AI Prompts for Due Diligence: Preparing Your Data Room Before a Buyer Asks

Due diligence is not a test you pass — it is a repricing mechanism. Buyers rarely walk away over what they find; they use it to renegotiate. So the way to prepare a data room is to generate the buyer’s question list yourself, months before they send it, and fix or disclose every answer that would otherwise cost you money at the table.

Best for: founders and owners of businesses roughly $2M–$100M in enterprise value who expect to sell, raise institutional capital, or take on a majority partner within three years — and who have never been through a diligence process from the sell side. Skip if: you’re already in an active process with a signed LOI and a banker running the data room; at that point only the AI vulnerability prompt and the deal team panel are still worth your time.

Why should I prepare a data room before anyone asks for it?

Because the alternative is preparing it under time pressure, while a buyer watches how long each answer takes. Response speed is itself a signal — a seller who produces a signed contract in an hour reads as organised; one who takes nine days reads as a business with records problems, and buyers price that perception.

The deeper reason is optionality. Findings discovered during a live process are leverage for the buyer. The same findings discovered eighteen months earlier are just problems you fixed.

What do buyers actually look for in due diligence?

Seven workstreams, in current middle-market practice: financial quality of earnings, working capital and net debt mechanics, customer concentration, AI vulnerability, EBITDA add-back defensibility, technology and cyber, and human-capital transferability.

The one most sellers underestimate is add-back defensibility. Owners routinely add back expenses they consider discretionary — vehicles, travel, family payroll, one-time projects — and buyers routinely reject a portion of them. Each rejected dollar of EBITDA is multiplied by the deal multiple, which is why a $200,000 disagreement can become a $1.4 million valuation gap.

How long does M&A due diligence take?

Financial due diligence in a typical middle-market deal runs roughly four to six weeks, though the full process from LOI to close usually spans considerably longer once legal, tax and confirmatory work are included.

Quality of Earnings consumes the largest share — around 30% of financial diligence effort — with working capital analysis at roughly 15%. That distribution tells you where to concentrate preparation: almost half of the work, and a disproportionate share of the price movement, sits in two areas.

What is a Quality of Earnings report and why does it matter?

A QoE analysis tests whether reported profit is sustainable and repeatable. The analyst typically examines three to five years of financials, isolates one-time revenue events, normalises owner add-backs to separate genuinely discretionary spending from costs the business actually requires, and arrives at a normalised EBITDA.

That normalised figure — not your P&L — is what gets multiplied. A QoE that reduces your EBITDA by 8% reduces your headline price by 8%, before anyone negotiates anything else.

What is a working capital peg, and why does it cost sellers money?

The working capital peg is a normalised target for net working capital at closing, established during diligence by analysing roughly 12 to 24 months of history. If you deliver less than the peg at close, you write a cheque for the difference. If you deliver more, you’re generally credited.

Its purpose is to stop sellers accelerating collections or delaying payables to strip cash before closing. Its effect, for unprepared sellers, is a six- or seven-figure post-closing surprise — because the peg was negotiated by people who understood the mechanic while the seller was focused on the headline number.

Prompt 1 — The Buyer’s Question List (start here)

The flagship. It generates the diligence request list before you receive one, tailored to your business rather than a generic template.

You are the deal team at a middle-market private equity firm
evaluating an acquisition. You are preparing the diligence
request list you will send after LOI.

You are not advising the seller. You do not work for them.

