Best AI for Family Law UK: Form E Prep and Client Updates

A UK family law firm can use AI to prepare Form E bank statement summaries safely by using only an approved business-grade tool under a data agreement, redacting what the task does not need, having the AI tabulate and flag rather than conclude, and checking every figure against the original statements before the client signs anything.

The confidentiality risk is real and the regulator has said so. The Solicitors Regulation Authority's warning notice on the misuse of AI, published on 17 August 2026, says client information should go into AI systems only where appropriate contractual, technical and organisational safeguards are in place, and that placing client material in an open, public AI tool is likely to breach confidentiality and may waive privilege permanently. Treat the steps below as practice, not legal advice, and check your own firm's obligations with your compliance officer.

Follow me on Instagram@sagnikteaches

What the AI is actually for in Form E prep

Form E is the financial statement each party completes in a financial remedy case. The version on gov.uk asks for all statements covering the last 12 months for each account listed, and the list includes any account held at any time in those twelve months, so an account closed in the spring still needs its statements. Form E2, the shorter statement used to vary an existing order, asks for six months of statements instead, so check which form your case uses and the current edition.

Connect on LinkedInSagnik Bhattacharya

The bank statement section is where fee earners lose hours. A client brings 14 accounts, several of them joint, as PDFs of various vintages, and someone has to turn them into a schedule of balances, regular income, large or unusual payments and gaps in the run of statements. That is a reading and tabulating job, and an AI is reasonable at reading and tabulating. It is poor at knowing what matters to the case, which is where the fee earner comes in.

Subscribe on YouTube@sagnikteaches

A division of labour that holds up

Note what is not on the AI side: no view on needs, no characterisation of spending, no suggestion about what the other party will argue. Keep the AI neutral and numerical.

Before any file goes in: the data decisions

Settle these before the first live case. They take a conversation with whoever handles data protection in the firm, not a project.

• Which tool. A business plan whose terms say workspace content is not used for model training by default. Both ChatGPT Business and Claude Team are sold for organisations and each needs at least two seats; Microsoft 365 Copilot Business is another route for firms on Microsoft 365. As price points, ChatGPT Business is listed at $25 a seat monthly or $20 on annual billing, and Microsoft 365 Copilot Business at $21 per user a month on annual billing; check the vendor pages for the current figure and its data terms.
• Which tool not. Free and consumer accounts. Their settings let users switch off training, but the label and default differ by vendor and change, and a firm cannot enforce a setting across staff.
• Agreement. A processing agreement with the vendor and a record of where data is stored. Some vendors offer a storage-region choice only on their enterprise tiers; ChatGPT Business, for one, has none. Ask, and record the answer.
• Impact assessment. Where a new tool will process clients' financial data regularly, a short written assessment before go-live is sensible. Your data-protection lead will know the format and the guidance that applies.
• Retention. How long inputs and outputs are kept, and how you delete them when the matter closes.
• Client transparency. A line in the engagement letter and privacy notice saying that approved AI tools may be used to process documents, under supervision.

Redact first, then summarise

Data minimisation is a core data-protection principle: process only what you need. For a statement summary you do not need the account holder's full address, the sort code and account number in full, or the names of the people the client paid for ordinary shopping. Redact before upload.

1. Replace the client's name with "[client]" and the spouse's with "[other party]" in the text layer.
2. Mask all but the last four digits of account numbers, and mask sort codes.
3. Leave payee names in place only where the task needs them, for example large transfers; otherwise rely on the assistant's description of the transaction type.
4. Keep an unredacted original in the matter file as the record, and keep a note of what was redacted and why.

Be realistic about PDFs. A scanned statement is an image, so redaction means drawing over it and then flattening the file so the text underneath is really gone. A text-based PDF needs a proper redaction function, not a black rectangle drawn on top, since the text below remains selectable. Test by opening the saved file and trying to copy the masked area.

The summary prompt

You are assisting a family law fee earner with a financial disclosure schedule. Below is the text of 12 monthly statements for one account, with personal identifiers removed. Do not give legal advice, do not comment on whether any spending is reasonable, and do not draw conclusions.
Produce:
1. Account name placeholder, statement period covered, and any month missing from the run (state the month).
2. A table of opening and closing balance for each month.
3. Regular credits: payee description, amount, frequency, and the months in which they appear.
4. Debits of 1,000 or more: date, description, amount.
5. Transfers that appear to be between the client's own accounts or to a named individual, listed separately without comment.
6. Cash withdrawals of 250 or more.
7. Any row where the figures do not reconcile (opening plus credits minus debits does not equal closing), quoting the page.
For every figure give the statement page or date so I can check it. If you are unsure of a figure, write "UNCLEAR" rather than guessing.

The "UNCLEAR" instruction and the page references are what make the output checkable. A figure with no reference is a figure you have to hunt for.

