Estate Planning Questionnaire: AI Intake That Feeds the First Draft

An estate planning firm can use AI to turn a client questionnaire into a first draft by designing structured questions, mapping each answer to a clause or a decision flag in rules the firm controls, letting AI list gaps and inconsistencies, and filling an approved template. A lawyer then reviews the draft and advises the client before anything is signed.

This is a design job for the firm, not a form for clients to download. The questionnaire, the mapping and the review point decide whether AI saves hours or creates a fresh pile of errors to untangle. The sections below set out each piece in the order you would build it.

Follow me on Instagram@sagnikteaches

Why the questionnaire is the hard part

When a first draft goes wrong, the cause is usually upstream. A client wrote "my children" and meant two children from a first marriage and one stepchild. A client listed a house but not that it is held jointly with a sibling. An answer sits in a free-text box and never reaches the clause that depends on it. The drafting is the easy part once the facts are captured cleanly.

Connect on LinkedInSagnik Bhattacharya

So the order of work is: fix the facts you need, design questions that capture them without ambiguity, decide what each answer does downstream, and only then bring AI in to chase gaps and assemble text. A firm that begins with the drafting tool and works backwards ends up with a tool that produces confident documents from vague data.

Subscribe on YouTube@sagnikteaches

Step 1: list the facts every plan needs

Start from the documents you actually draft, not from a generic list. For a firm producing wills, incapacity documents and simple trusts, the fact groups are roughly these:

• The client. Full legal name, any other names used, date of birth, marital or partnership status, residence.
• Family. Spouse or partner, children (including from previous relationships, adopted and stepchildren), their ages, any who have special needs or are vulnerable to creditors or divorce.
• Assets. Property and how it is held, accounts, pensions and retirement accounts with named beneficiaries, business interests, life insurance, digital assets, items of sentimental value.
• Liabilities. Mortgages, loans, guarantees.
• Wishes. Who inherits, in what shares, at what ages, who gets specific gifts, what happens if a beneficiary dies first.
• Roles. Executor or trustee, a backup, guardians for minor children, people to act if the client loses capacity.
• Existing documents. Earlier wills, trusts, beneficiary designations.

For each fact, write down which document or clause uses it. A fact that feeds nothing should not be on the form. Clients abandon long questionnaires, and every unnecessary question costs completion.

Step 2: design questions that remove ambiguity

Good intake questions are narrow and ask for one thing. Compare a vague question with a better one:

Use dropdowns, repeating blocks and conditional questions (ask about guardians only if there are children under 18) wherever answers feed a clause. Keep free text for the things that truly need it: wishes about personal items, family circumstances in the client's own words, anything the client wants the lawyer to know.

AI earns its place here at two points. It can rephrase a question for plain language when a client says they do not understand, and it can ask one follow-up when an answer is thin. Both uses stay inside the form, and neither invents facts.

Step 3: map answers to clauses in a rules table

This is the heart of the build and the part generic articles skip. The mapping table says, for each answer pattern, which clause to insert, which variables to fill and which questions to put to the lawyer. It belongs to the firm, written once by a lawyer, kept under version control.

An illustrative extract:

Notice how many rows end with a flag rather than a clause. That is the right shape. The simple cases get a clean first draft, and the complicated ones are routed to the lawyer early with the reason attached. The firm decides which patterns are simple enough to automate; AI should not decide that on the fly.

The AI's job in this step is mostly clerical. It reads the structured answers, applies the table, assembles the draft from the approved clause library, and reports which rows it used. Where answers do not match any row, it reports "no rule" instead of improvising. Ask for that behaviour explicitly in the instructions you give it.

A worked example with numbers

Say a client in her fifties completes the questionnaire for a will. She is married, has two adult children together, owns a home worth $540,000 with $120,000 outstanding on the mortgage, holds $85,000 in savings, a retirement account of $210,000 with her spouse named as beneficiary, and a small business interest she values at $60,000. She wants everything to her spouse, then to the children equally, and names her sister as executor with no backup.

The intake system produces a summary the lawyer sees first:

CLIENT SUMMARY (illustrative)
Married; 2 adult children (joint); no special needs reported.
Estate: home $540,000 less mortgage $120,000 = $420,000 equity; savings $85,000; business interest $60,000 (client's valuation).
Total net estate, as stated: $565,000 (excludes retirement account, which passes by beneficiary designation: $210,000).
Pattern selected: all to spouse, then children equally -> standard clause set applied.
GAPS: (1) no backup executor; (2) home title: joint or sole? not answered; (3) business interest: structure and any partner agreement not stated.
FLAGS: business interest may need separate advice.