TARGET BUSINESS
  Industry:              [INDUSTRY]
  Revenue:               [ANNUAL REVENUE]
  EBITDA:                [EBITDA, OR "unclear"]
  Revenue model:         [SUBSCRIPTION / PROJECT / RETAINER / TRANSACTIONAL]
  Employees:             [HEADCOUNT], [W-2 vs CONTRACTOR SPLIT]
  Entity type & state:   [E.G. DELAWARE C-CORP]
  Years operating:       [N]
  Customer count:        [N]
  Largest customer:      [% OF REVENUE]
  Prior outside capital: [YES/NO — DETAILS]
  Geographies sold into: [STATES/COUNTRIES]

Produce your diligence request list organised into these
workstreams:
  1. Financial / Quality of Earnings
  2. Working capital and net debt
  3. Commercial — customers, contracts, concentration, churn
  4. Legal — corporate records, cap table, litigation, contracts
  5. Tax — income, sales/use, payroll, nexus
  6. Employment — classification, agreements, benefits, key person
  7. Technology, IP and cyber
  8. AI exposure and vulnerability

For EACH request, give:
  - The exact document or data you would ask for
  - What you are actually testing for (the real question
    behind the request)
  - What a red flag would look like
  - Roughly how that red flag would affect price, structure,
    escrow or indemnity

Rank the full list by how likely each item is to change your
offer. Put the ten highest-impact items first.

Be specific to this business. Do not produce a generic
checklist — if my industry or revenue model creates unusual
exposure, say so explicitly.

Why this works: the role assignment is load-bearing. Ask a model to “help me prepare for diligence” and it produces a generic checklist. Ask it to be the buyer writing the request list and it produces the adversarial version, including the “what I’m really testing for” column that generic checklists never contain.

Prompt 2 — The Repricing Hunt (the PE associate)

Run this after Prompt 1, with real answers. This is the prompt that finds what your accountant won’t tell you, because your accountant likes you.

You are a private equity associate. Your managing director has
signed an LOI at [HEADLINE PRICE] and has asked you to find
every legitimate reason to reduce it before closing.

You are not being unethical. You are being thorough, which in
your job is the same thing. Your bonus depends on the firm
paying less than it agreed to pay.

HERE IS WHAT DILIGENCE HAS SURFACED:
[PASTE YOUR HONEST ANSWERS TO PROMPT 1 — INCLUDING THE
 ONES YOU'D RATHER NOT WRITE DOWN]

Produce a repricing memo:

1. Every finding that supports a price reduction, ranked by
   dollar impact.
2. For each: the mechanism you'd use — EBITDA adjustment,
   multiple compression, escrow, indemnity, earnout, working
   capital peg adjustment, or walk-away threat.
3. For each: the specific dollar or percentage impact, and
   show the arithmetic.
4. The three findings you'd lead with in the renegotiation
   conversation, and the order you'd raise them in.
5. The findings you'd hold back to trade away later.
6. Anything that would make you recommend walking rather than
   repricing.

Then, separately, in one paragraph: what is the seller's
strongest counter-argument on each of your top three, and
how would you respond to it?

Do not soften anything. Do not add encouragement. If a finding
is not actually material, say so — false positives waste your
MD's time.

Why the framing matters: the instruction that this is “thorough, not unethical” prevents the model from refusing or hedging. Without it, models frequently produce a balanced view of the transaction, which is precisely the thing you already have and do not need.

Prompt 3 — The Add-Back Defensibility Test

The highest-leverage arithmetic in the whole process. Each rejected add-back is multiplied.

I am preparing a Quality of Earnings analysis on my own
business before going to market.

Reported EBITDA: [FIGURE]
Expected multiple: [E.G. 6x]

MY PROPOSED ADD-BACKS:
[LIST EACH: description, annual amount, and your one-line
 justification. Include the ones you're unsure about.]

For each add-back, tell me:
  1. Would a buyer's QoE accountant accept, reject, or
     partially accept it? State which.
  2. What evidence would I need to defend it? Be specific
     about documents.
  3. If rejected, the EBITDA impact AND the valuation impact
     at my multiple. Show the multiplication.
  4. Whether it is worth defending or better dropped
     pre-emptively — some add-backs cost more in credibility
     than they add in value.

Then give me:
  - Total EBITDA if every add-back is accepted
  - Total if the likely-rejected ones are removed
  - The valuation gap between those two numbers
  - The three add-backs where the evidence gap is largest
    relative to the dollars at stake

Flag any add-back that would make a buyer question the
integrity of the whole set. One indefensible add-back causes
scrutiny of all of them.

