Prompting Guide for Project Assist

Background

Project Assist is designed to reason across your entire project document set — connecting information scattered across contracts, subcontracts, specs, addenda, and change orders to surface risks, obligations, and inconsistencies in one place. Because it works across multiple related documents simultaneously, the quality of your answers depends heavily on how your project is set up and how your questions are framed. This guide covers best practices for loading your project with the right documents and writing prompts that get you the most accurate, complete, and actionable results.

Things to Consider

• Project Assist works within a single project — it cannot query across multiple projects at once

• The more complete your document set, the more connected and reliable your insights will be

• CrunchAI may pull from any document in the project, so outdated drafts alongside executed versions can produce answers that don't reflect your actual governing terms

• Focused, topic-specific questions return stronger answers than broad, open-ended ones

• You can deselect specific documents before running a query to narrow scope and speed up results

• You can request a specific output format (table, matrix, bulleted list, short summary) and Project Assist will structure its answer accordingly

• Responses on large document sets may take longer — this is expected as CrunchAI reasons thoroughly across all uploaded files

Setting Up Your Project for Best Results

• Upload every document relevant to the project — bid packages, addenda, executed contracts, amendments, specs, change orders, and upstream/downstream contracts

• Replace draft documents with executed versions as they become available to ensure CrunchAI reasons from what's actually governing

  Note: Keeping outdated drafts alongside final versions can cause Project Assist to pull from superseded terms. Remove or replace drafts as the project progresses.

• Add new documents as they arrive — upload incoming change orders, amendments, or addenda so Project Assist always has the latest picture of the project

  Tip! If you're a subcontractor with access to the prime contract, upload it alongside your subcontract so Project Assist can cross-reference obligations between them.

Writing Effective Prompts

• Ask focused, topic-specific questions rather than broad ones

  • Instead of: "Tell me about indemnity"

  • Try: "What are my indemnity obligations under the subcontract with Acme, and do they align with what the prime contract requires?"

• Provide enough context in your question so CrunchAI understands what you're looking for and why it matters — avoid one-word or vague prompts

• Request a specific output format when it will make the answer more useful

  Tip! Asking for a table, matrix, or bulleted list upfront will return answers structured the way you'll actually use them in a meeting or report.

• Narrow your document scope when you only need answers from a specific subset — deselect documents that aren't relevant to your query for faster, more focused results

• Iterate and refine — if an answer doesn't fully address your question, add more detail or reframe your prompt with additional context

Sample Prompts by Use Case

• To locate a specific clause or requirement across documents:

  • "What does the contract, specs, and subcontracts state about who is paying for third-party testing of structural steel?"

• To synthesize requirements scattered across documents:

  • "Review all my contract documents — is there anything that says we are responsible for fire caulking at HVAC penetrations?"

• To build a comparison matrix:

  • "Build a matrix comparing liquidated damages provisions across all subcontracts in this project."

• To catch inconsistencies in values, dates, or terms:

  • "Summarize the contract amount established in the GMP followed by the changes to the contract amount by each change order and note if there are any discrepancies that have carried through."

• To gather and structure project data:

  • "Pull together all procurement and long-lead item requirements from the specs and contract into a table with responsible party, lead time, and spec section."