The sum is checkable: $420,000 plus $85,000 plus $60,000 is $565,000, and the retirement account is shown separately because it passes by designation, not by will. Putting that line in the summary is a design decision. It stops the client and the draft from treating money the will does not control as part of the estate.

The first draft is assembled from the standard clause set, with the executor clause carrying a visible placeholder for the backup. The lawyer sees three gaps and one flag, calls the client about them, and amends the draft. The time saved is in not retyping facts and not hunting for what is missing, not in the lawyer's judgement.

Step 4: let AI hunt for gaps and contradictions

Once answers are structured, a second pass can look across them for problems. This is where a language model is genuinely useful, because contradictions often sit between fields, not inside one. A prompt along these lines, run on the completed questionnaire, returns a list the intake staff check before the lawyer sees the file.

Review the completed estate planning questionnaire below. List only:
1. Required fields left blank.
2. Answers that contradict each other (for example, "no children" but a guardian named).
3. Assets mentioned in free text that are not in the asset table.
4. Roles with no backup.
5. Any beneficiary who is a minor, or who may need protection.
Do not give legal advice and do not suggest clauses. Quote the answer each item refers to.

A realistic output, with the sort of items it finds:

- Free text mentions "the boat at the marina"; not in the asset table.
- Marital status "divorced" but spouse named as sole beneficiary and executor.
- Guardian section completed, but children listed as aged 19 and 22.
- No backup trustee named for the trust for the 15-year-old.

The second and third items are exactly what a tired person misses on page four. Each becomes a question to the client, asked by staff in a message the firm approves in advance, which keeps the AI out of direct advice.

What stays out of the gap check

The model does not tell the client what they should do, estimate tax or say that a clause is enough. If the output starts to include recommendations, tighten the prompt and discard those lines.

Step 5: the draft and the lawyer's review

The first draft is a document assembled from clauses the firm has already approved, with variables filled from structured answers. A model can help with the assembly and with a plain-English cover note, but the clause wording itself should come from your library. Free-form generation of legal clauses is where inventive errors creep in.

The lawyer's review is a checklist, not a read-through:

1. Compare the client's stated wishes in their own words with the pattern selected. If they differ, the pattern is wrong.
2. Read every flag and gap and resolve it with the client.
3. Check names, shares and ages against the questionnaire, line by line.
4. Confirm that every asset that matters is either dealt with by the document or noted as passing another way.
5. Read the whole document once as the client would, for anything surprising.
6. Advise the client, and record the advice.

The review point must be a hard stop in the process: no document is released to a client without a named lawyer's sign-off recorded in the file. If you are planning the firm's whole AI set-up, the two-lawyer estate planning firm first-month plan puts intake alongside the other early steps.

A second example: a blended family, routed rather than drafted

Take another illustrative client, divorced and remarried, with one adult child from the first marriage, one teenage child with the new spouse, and a stepchild the client helped raise. On the form the client chooses "everything to spouse, then children equally" because it looks simplest, and lists "my children" as three names.

The mapping table catches this on two rows at once: a child from a prior relationship, and a beneficiary under 18. No clause is inserted. The system produces a routed summary instead:

ROUTED TO LAWYER (illustrative)
Reason 1: child from a prior relationship (blended-family rule).
Reason 2: beneficiary aged 15 (trust-until-age rule).
Reason 3: stepchild listed as "my children"; relationship not legally defined on file.
Client's stated wish: everything to spouse, then children equally.
Question for lawyer: does "children" include the stepchild? What happens to the first child's share if the spouse remarries?
No draft generated.

This is the system working as intended. A model left to draft would have produced a perfectly tidy will that quietly cut out a stepchild or left an adult child dependent on the goodwill of a surviving spouse. The firm's mapping table said "stop", and the lawyer got the question stated in plain terms before the first meeting.

Writing the follow-up messages the firm approves in advance

Gaps turn into client messages, and a firm can pre-approve the wording so staff send them quickly and consistently. A model can personalise the blanks from the gap list, but the text itself comes from the library. Two samples, in the voice the firm might choose:

Hi [first name], thanks for completing your questionnaire. Before we can draft, we need two details. Who would you like as a backup executor if your first choice cannot act? And is your home held in your sole name or jointly with someone? If you are not sure, a copy of the title paperwork will tell us. Thanks, [your name]

Hi [first name], one of your answers lists a trust. Could you send us the name of the adviser who set it up, or a copy of the document? We need to see it before we can plan around it. Thanks, [your name]

Each message asks for one or two things, explains why, and offers an easy way to answer. The firm's rule is that these messages come from approved wording and never contain advice about what the client should do.

