Skip to content

Guide

AI prompts for due diligence review: practical examples for contracts, financials and Q&A

Published November 16, 2026

Key points

  • The best diligence prompts are specific: name the document set, the clause or metric, the output format and what to do when the answer is not found.
  • Ask for citations to the exact document and page, so every answer can be checked.
  • Use AI for finding and extracting; keep judgement and conclusions with people.
  • Tell the model to say "not found" rather than guess; that one line prevents many false answers.
  • 4 of the 11 providers we review list built-in AI help; check what each one can actually do with your documents.

AI assistants built into data rooms, and general AI tools used on exported documents, can speed up due diligence. How much they help depends a lot on how you ask. A vague prompt (“summarize the contracts”) gets a vague answer. A specific prompt (“list every customer contract with a change of control clause, with document number, clause number and a one-line summary; if none, say none found”) gets something a lawyer can check in minutes.

This guide gives prompts that work for common tasks, and rules for checking the output. Before using any AI tool on deal documents, confirm that it is approved for confidential data and that your agreements allow it.

Anatomy of a good diligence prompt

Five parts of a prompt that produces checkable output

Scope Which folder or documents to search
Task What exactly to find or extract
Format Table, list or short answer
Citations Document and page for every point
Not found What to say when there is no answer

Prompts for contracts

TaskExample prompt
Change of control”In folder 4 (Customer contracts), list every agreement containing a change of control, assignment or termination-on-sale clause. Output a table: document number, counterparty, clause reference, quoted clause text, one-line effect. If none, say ‘None found’.”
Exclusivity”Identify contracts in folder 4 that grant exclusivity or most favoured customer terms. Quote the clause and cite document and page.”
Termination for convenience”Which supplier contracts in folder 5 can the supplier terminate without cause, and on how much notice? Table with citations.”
Renewal dates”Extract term, renewal mechanism and next renewal date for each lease in folder 7. Use YYYY-MM-DD. Mark unclear entries ‘check manually’.”
Liability caps”For each customer contract over 100,000 in annual value per the revenue schedule in 2.4, extract the limitation of liability clause and cap amount.”

Prompts for financial documents

TaskExample prompt
Consistency check”Compare revenue by year in the audited statements (2.1) with the management accounts (2.3). List any year where they differ by more than 1%, with both figures and sources.”
Customer concentration”Using the revenue by customer schedule (2.5), list the top 10 customers and their share of revenue for each year shown.”
One-off items”Identify items described as non-recurring, exceptional or one-off in the financial statements and management accounts. Table with amount, year, description and source.”
Debt”List all borrowings, guarantees and security interests mentioned in folders 2 and 5, with lender, amount and document reference.”
Check the numbers yourself

AI tools can misread tables, mix up years or round incorrectly. Use AI to locate figures and flag differences, then check every number that matters against the source document.

Prompts for red flags and Q&A

TaskExample prompt
Litigation mentions”Search all folders for references to claims, disputes, litigation, arbitration or regulatory investigations. List each with document, page and a quoted sentence.”
Missing documents”Compare the index in folder 0 with the request list in the Q&A log. List requested items that do not appear in the index.”
Draft follow-up questions”Based on the employment folder (6), draft five follow-up questions a buyer’s employment lawyer would ask. Cite the document prompting each question.”
Answer a buyer question (sell side)“Draft an answer to Q&A item 47 using only documents in the room. Cite sources. If the documents do not answer it, say so and suggest who internally should answer.”

The sell-side drafting workflow is covered in detail in our guide to AI-drafted answers to buyer questions.

A safe workflow

Prompt, check, record

  1. 01 Prompt Specific task, scope, format, citations, not-found rule.
  2. 02 Review output Open every cited document and check the passage.
  3. 03 Fill gaps Search manually where the AI said not found but you have doubts.
  4. 04 Record Save the checked result in your diligence notes, not the raw AI output.

Use AI prompts in due diligence review safely

  1. Confirm approvalMake sure the AI tool is approved for confidential deal data and that agreements allow its use.
  2. Narrow the scopePoint the prompt at specific folders rather than the whole room.
  3. Ask for citationsRequire document and page references for every point.
  4. Add a not-found ruleTell the model to answer "not found" rather than guess.
  5. VerifyCheck each cited passage and every number before relying on it.
  6. Keep humans on judgementUse outputs as inputs to a reviewer's conclusion, not as the conclusion.

Signs an AI answer needs more checking

  • No citation, or a citation to a document that does not contain the text.
  • Confident statements that something does not exist.
  • Figures that do not match the source table.
  • Answers that mix up parties, years or entities.
  • Summaries that leave out qualifications or exceptions in the clause.

Our AI features guide explains what AI tools in data rooms can and cannot do, and each provider review notes whether AI help is listed.

Frequently asked questions

What is the best AI prompt for contract review in due diligence?

One that names the folder, the exact clause type, the output format and asks for citations and a "not found" answer when appropriate.

Can AI replace lawyers in due diligence?

No. It can find and extract information quickly, but legal judgement, materiality and advice stay with people.

Should I use a general AI tool or the data room's built-in AI?

Built-in AI keeps documents inside the data room environment, which is often easier to approve. Either way, check data handling and training terms.

How do I stop AI from inventing answers?

Ask for citations, limit the scope, tell it to say "not found" when unsure and check every cited passage.

Can AI check financial statements?

It can flag differences and extract figures, but every important number should be verified against the source.