
- Type (product data, shop drawing, sample, etc.)
- Estimated lead time based on equipment type
Organize by specification section.
Save these templates and reuse them across projects.
What's Next
Good prompts get good outputs. The next skill is learning how to verify AI outputs—because even perfect prompts can produce imperfect results.
TL;DR
- Good prompts have three parts: context, task, and format
- Use patterns: role assignment, step-by-step, examples, constraints
- Be specific—vague prompts get vague answers
- Iterate: start simple, identify gaps, add specificity
- Build a library of prompts for common tasks