How firms usually build this

You do not need custom software on day one. A workable low-cost build has four parts. A form tool that supports conditional questions and repeating blocks holds the questionnaire. A spreadsheet holds the mapping table, one row per rule, owned by a named lawyer. A document-assembly feature in your word processor or practice software holds the clause library with variables. And a general-purpose AI assistant, used under the firm's data rules, runs the gap check and drafts the plain-English cover note. The software comparison weighs dedicated tools against this kind of assembly, and the right choice depends on volume and on how often your clients' situations fall outside the simple patterns.

Whichever route you choose, keep the mapping table and the clause library in one place that both the lawyers and the person running the system can read. A rule that lives only in someone's head or inside a vendor's black box is a rule you cannot audit.

Where intake goes wrong

Questions the client cannot answer

"Is the property held as joint tenants or tenants in common?" stumps most people. Offer a short explanation and an "I don't know" option that routes to a document request, such as the title paperwork. An "I don't know" captured honestly is better than a guess that flows into a draft.

Answers that change after submission

People remember a second property or a new grandchild after the form is in. Build a simple revision flow: a changed answer creates a dated new version, and the draft regenerates from the new version while the old one is kept. Keep the earlier version beside the new one, so the reviewing lawyer can see what changed, when and at whose request, and can check that the draft followed the change.

A "yes" that means different things

"Do you have a trust?" may mean a family trust, a pension scheme or a savings account for a child. Ask for the document or the name of the adviser who set it up, and treat the bare "yes" as an open question for the lawyer.

Over-trusting the summary

A tidy client summary invites the reviewer to skim. Require the lawyer to open at least the questionnaire answers for any flagged item, and sample unflagged files regularly to see whether the flags missed things.

Keeping client information protected

Estate planning intake is some of the most personal material a firm holds: family circumstances, health hints, finances and relationships. Decide where it may go. Use tools whose terms you have read, whose retention and training settings you control, and which your firm's rules and its professional body allow. Do not paste questionnaires into a personal chatbot account. Limit access inside the firm to the people working on the file, and keep a list of systems that hold answers so you can respond properly if a client asks where their data is, or asks for it to be deleted.

If you use AI to assist with taking notes or recording calls during intake, the issues are covered in whether lawyers can use AI note takers in client meetings.

Measuring whether it works

After the first ten clients, collect four figures: the minutes from questionnaire complete to a draft the lawyer is ready to review, the number of gaps the AI found that staff confirmed were real, the number of corrections the lawyer made to the first draft, and the number of times the client had to be re-contacted for something a better question would have captured. The last two numbers drive improvement. Each correction or re-contact points at a question to redesign or a mapping row to add.

An illustrative pattern, not a measured result: early files need many questions reworded, and by the fifth or sixth client the form stabilises. If the corrections stay high past ten clients, narrow the scope to the simplest plan type until the questionnaire and mapping settle.

Starting small

Pick one document type, such as a simple will for a married couple with adult children. Build the questionnaire, the mapping table and the checklist for that alone. Run three real clients through it with a lawyer reviewing everything as usual. Only then add a second plan type. If you want a hand shaping the field list and mapping table for your own documents, my AI implementation consultation is a good place to start, and the intake method for enquiries is covered in AI client intake for law firms.

Questions firms ask when designing AI intake

Should clients talk to a chatbot or fill in a form?

Start with a structured form with some AI help, such as follow-up prompts for unclear answers. A free-form chatbot gives rich answers but is harder to map to clauses and to audit. You can add conversation later, once the field list is stable.

Can AI decide which clauses a client needs?

It can propose clauses from rules you write and flag unusual answers. The lawyer decides what the client needs and advises on it. Keep the mapping rules in a document the firm owns and a lawyer has approved.

What about client confidentiality?

Treat intake data as privileged client material. Use only tools whose terms you have read, switch off training on your data where possible, and keep a record of which systems hold questionnaire answers.

Do clients need to be told AI is used?

Check your professional rules and your engagement terms. A plain sentence in the engagement letter, saying that AI assists with drafting and that a lawyer reviews every document, is one simple way to cover it.

Further reads

• How to Write Terms of Business With AI and What Needs a Lawyer — Where AI drafting stops and a lawyer starts.
• AI Governance Checklist for Small Financial Advice Firms — Writing down the rules your staff follow.
• AI Time Capture for Solicitors: Recover Hours You Never Billed — Recovering unbilled time in a small law practice.
• Can AI Sort and File Incoming Documents for Your Business? — Filing the paperwork clients send with their answers.
• AI Compliance Checklist for Small Businesses: What Applies to You — A wider check on how AI use is governed.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: a retirement plan provider's member notice (a beneficiary designation overrides a will and passes outside probate) and a public legal-education page on joint tenancy and survivorship, both read 4 November 2026; general document-assembly practice. Legal rules vary by jurisdiction, so the lawyer confirms them for each client.