Label all figures as illustrative pending accountant review.

Prompt 4 — The Customer Concentration Stress Test

Analyse my customer concentration as a buyer would.

CUSTOMER DATA:
[FOR TOP 10: customer, % of revenue, % of gross profit,
 years as a customer, contract end date, notice period,
 auto-renew yes/no, change-of-control clause yes/no,
 relationship owner — me or an employee]

Total customers: [N]
Revenue retention rate: [%, OR "unknown"]

Assess:
  1. Where does my concentration sit relative to what a
     buyer would consider acceptable in [INDUSTRY]?
  2. Which specific customers would trigger a price
     adjustment, and what mechanism would a buyer use?
  3. Which contracts contain change-of-control provisions
     that let the customer exit or require consent on a
     sale? What is that worth to a buyer as leverage?
  4. Which relationships depend on me personally? That is
     key-person risk and it usually converts into an earnout
     rather than cash at close.
  5. What could I do in the next 12 months to materially
     improve this picture — ranked by impact per unit of
     effort?

Be direct about which risks are structural and cannot be
fixed before a sale. Those need to be priced and disclosed,
not managed.

Prompt 5 — The IP Chain-of-Title Audit

Deal-critical for technology businesses and routinely missed until it is expensive.

Audit my intellectual property chain of title as an
acquirer's counsel would.

WHO HAS WRITTEN CODE, CREATED DESIGNS, OR PRODUCED
COPYRIGHTABLE WORK FOR THIS BUSINESS:
[FOR EACH: name/role, employee or contractor, dates,
 country, whether they signed an IP assignment or
 work-for-hire agreement, whether you can locate that
 document today]

ALSO:
  Open-source components in the product: [LIST OR "unknown"]
  Any code written before the entity existed: [YES/NO]
  Any code from a prior employer's time: [YES/NO]
  Trademarks registered: [LIST]
  Anything built with AI code assistants: [YES/NO, DETAIL]

Identify:
  1. Every gap where the company may not own what it
     believes it owns
  2. Which gaps are fixable now, and the mechanism —
     confirmatory assignment, consulting agreement, buyout
  3. Which are difficult or impossible to fix retroactively
  4. How an acquirer would treat each gap: escrow,
     indemnity, specific representation, or a condition
     to closing
  5. Open-source licence obligations that could affect
     the product's distribution or a buyer's plans

Be blunt about the ones that are genuinely serious. In
technology deals, IP ownership gaps are among the small
number of findings that can stop a transaction rather than
merely reprice it.

Prompt 6 — The Tax Exposure Sweep

Identify tax exposures a buyer's diligence would surface.

BUSINESS PROFILE:
  Entity type and state of formation: [DETAIL]
  States/countries with employees:    [LIST]
  States/countries with customers:    [LIST]
  Revenue by state, roughly:          [DETAIL]
  Do you sell software, SaaS, digital goods, or services?
                                      [WHICH]
  Do you collect sales tax anywhere?  [WHERE]
  Contractor headcount and roles:     [DETAIL]
  Any equity compensation issued?     [DETAIL]
  Prior year returns filed on time?   [YES/NO]

Assess exposure in each area:
  1. Sales and use tax — economic nexus. Since the 2018
     Wayfair decision, physical presence is not required to
     create a filing obligation. Which states am I likely
     to have triggered, given my revenue distribution and
     what I sell?
  2. Worker classification — which contractor arrangements
     would a buyer's counsel flag as likely misclassification?
  3. Payroll tax and state registration in employee states
  4. Equity compensation — missing 83(b) elections,
     unissued promised options, valuation support for
     strike prices
  5. Income tax nexus and apportionment

For each exposure, estimate: rough magnitude, how a buyer
would handle it (escrow, indemnity, purchase price
reduction, or condition to close), and whether voluntary
disclosure before a sale would reduce total cost.

