Business Broker NDA Process: Qualify Buyers With AI Screening

A business broker can use AI to qualify buyers by running four gates in order: a short screening questionnaire, a signed NDA, a proof-of-funds check and a scorecard. AI drafts the questions, the checklists and the summaries; the broker verifies funds with the issuing institution and makes every accept or decline decision.

The reason to run gates in order is that each one costs the buyer a little effort and costs you a little confidentiality. A casual enquirer drops out at gate one, a hobbyist at gate three. By the time someone reaches the information pack, you have a short list who can afford the business and behave sensibly with confidential material. Below is each gate, with a sample you can adapt, and the limits on what AI should do.

Follow me on Instagram@sagnikteaches

The four gates, and who does what

Notice that the second column is mostly drafting and the third and fourth are mostly checking. The aim is not to hand any decision to a model. It is to spend your time on the decisions and not on the paperwork around them.

Connect on LinkedInSagnik Bhattacharya

Gate 1: The screening questionnaire

The questionnaire arrives after an enquiry, before the NDA, and must not reveal anything about the seller. Keep it to ten or twelve questions that a serious buyer can answer in five minutes. Ask the assistant for a draft from criteria you supply:

Subscribe on YouTube@sagnikteaches

Draft a 10-question screening form for buyers of small service businesses valued between $1m and $1.5m. Cover: relevant experience, whether they will run the business day to day, budget range, how they plan to fund the purchase (cash, bank loan, seller financing, investors), timeline, who else must approve, what they are looking for in a business, other businesses they are considering, and any conditions on the purchase. Use plain wording and no jargon. Mark which questions must be answered before an NDA is sent.

An illustrative question set, after editing: 1. What businesses have you owned or managed, and for how long? 2. Will you run this business yourself? 3. What is your total budget, including working capital? 4. How much will be cash from you, and how much borrowed? 5. If borrowing, have you spoken to a lender? 6. Who else must approve a purchase? 7. When do you want to complete? 8. What sectors interest you most? 9. What would make you walk away? 10. Is there anything about timing or conditions I should know?

Reading the answers

Paste the buyers' answers into an assistant only where the data rules allow it. Buyer names and contact details are personal data, so strip them first: label buyers "Buyer 1" to "Buyer 8", and keep the key in your CRM. Ask for a table with one row per buyer and columns for experience, funds, funding mix, timeline and any red flags, quoting their words. For example: "Buyer 3: 'we're looking at a few options but money isn't an issue' (no figure given); flag: no budget." That is a useful summary, because it shows what was avoided as well as what was said. It remains your reading that decides; the table saves you from searching through twenty emails.

Which tool you use matters. Consumer ChatGPT and Claude accounts let you switch off training on your chats. Business plans, including ChatGPT Business and Claude Team, do not train on content by default, and both require at least two seats. If your seller agreements or your regulator restrict third-party tools, check before pasting any buyer detail.

Gate 2: The NDA step

The NDA protects the seller. Its wording is a legal matter, so it should come from a solicitor or a template that your professional body supplies, and not from an AI draft. What AI does well is the process around it.

• A plain-English summary for you. Paste the clause text of your own standard NDA (not a signed copy with names) and ask for a one-page summary of what the buyer agrees: purpose, definition of confidential information, no contact with staff, customers or suppliers, how long it lasts, return or deletion of documents. Use it to brief new staff.
• A buyer-facing cover note. Two short paragraphs explaining why the NDA exists and what happens after signing. An illustrative version: "Before I can share the business's name and financial information, I need you to sign the attached agreement. It's a standard confidentiality agreement and takes about five minutes. After I receive it, I'll send the information pack within one working day."
• A tracking table. Buyer, date sent, date signed, version used, signatory name (checked against the questionnaire), and the date documents must be returned or deleted.

Watch two things. First, the signatory: if a company signs, check who is authorised, and be wary of a buyer who signs for "an associate". Second, make sure no document has been sent before the signed copy is on file; a rule of "no pack without a filed signature" prevents the most common leak. The stage before this one, getting anonymous outreach in front of the right buyers, is covered in AI buyer outreach and anonymous listings.

Gate 3: The proof-of-funds check

Proof of funds is the gate where fraud and fantasy are most common, and the one where AI is the least able to help with the decision. A model reading a letter can tell you what the letter says; it cannot tell you whether the letter is real. The check is a human process.

A checklist for what a credible document shows

1. The name on the document matches the buyer, or the legal entity signing the NDA.
2. The amount covers the purchase price plus working capital, or, for a loan, states a ceiling that does.
3. It is dated within the last 30 days (many brokers use 30; choose your own window and apply it consistently).
4. It comes from a regulated institution, on letterhead, with a named contact.
5. For a financing letter, it states conditions; an "indicative" or "subject to valuation" letter is not the same as proof of cash.
6. For bank statements, balances are consistent with the claim and large recent deposits can be explained.
7. For seller financing or investor money, written commitment exists, not just a promise.