Illustrative output and what you would fix

For a joint current account, a realistic first pass looks like this: months covered, 11 of 12, with March missing; monthly balances in a table; regular credits of a salary figure on a recurring date and a smaller monthly credit labelled as a standing order from a named individual; four debits over 1,000, including one described as "transfer to savings"; two cash withdrawals of 300. One reconciliation flag: October's closing balance in the table does not equal the next month's opening balance.

What the fee earner does with it: asks the client for the March statement; checks the October flag against the PDF and finds the assistant had mis-read a "1" as a "7"; confirms that "transfer to savings" matches an account already on the client's list; and asks the client about the standing order from the named individual. Each of those steps is human work. The assistant gave the person a head start of roughly an hour per account on the reading.

Verifying every figure

The statement of truth on Form E is the client's, and the firm answers for the disclosure. A wrong figure caused by a misread digit is an avoidable embarrassment. Use a verification pass that does not rely on the AI.

1. Balances. Check the opening balance of the first month and the closing balance of the last month against the PDF directly. Check at least every third month in between.
2. Totals. Add the credits and debits for two sample months in a spreadsheet and compare with the statement totals if shown.
3. Large items. Open the PDF at each page the assistant cited for debits over the threshold. Confirm date, amount and description.
4. Sum of accounts. Add the assistant's closing balances for every account and compare with the total the client believes they hold. A mismatch suggests a missing account.
5. Sign-off. The fee earner initials and dates the schedule. A second person checks anything above an internal threshold.

If your team converts statements into a spreadsheet at any stage, checking every line when turning a statement PDF into Excel describes a line-by-line routine that fits here. For a second view on the same job, see summarising financial disclosure safely.

Where the summary goes wrong

The misread digit. OCR errors on older or faint scans. Defence: the page references and the spot-check pass.

The duplicated or dropped month. When 12 statements are pasted in one go, the assistant may skip one or count one twice. Defence: ask for the month list first and count it yourself before looking at any figure.

The confident label. A payment to a person is described as "gift" or "loan repayment" when the statement says only a name. Defence: the prompt forbids characterisation, and you ask the client.

The joint account blind spot. The assistant cannot tell which transactions belong to which party on a joint account. Defence: never let it attribute; ask the client.

These are failures of reading, and every one is caught by a person looking at the original. That is why the process keeps the original in front of the checker.

Client update emails that stay calm and consistent

Financial remedy cases run for months, and clients under strain want to know what is happening. A regular, calm update reduces anxious calls. An assistant is a good drafting aid for the wording, provided the update contains no confidential detail that does not need to be there and a fee earner approves it.

Draft a short client update email (under 150 words) in plain, calm English for a client in a financial remedy case.
Facts: we have received [number] of the [number] statements we asked for; we are still waiting for [item]; we will send the draft financial statement for review by [date]; their next step is [action].
Do not mention amounts, account names, the other party, or any legal opinion. Use [first name] and [your name] as placeholders. Do not promise outcomes or timescales beyond those stated.

An illustrative result: "Hello [first name], a quick update on your financial disclosure. We've received most of the statements we asked for and are waiting only for [item]. Once we have it, we'll send you the draft statement to read through by [date]. The only thing we need from you before then is [action]. If anything is hard to find, tell me and we'll work out another way. [your name]". It says what is happening, what is next and what the client must do, without a single figure. For the wider communication routine, see calm, consistent family law client emails with AI.

Emails the assistant should not draft alone

• Anything giving advice on settlement, risk or likely outcome.
• Anything responding to a client in distress or raising safety concerns. Those need a person, immediately.
• Anything discussing the other party's conduct or disclosure.

Keep the AI for logistics and tone. Judgement and care stay with the lawyer.

A worked month for a four-solicitor practice

Say a four-solicitor family practice has 12 financial remedy cases at the disclosure stage in a month, each averaging 6 accounts. The firm estimates the following, illustratively and not from a measured study.

The verification line is the honest one. A firm that leaves it out will report a larger saving until the first wrong figure appears. Subtract the tool cost and the one-off hours to write the policy and prompts, and the net remains positive but smaller. Test on three real cases before widening.

Beyond the current account: other documents in the bundle

Form E asks for more than current accounts, and the same approach works on other documents with different prompts. Keep each prompt narrow, because a prompt that tries to read a mortgage statement, a credit card bill and a pension letter together produces muddle.

Mortgage and loan statements

Ask for four things only: the lender description, the outstanding balance at the statement date, the monthly payment, and any stated redemption figure with its date. Redemption statements are time-limited, so the date matters as much as the number. An illustrative output: "Outstanding balance on statement date: figure, page 1. Monthly payment: figure, page 2. Redemption figure: not shown, UNCLEAR." The fee earner asks the lender for a redemption statement, which is a request the AI could have prompted but should not make.

Credit cards and savings accounts

For a credit card, the useful questions are the balance, the limit, and the largest payments in the period. For savings, balances and any large transfers in or out. A comparison across documents helps here: ask the assistant to list every transfer in one account and check whether a matching entry exists in another account's statements. Unmatched transfers are flags for the client conversation, not conclusions.