State clearly where you are uncertain. Do not guess at
state-specific thresholds — name the states to check and
what to check for.

Note on this one: sales tax nexus is the exposure most commonly missed by software and e-commerce sellers, and it compounds silently for years. Treat the output as a list of questions for a state and local tax specialist, not as a conclusion.

Prompt 7 — The AI Vulnerability Assessment

New in 2026 diligence, and absent from almost every checklist currently published.

You are a private equity associate assessing whether this
business is exposed to disruption by AI over a five-year
hold period. Your firm will own it for that long.

BUSINESS:
  What we sell:            [DESCRIPTION]
  How customers use it:    [DESCRIPTION]
  What our staff spend
    their time doing:      [BREAKDOWN BY FUNCTION]
  Pricing model:           [DETAIL]
  What makes customers
    stay:                  [YOUR HONEST ANSWER]

Assess:
  1. Which parts of our value proposition could a competent
     team replicate with current AI tooling in 18 months?
  2. Which of our roles are most exposed to automation, and
     what share of our cost base do they represent?
  3. Is AI a margin opportunity for us, a threat to our
     pricing, or both? Be specific about mechanism.
  4. What would a buyer worry about most here?
  5. What evidence would reassure them — and do we have it?
  6. If our moat is a data asset, a regulatory position, or
     a relationship, say which. If we don't have one, say
     that plainly.

Then: how would you, as the buyer, structure around this
risk if you still wanted the deal? Earnout? Shorter hold
assumptions? Lower multiple?

Do not be reassuring. A buyer will not be.

Prompt 8 — The Disclosure Schedule Builder

Help me build the disclosure schedules that will accompany
a purchase agreement.

Disclosure schedules qualify the seller's representations.
Anything properly disclosed generally cannot later be
claimed as a breach. Anything omitted can.

MY SITUATION:
[PASTE THE FULL OUTPUT OF PROMPTS 1-7, PLUS ANYTHING
 UNCOMFORTABLE THAT HASN'T APPEARED YET]

For each standard representation category — organisation and
good standing, capitalisation, financial statements,
undisclosed liabilities, material contracts, litigation,
compliance, tax, employment, intellectual property, customers
and suppliers, insurance, related-party transactions —
tell me:

  1. What I would need to disclose given what you know
  2. What documents support each disclosure
  3. Where I currently have a gap between what I'd need to
     say and what I can evidence
  4. Which disclosures are routine and which will attract
     buyer attention

Then: rank every gap by the risk of it being discovered
independently. Anything a buyer would find on their own is
something I should disclose first, because a discovered
omission damages trust across the entire process.

Do not draft the schedules. Counsel does that. Produce the
inventory and the gaps.

What separates a strong diligence prompt from a weak one?

The pattern: weak prompts ask the model what it knows. Strong prompts assign an adversarial role, supply your actual numbers, demand the arithmetic, and force a ranking by dollar impact.

Real example: input → output

Input to Prompt 2 (figures illustrative; composite of common findings, not a real transaction):

Excerpt from the repricing memo:

Total identified movement: roughly $1.4M of headline reduction, $6–7M of consideration shifted to earnout, and $400–900K into escrow — against a $24M LOI.

That is what a competent associate produces in an afternoon. Every finding was discoverable by the seller a year earlier, and three of the four were fixable.

Level-up: the Deal Team Panel

The advanced version, and the part competing guides won’t have. A single adversary finds one category of problem. A real diligence team is three specialists with different incentives who disagree with each other, and the disagreements are where the useful information sits.

Convene the buyer's full deal team to review this business.
Each member has a different mandate and a different blind
spot. They must genuinely disagree.