Ask the assistant to turn your own list into a form you tick per buyer, but verify each item yourself. Contact the institution using a phone number or address you find independently, never one printed on the document. Illustratively, if a buyer sends a letter on a bank's letterhead with a mobile number for the "manager", a call to the bank's published switchboard takes five minutes and settles it.

Do not paste a buyer's bank statements into a chat assistant. They contain account numbers and personal data about the buyer and others, and most privacy rules would regard this as unnecessary disclosure. Read them yourself.

Gate 4: The buyer scorecard

A scorecard replaces a vague feeling with a consistent method. It also protects you if a seller asks why a particular buyer got the pack and another did not. Draft it with an assistant, then adapt the weights to your market.

A worked example with illustrative numbers. Buyer 5 scores: funds 5, experience 3, plan 4, timeline 4, authority 5, conduct 5. Weighted total: 5 x 0.30 = 1.5; 3 x 0.20 = 0.6; 4 x 0.15 = 0.6; 4 x 0.10 = 0.4; 5 x 0.10 = 0.5; 5 x 0.15 = 0.75. Sum is 4.35 out of 5, or 87%. Buyer 2 scores funds 3, experience 5, plan 2, timeline 3, authority 2, conduct 3, which gives 0.9 + 1.0 + 0.3 + 0.3 + 0.2 + 0.45 = 3.15, or 63%. You might set a threshold of 70% for the information pack, and a conversation for anyone between 55% and 70%. The thresholds are yours; the model only does the arithmetic, and you should check it, because language models make sums mistakes.

Ask the assistant to generate a spreadsheet formula instead of totalling by itself. A simple SUMPRODUCT of weights and scores is dependable and auditable. The ChatGPT Excel guide shows how to ask for formulas like this safely.

Fairness and the human decision

Do not include characteristics such as age, nationality or family status in any criterion, and do not let an AI tool make the final decision. Declining a buyer on a tool's word alone is hard to defend and may breach rules on automated decisions that affect people. Record your reason in a neutral sentence ("funds not verified within 14 days") and keep it with the file.

Red flags to put on the form, and what each usually means

Ask the assistant to turn a list of your own past experiences into a red-flag sheet, then trim it. A workable version, with the usual explanation beside each:

• Refuses to give any budget figure. Often a browser, not a buyer. Ask once more; if there is still no figure, score funds at zero.
• Asks for the business name before the NDA, three times. Either impatient or collecting leads. Hold the line.
• Funds are "with a partner overseas" or "in a trust". Not disqualifying, but the verification takes longer and the signatory must be the person who controls the money.
• Wants customer and staff lists in the first call. Typical of a competitor. These come late, after an offer, and often under extra protections.
• Pushes for exclusivity or a rushed deadline before seeing figures. Pressure instead of evidence.
• Answers read like another buyer's. Identical wording in two questionnaires can mean an intermediary. Ask who they represent.

None of these is proof of bad faith. They are prompts for one more question. Record each in the buyer's row, so the pattern across a listing is visible.

A filled-in tabulation, so you can see the output

Illustratively, after you strip names and paste eight questionnaires, the assistant returns a table like this (three rows shown):

Buyer | Experience | Budget | Funding | Timeline | Flags
B1 | Ran two cafes 6 yrs | $1.2m total | $400k cash, bank loan discussed | 4 months | none
B2 | None stated | "flexible" | not stated | "asap" | no budget; no plan
B3 | Managed a cleaning franchise | $1.4m | $700k cash, $700k loan pre-approved | 3 months | none

Your reading: B3 proceeds to the NDA straight away, B1 proceeds with a question about the loan, and B2 receives a polite request for a budget figure before anything else. Compare the table against the original answers for the two you are least sure about. If the assistant has shortened a hedge ("probably") into a certainty, correct it in the table, because that nuance matters.

Messages you will send at each gate

Ask the assistant for starting drafts; edit them to your voice and always add real names yourself.

A polite decline that leaves a door open

Hi [first name], thank you for completing the questionnaire. At this stage I'm sharing details with buyers whose funding is already in place for a purchase of this size. If your situation changes, please contact me and I'll be happy to look again. Best wishes, [your name]

It states a reason without criticising, and it is true. Do not let the model add flattery or false encouragement ("you're a strong candidate, but...").

A reminder for an unsigned NDA

Hi [first name], just a reminder that I can't share the business's name or figures until the signed NDA is back with me. It takes about five minutes and I'll send the information pack within one working day of receiving it.

A request for fresh proof of funds

Hi [first name], the letter you sent is dated more than 30 days ago. Could you send a current one? I also need to speak to the issuing institution to confirm it, so please include a named contact. This is a standard step for every buyer.

The last sentence matters. Telling buyers the check applies to everyone makes it routine, not personal, and tends to reduce pushback from genuine buyers while putting off the others.

