Creating Accurate Prompts for Inspired Query [IQ]

How to Build Document-Grounded Prompts That Deliver Trusted Results

Overview

Inspired Query (IQ) gives electrical contractors the power to ask project-specific questions and receive structured, actionable answers — instantly.
But the quality of your results depends directly on how your prompt is written.
A well-crafted prompt ensures IQ reads only from your project documents, not from generic knowledge or external data.

This guide explains how to build accurate, document-grounded prompts — like the Feeder Schedule, Raceway Analysis, and Breaker Count prompts — that deliver precision and repeatability.

Why Prompt Structure Matters

IQ doesn’t guess. It follows instructions.
When a prompt is vague (“analyze the drawings”), IQ may summarize or rely on general construction knowledge.
When a prompt is clear (“only use the uploaded project documents”), IQ acts like a focused reviewer, extracting data exactly as it appears.

A properly structured prompt:

Eliminates hallucination (AI filling gaps with assumptions)
Improves consistency across users and projects
Makes outputs traceable to actual specification and drawing sources
Saves hours of validation time during coordination and prefab planning

Core Concepts for Building Reliable Prompts

1. Ground IQ in the Uploaded Project Documents

Every prompt must explicitly tell IQ where to look for data and where not to look.

✅ Use phrases like:

“Use only the uploaded project documents.”
“Do not reference or infer data from outside sources or web content.”
“If information is missing, write ‘Not Defined in Project Documents.’”

These phrases force IQ to stay inside your uploaded specifications, drawings, and schedules.

2. Define What You Want – and How You Want It

Be specific. Tell IQ exactly what to extract and how to organize it.

✅ Example phrases:

“Create one row per physical conduit.”
“List every breaker size and count per panel.”
“Include manufacturer and model only if listed; leave blank otherwise.”
“Organize the output into Markdown tables grouped by equipment type.”

This gives IQ a repeatable template — ensuring output that’s consistent across projects.

3. Include Completeness Instructions

LLMs often summarize. That’s not acceptable for electrical data.
Always add completeness and verification steps.

✅ Example phrases:

“Identify and list every instance, not a representative sample.”
“Re-scan all uploaded files to verify that no items were missed.”
“End output with a completeness statement confirming all project documents were reviewed.”

This transforms IQ from a “summarizer” into a procedural extractor.

4. Handle Missing or Conflicting Data Gracefully

Construction drawings aren’t always clean.
Teach IQ what to do when it encounters unclear, missing, or conflicting data.

✅ Use phrasing like:

“If values differ between sheets, list both and mark ‘Conflict – Verify in Field.’”
“If a field is not defined, write ‘Not Defined in Project Documents.’”
“Do not calculate or assume missing values.”

These instructions keep IQ outputs trustworthy and verifiable.

5. Control Formatting and Output Type

Ask for tables, not narratives, when you need structured data.
Define columns, order, and examples.

✅ Example:

“Provide a Markdown table with columns for:
Conduit ID, Feed From, Feed To, Voltage, Amperage, Conductor Size, Conduit Size, Number of Conduits, Source Reference.”

That single line gives IQ a visual blueprint for the answer you expect.

Example: How This Works in Practice

🔹 Breaker Count Prompt Summary

Goal: Count every breaker size across panel schedules — including GF and GFI breakers.
Why It Works:

Anchors IQ to panel schedule sheets only
Accounts for mirrored layouts
Normalizes “1P-20A” and “2P-20A-GFI” variations
Ends with a completeness statement

Key Accuracy Phrases Used:

“Use only uploaded project documents.”
“Identify and count every breaker.”
“Interpret both sides of the panel independently.”
“Detect GF/GFI/GFCI abbreviations in any format.”
“Re-check for mirrored schedules before output.”

6. Include Verification Statements

IQ will include a final line like:

“All uploaded project documents were reviewed; no additional breakers identified.”

That statement signals completeness and gives contractors confidence that nothing was missed.

7. Avoid Phrases That Cause Outside Research

❌ Avoid:

“According to NEC…”
“Based on standard practice…”
“Typically, in electrical design…”

These invite IQ to fill in blanks using public knowledge instead of project data.

✅ Instead, use:

“Report only what is explicitly shown in the uploaded project documents.”

This single sentence prevents IQ from ever hallucinating external data.

Putting It All Together

A strong IQ prompt follows this structure:

Objective – What IQ should do
Scope of Documents – What to reference
Interpretation Rules – How to read the content
Output Table – Columns and order
Completeness Step – How IQ confirms it found everything
IQ Directives – Hard rules (no calculations, no external data)

Example Template You Can Reuse

Objective:
Analyze uploaded electrical documents to identify [specific data goal].

Scope:
Use only uploaded drawings, schedules, and specs. Do not reference external data.

Instructions:
1. Identify every [item type] mentioned in documents.
2. List each item as an individual row.
3. For missing info, write “Not Defined in Project Documents.”
4. For conflicting data, mark “Verify in Field.”
5. Re-scan documents to confirm all instances are captured.

Output Format:
Provide a Markdown table with columns for:
[define columns here]

Completeness:
End output with: “All uploaded project documents reviewed; no additional [items] identified.”

Final Takeaway

When writing IQ prompts, think like a field-proven estimator talking to an intern:

Tell it exactly where to look
Tell it exactly how to report
Tell it what not to assume

Following these steps will produce consistent, accurate, document-grounded results that your teams can trust — every time.

✅ Key Phrases to Remember

Purpose	Phrase to Include
Data Source	“Use only uploaded project documents.”
No Web Content	“Do not reference or infer data from external sources.”
Missing Data	“If missing, write ‘Not Defined in Project Documents.’”
Conflicts	“If data differs between sheets, list both and mark ‘Verify in Field.’”
Completeness	“Identify and list every instance; re-scan before finalizing.”
Output Control	“Provide a Markdown table with columns for…”