PANEL:
1. THE PE ASSOCIATE — optimises for price and structure.
   Thinks in multiples and mechanisms. Blind spot: will
   propose structures the seller would never sign.
2. THE QoE ACCOUNTANT — optimises for defensible numbers.
   Cares only about what survives scrutiny. Blind spot:
   ignores commercial reality and relationship value.
3. THE M&A LAWYER — optimises for reducing the buyer's
   post-closing risk. Blind spot: would paper the deal to
   death and lose it.
4. THE OPERATING PARTNER — will actually run this business
   after close. Cares whether it works on Monday morning.
   Blind spot: falls in love with the asset.

MATERIAL: [PASTE OUTPUT OF PROMPTS 1-7]

ROUND 1 — Each gives their assessment in under 200 words,
including the single item they most want addressed before
closing.

ROUND 2 — Each responds to the panellist they most disagree
with, by name. Require at least two genuine disagreements.
The associate and the operating partner should conflict on
earnout structure. The lawyer and the associate should
conflict on how much risk is worth papering.

ROUND 3 — Each states the one condition under which they'd
change their position.

ROUND 4 — Joint recommendation to the investment committee:
   · Revised offer and structure
   · What must be resolved before signing
   · What can be handled by indemnity or insurance
   · What they suspect the seller has not disclosed
   · The single item most likely to break the deal

Then, in your own voice as facilitator: which panellist is
most right, and why? Name them.

Finally — and this is the part I actually need — reverse it.
Given everything above, what should the SELLER have done in
the twelve months before this process to eliminate each
issue? Rank by cost of the fix relative to the value it
would have preserved.

Why the reversal at the end matters: it converts an adversarial exercise into a work plan. The panel identifies what a buyer will extract; the final instruction converts each item into a task with a deadline you still have time to hit.

How do I actually run this end to end?

1. Run Prompt 1 cold, before gathering anything. It tells you what to collect and in what priority order.
2. Spend two weeks gathering the top ten items. Note precisely which ones you cannot produce — that list is your real problem inventory.
3. Run Prompt 2 with honest inputs, including the answers you’d prefer not to write down. Dishonest input produces a comfortable output and wastes the exercise.
4. Run Prompts 3–7 in whichever order matches your exposure. Technology businesses should prioritise 5 and 7; multi-state sellers should prioritise 6.
5. Run Prompt 8 to convert findings into a disclosure inventory.
6. Run the Deal Team Panel, and use the reversal at the end as your twelve-month work plan.
7. Take the output to your attorney, your CPA and a transaction advisor. You will have compressed several billable meetings into one and arrived with better questions than most sellers ever ask.

Which model should I use for each prompt?

Model-version note (August 2026): assignments validated 13 August 2026. Enable extended reasoning for Prompts 2, 3 and the panel. Any prompt touching tax thresholds, state registration or current regulation should be run with web access, and every figure it returns must be verified against a primary source. Models produce confident, incorrect tax and legal figures routinely — treat every number as a claim to check, not a fact.

Frequently asked questions

What is a due diligence checklist?

A structured list of documents and data a buyer requests to verify what a seller has claimed. In middle-market M&A it typically spans eight workstreams: financial and quality of earnings, working capital and net debt, commercial and customer, legal and corporate, tax, employment, technology and IP, and — increasingly in 2026 — AI exposure. The list is not neutral; each item exists because it has produced findings in prior deals.

What do buyers look for in due diligence?

Evidence that reported earnings are sustainable and that no undisclosed liabilities exist. In practice that means testing add-backs, revenue quality, customer concentration and contract terms, worker classification, IP ownership, tax registration and filings, and key-person dependency. Buyers are not looking for perfection — they are looking for anything undisclosed, because a discovered omission damages credibility across every other answer you give.

How long does M&A due diligence take?

Financial due diligence in a typical middle-market transaction runs approximately four to six weeks. The full period from signed letter of intent to closing is usually longer once legal, tax and confirmatory workstreams are included, and timelines extend when the seller cannot produce documents promptly. Response speed is itself observed by buyers and shapes their view of operational quality.

What kills M&A deals?