After the NDA: watching for breaches

An NDA is only as good as its follow-through. Set three habits. Log every document sent, to whom and on what date. Use a different version marker or watermark per buyer in the pack where your process allows it, so a leaked copy can be traced; ask your solicitor whether and how that fits your agreements. Ask returning buyers to confirm deletion or return when they withdraw, and record the date. An assistant can keep the tracking table tidy and remind you of dates, but never put the documents themselves in a chat.

If a leak is suspected, do not ask the assistant for legal advice about the buyer; contact the seller and your solicitor first. This is also where your documented gates help, because you can show exactly what the buyer signed and received.

Putting the gates into one flow

Here is an illustrative sequence for a broker handling a listing with 12 initial enquiries. Day 0: send the questionnaire to all 12 (about 15 minutes with a template). Day 2: 8 respond; you read and tabulate them (45 minutes) and send NDAs to 6. Day 4: 5 NDAs signed; send the information pack's first part to nobody yet, and request proof of funds from the 5. Day 8: 4 provide it; you verify 3 by phone (90 minutes) and score the 4. Day 9: 3 score above 70% and receive the pack. Of 12 enquiries, 3 reach the pack, and your total screening time is under four hours. Those numbers are for illustration, but the shape holds: a funnel that narrows by evidence, not by instinct.

The pack those buyers see is covered in building a confidential information memorandum with AI, and what comes next is the buyer's own checking, as in a due diligence checklist built with AI.

Adjusting the process by deal size

The gates stay the same, but their weight changes with the price. For a small business under a few hundred thousand dollars, the buyer is often a single person using savings and a loan, so the questionnaire can be shorter, proof of funds is mostly a bank letter plus a lender's indication, and the scorecard may be a five-line version. For a mid-sized business of one to five million, expect partners, investors or a company buyer, so add a question on who signs and who funds, check corporate authority on the NDA, and ask for evidence of each funding source separately. For anything larger, or where a competitor may be the buyer, add a clean-team step: a small named group sees sensitive figures such as customer pricing, and only after an offer. Ask the assistant to produce the short, medium and large versions of your questionnaire from one master list, then you review them. It is a good use of its drafting strength, because the three versions differ in a few questions and are tedious to maintain by hand.

Whatever the size, write down the version of the process you used for each listing. If a seller later asks why a buyer was or was not admitted, the record shows a consistent method, which is the best protection you have.

Realistic mistakes and how they showed up

The fluent verification. An assistant was asked "does this proof of funds letter look genuine?" and answered that the format was standard. The letter was a forgery. The mistake was asking a language model to judge authenticity. The fix is a rule: nobody says a document is genuine until the institution has confirmed it.

The early pack. A broker emailed the information pack to a buyer who had returned the questionnaire but not the NDA, because the buyer was keen and the seller was waiting. The buyer was a competitor of the seller. The fix is the filing rule from gate 2: no pack without the signed copy on file.

The over-long form. A 30-question form cost a broker two good buyers, who stopped at question 14. Ten or twelve questions is enough for gate 1; ask the rest after the NDA.

Checking that the process works

After each listing, review four things: how many enquiries reached each gate, how many buyers withdrew at each, whether any buyer who passed turned out to be unable to fund the purchase, and whether any seller complained about a leak. If funded buyers are dropping out at the questionnaire, it is too long or too intrusive. If unfunded buyers reach the pack, tighten gate 3. Adjust one thing at a time, and keep a log so the next listing starts from what you learnt.

Buyer qualification and AI: common questions

Can AI draft a business broker's NDA?

It can help you structure a plain-English summary or a checklist of clauses to ask your solicitor about, but the NDA itself should come from a solicitor or your professional body's template. Never use an AI-written NDA unreviewed, because a gap can leave the seller exposed.

Can AI verify proof of funds?

No. A person verifies it, by contacting the issuing institution through details found independently, not from the letter. AI can build the checklist and summarise what a document claims, but it cannot tell you whether a letter is genuine.

What should a screening questionnaire ask?

Experience, budget range, source of funds, financing plan, timeline, decision-makers, and what they want from a business. Keep it to ten or twelve questions; longer forms put off good buyers and add little.

Should I let AI score buyers automatically?

Use it to draft a scorecard and to total the scores you enter. The scores themselves, and any decision to decline a buyer, should come from you, because an automated rejection can be unfair or wrong and may breach rules on automated decisions.

Further reads

• Seller Discretionary Earnings With AI: Recast and Find Add-Backs — Recast the numbers the qualified buyer will see.
• Do AI Business Valuation Tools Work? What They Miss for Brokers — Where AI valuation tools mislead on price.
• Business Broker CRM With AI: What to Compare Before You Buy — Track buyers, NDAs and scores in one place.
• How to Automate Buyer and Vendor Follow-Up With AI — Follow-up routines you can automate safely.
• Business Teaser Template With AI: Attract Buyers, Hide the Seller — An anonymous teaser template with five anonymisation rules, a before-and-after rewrite and a leak check for any AI-drafted listing.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT Business and Claude pricing pages and OpenAI's data-controls help page (checked 4 November 2026). Legal wording must come from a solicitor.