Below are summaries of three accounts (redacted). List every transfer of 1,000 or more that appears in one account and look for a matching amount and date (within 3 days) in another.
Return two lists: matched transfers (both accounts, dates, amount) and unmatched transfers (account, date, amount).
Do not guess what an unmatched transfer is for.

In a realistic run (illustrative) nine transfers match and two do not. One of the two is a payment to an account that is not on the client's list. That is exactly the question to raise with the client in the next meeting: is there an account that was not mentioned, or a payment to a third party? The assistant found a lead. It decided nothing.

Third-party information and sensitive details

Bank statements hold other people's data: the other party's, children's, payees'. They can also reveal special-category information, for instance payments to a medical provider or a religious organisation. Data-protection law gives that kind of data extra protection, and a family firm should handle it with the same care it already applies on paper.

• Redact payee names where the task does not need them, as described above.
• Do not upload anything that identifies a child. School names, children's account names and contact details stay out.
• Where a statement reveals something sensitive, the fee earner decides how it is handled and does not paste a reasoned discussion of it into any tool.
• Do not ask any assistant to speculate about the reason for a payment.

If something that should not have been uploaded does go in, follow the incident step in your policy straight away: stop, record what and when, tell your data-protection lead, and check the vendor's deletion route.

A schedule the fee earner can adapt

A tidy schedule is more useful to the client and the other side than a pile of summaries. Ask the assistant to produce this shape once the figures are verified, then drop it into the firm's own template.

The last two columns matter. They give the fee earner a one-glance view of what is missing and what has been checked, which is also what a supervisor wants to see in a file review.

Supervision and policy: what to write down

The SRA notice says supervisors remain accountable for work done through others, and that firms need effective systems, controls and governance for AI risk. It also says that relying on AI is no defence if false material reaches a court. A one-page policy for the family team can cover:

1. Approved tools and who may use them.
2. What may be uploaded (redacted statements, with the assessment completed) and what may not (anything involving children, safeguarding or contact details).
3. The verification pass for every AI-produced schedule, with the checker named on the file.
4. No AI use for legal advice, case analysis or drafting statements for court without a solicitor's full review.
5. Incident handling: what to do if something is uploaded in error, including who to tell and when.
6. A quarterly review of the tools, terms and any error log.

Read the SRA notice yourself on the regulator's site, ask your compliance officer to map it to your firm, and note that wording and rules develop; confirm the current position before you rely on this summary.

Choosing a tool, and what to test on one real case

General business assistants suit the summarising and drafting described here. Legal-specific platforms add case-management integration and may handle documents in ways a general tool does not; ask for written pricing, data terms and a demonstration on one of your own redacted files. Our tutorial on what to pay for and what to skip in family law AI compares the options, and a checklist for buying family law AI software turns the questions into a form you can score.

Run a four-week pilot on three live disclosure bundles. Score the tool on: figures wrong at first pass, months missed, minutes saved per account, and how long the verification pass took. If you would rather choose the first step with someone, that is what my AI implementation consultation covers; see the contact page.

Questions family solicitors ask about AI and Form E

Can I upload a client's bank statements to ChatGPT?

Not to a free or consumer account. Bank statements are sensitive personal and financial data. Use only a tool your firm has approved under a business agreement that says content is not used for training, redact what you can, and keep the original statements as the record.

Does AI replace the fee earner's review of Form E?

No. The client signs a statement of truth, and the firm remains responsible for the disclosure. An AI summary is a working aid. Every figure and every account in the draft is checked against the statements by a person before anything is filed or served.

How many months of bank statements does Form E need?

The gov.uk Form E asks for all statements covering the last 12 months for each account, and it covers any account held at any time in those twelve months, so accounts closed during the year still count. Form E2, used to vary an existing order, asks for six months instead, so check which form applies and its current version.

Do clients need to know we use AI?

Many firms say so in their engagement letter or privacy notice, and data-protection rules expect transparency about how personal data is handled. Ask your data-protection lead for the wording, and answer honestly if a client asks which tools touched their documents.

Further reads

• AI Legal Chronologies for Family Cases: Events Linked to Evidence — Link events to evidence in a case chronology.
• Family Law Client Intake: An AI Questionnaire and Urgency Triage — Intake questionnaire and urgency triage.
• Family Law Retainer Agreements: AI Drafts, Lawyer Sign-Off — Retainer drafts with lawyer sign-off.
• AI Receptionist for Family Law Firms: Handling Callers in Crisis — Handling callers in crisis.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: gov.uk Form E financial statement (section 2.3 and the documents checklist) and Form E2, Solicitors Regulation Authority warning notice on the misuse of AI (17 August 2026), OpenAI ChatGPT Business and Claude pricing pages, Microsoft 365 Copilot Business pricing page.