Genuine deal-breakers are rarer than repricing events. The findings that most often stop a transaction rather than adjust it are intellectual property the company does not actually own, undisclosed litigation with material exposure, revenue that cannot be substantiated, and discovered omissions that destroy trust. Most other findings — concentration, tax exposure, add-back disputes — become price adjustments, escrow, indemnities or earnouts.

What is a working capital peg?

A normalised net working capital target set at closing, calculated during diligence from roughly 12 to 24 months of historical balance sheets. Deliver less than the peg and you pay the difference; deliver more and you are generally credited. It exists to prevent sellers stripping cash before close by accelerating collections or delaying payables, and it is a frequent source of post-closing surprises for sellers who did not negotiate it closely.

When should I start preparing for due diligence?

Twelve to twenty-four months before you intend to go to market. That window matters because the highest-value fixes have long lead times — voluntary tax disclosure agreements, confirmatory IP assignments from former contractors, renegotiating customer contracts to remove change-of-control clauses, and building three years of clean, consistently presented financials all take months rather than weeks.

What is representation and warranty insurance?

An insurance policy covering losses from breaches of the seller’s representations in the purchase agreement — typically financial statements, tax, undisclosed liabilities, compliance and material contracts. Buy-side policies are the market-standard structure, and RWI is now used in roughly two-thirds of middle-market private transactions, with higher adoption in private equity deals. It can allow sellers to exit without ongoing indemnity obligations or escrow.

Can AI help with M&A due diligence?

For preparation, genuinely yes — generating the buyer’s likely question list, stress-testing add-backs, and modelling repricing scenarios are all tasks where a structured adversarial prompt outperforms an unstructured conversation with an advisor who is invested in your good mood. For execution, no. Models produce confident and incorrect tax, legal and accounting figures, and diligence findings carry contractual consequences. Use AI to prepare; use professionals to conclude.

About the author

E-E-A-T note for whoever publishes this: on a transactional finance topic the four highest-value signals, in order, are — a named human with a linkable professional profile; a quantified transaction record; an explicit conflict-of-interest disclosure; and a credential specific to M&A rather than general business. Add a visible “Reviewed by [CPA / transaction attorney, name and credential]” line if you can obtain one. On YMYL finance content, reviewer attribution is increasingly what separates cited sources from ignored ones.

Sources and further reading

• Valutico — What Buyers Actually Look For in 2026 Due Diligence
• Intralinks — Financial Due Diligence in M&A: Definition, Importance, and 2026 Best Practices
• Dealroom — How to Conduct Financial Due Diligence
• Rehmann — Navigating M&A Diligence: Unlocking True Deal Value with Quality of Earnings
• Warren Averett — What Happens in a Quality of Earnings Analysis?
• Anders — Quality of Earnings Report: What It Is and Why It Matters in M&A
• CBIZ — Representations and Warranties Insurance: Trends and Best Practices
• IMAP — The Sell-Side Process in Mid-Market M&A: How Founders Can Prepare Strategically

Primary sources to consult directly where possible: the ABA Model Stock Purchase Agreement and its commentary; IRS guidance on worker classification, including the common-law control test; state departments of revenue for economic nexus thresholds following South Dakota v. Wayfair (2018); and ASC 606 for revenue recognition in subscription businesses.

Last updated

Version notes

• Deal-term figures — RWI adoption (~two-thirds of middle-market private deals), QoE effort share (~30%), working capital analysis share (~15%) and the 4–6 week financial diligence window were verified against the sources above on 13 August 2026. These shift with market conditions; re-verify quarterly rather than annually.
• Tax thresholds — deliberately not stated. Economic nexus thresholds vary by state and change; the guide names the question rather than the number for that reason. Do not add specific thresholds without a state-by-state citation and a checked date.
• Model assignments — written and tested against [MODEL NAME AND VERSION] on [DATE]. Re-test after any major release; multi-role prompts like the Deal Team Panel are the most fragile.
• Worked example — figures are illustrative and constructed from common findings, not a real transaction, and are labelled as such in the text. If replaced with a real anonymised deal, confirm confidentiality obligations first.