AI for Divorce Lawyers: Turn Intake Notes Into a Case Chronology

Strip every identifier from the intake notes first, replace people and accounts with codes you keep in a separate key, then give a business-grade AI tool a fixed chronology template and ask it only to sort, date and tag what you supplied. Every row is checked against a source document before anything touches the file.

The order matters more than the tool. The usual way this goes wrong is pasting raw notes into a free chat window and tidying up afterwards. Do it the other way round: the redaction step is the control, the template is the quality lever, and the review step is where your professional judgement sits. The AI never decides what happened; it arranges what you told it.

Follow me on Instagram@sagnikteaches

What a chronology should do, and what AI should not touch

A divorce chronology is a working tool. It lets the fee-earner see at a glance when the parties met, married, separated, moved, bought or sold property, changed jobs, and when each child's arrangements shifted. It supports the first meeting with the client, the statement of the issues, disclosure requests and settlement discussions. Done well, it exposes gaps: "the notes say the house was bought in 2016 but there is no completion date, ask for the transfer".

Connect on LinkedInSagnik Bhattacharya

AI is good at three jobs here: pulling dated events out of rambling narrative, putting them in order, and flagging where the narrative contradicts itself. It is poor, and in places dangerous, at four others. It should not infer a date the client never gave, characterise conduct, weigh credibility, or fill a financial gap with a plausible figure. Write those four prohibitions into the prompt and into your firm's procedure.

Subscribe on YouTube@sagnikteaches

Keep the output internal. A chronology produced this way is a first draft for the fee-earner, comparable to a trainee's first pass, and it needs the same supervision. Anything served, filed or sent to the client as a statement of fact must be rebuilt or fully verified by a qualified lawyer.

Step 1: Redact before the notes leave your system

Redaction is the part most guides skip, and it is the part that decides whether the workflow is defensible. The goal is notes that are useful for sequencing but cannot identify the client, the spouse, the children or the assets if they leak.

What to replace and with what

Create a short key in your case management system (not in the AI tool) and swap each identifier for a code. Keep the codes meaningful so the output reads naturally.

Keep real dates of events (marriage, separation, completion) because the whole point is sequence. If the date itself is sensitive and unusual, shift every date by the same private offset and record the offset in the key. That preserves the order and the gaps while hiding the real calendar.

Redact in two passes

Pass one is mechanical: a find-and-replace list built from the key (names, addresses, employers). Pass two is by eye, because notes contain identifiers you did not list, such as a nickname, the name of the family's vet, or "the cafe on the corner of the high street where we met". Read the redacted notes once, slowly, as if you were a stranger. If you could work out who it is, redact more.

Do the redaction in your own document system. Some firms use a tool with a redaction feature; others do it in Word with a macro. Either is fine. What is not fine is asking a consumer chat tool to "remove the names for me", because by then the names are already in it.

Step 2: Choose where the redacted notes go

Even redacted notes are client information, so choose the destination deliberately. Three tiers are realistic for a small family practice.

• A legal-specific product with a data processing agreement, no training on your content and audit logs. Dearest to buy, easiest to defend. See the comparison in ChatGPT or a legal AI tool for a small firm.
• A business plan of a general assistant. As of November 2026, a ChatGPT Business Standard seat costs $20 a month on annual billing or $25 month to month, for teams of 2 to 200; a Claude Team Standard seat is priced the same, for 2 to 150 seats; and the Microsoft 365 Copilot Business add-on lists at $21 a user a month on an annual commitment, with an $18 first-year offer running to 31 December 2026. OpenAI and Anthropic both state that business content is not used for training by default, and Microsoft says Copilot prompts and responses are not used to train its foundation models. Read the current wording, and the contract, before relying on it.
• A consumer plan with training switched off. Cheapest, weakest. Switching off model training is a setting, not a contract, and retention periods differ by vendor. I would not put even redacted matrimonial material here.

For most small firms the middle option is a sensible start, provided the redaction step is genuinely followed. If you already live in Microsoft 365, the Copilot route keeps documents inside the tenancy you already govern.

Step 3: The chronology template

A fixed template does two things: it forces consistent output across cases and it makes review quick, because every row has the same fields. This is the template I would give a small family team.

Date or date range | Event | Parties involved | Source in the notes | Certainty | Follow-up needed

Certainty has only three values: Stated (client gave an exact date), Approximate (client gave a month, season or "around") and Unknown (the notes mention the event with no date at all). Unknown rows go in a separate section at the end rather than being slotted in by guesswork. That single rule prevents the most common AI failure in chronologies, which is a confident date that nobody supplied.

Add a column for category if you want to filter later: relationship, children, property, income and employment, health, conduct allegations raised by the client, and legal steps taken. Keep conduct allegations labelled as "client states", never as fact.

Step 4: The prompt that does the sorting

Prompts for this work should be short on personality and long on constraints. Here is one to adapt and test on redacted sample notes before using it live:

You are helping a family lawyer prepare an internal working chronology.
Use ONLY the redacted notes below. Do not use outside knowledge.

Rules:
1. Output a table: Date or range | Event | Parties | Source phrase from notes | Certainty (Stated / Approximate / Unknown) | Follow-up needed.
2. Never invent or infer a date. If the notes give none, mark Unknown and list the row in a separate "Undated events" section.
3. Quote the exact phrase from the notes that supports each row in the Source column.
4. Describe allegations as "CLIENT states ..." not as fact.
5. Do not comment on credibility, fault or likely outcome.
6. After the table, list every contradiction or gap you can see (two dates for the same event, a missing completion date, a child's age that does not fit).
7. Keep the codes (CLIENT, SPOUSE, CHILD-A, PROPERTY-1) exactly as written.

NOTES:
[paste redacted notes]

The "source phrase" requirement is the useful trick. It forces the tool to point back at the words you gave it, which turns the review step from "does this sound right?" into "does this quote say that?". If a row has no quote, or the quote does not support the event, you have found a hallucination in five seconds.

One intake, timed: 4,800 words of notes to a 55-row chronology

Take an illustrative case: a first meeting with a client after a 14-year marriage, two children, one family home, a small business owned by the spouse. The fee-earner's handwritten and dictated notes run to about 4,800 words across the first meeting and two follow-up calls.

Done the old way, a trainee reads the notes and types the chronology in about 2 hours 30 minutes, then the fee-earner spends 20 minutes checking it. Total 2 hours 50 minutes of mixed-grade time.

Done with the workflow above, the steps look like this (all timings illustrative):

1. Redaction pass one with the key list: 10 minutes.
2. Redaction pass two by eye: 15 minutes.
3. Run the prompt and read the first output: 5 minutes.
4. Row-by-row review against the original notes: about 55 rows at 40 seconds each, roughly 35 minutes.
5. Re-identify (swap codes back using the key) and add to the file: 10 minutes.

Total about 75 minutes, with the legal judgement concentrated in step 4. The saving is real but modest on a first case; it grows once the template is familiar and the redaction list is reusable. Do not promise yourself more than a 40 to 50 per cent time cut at the start.

What the output can look like on a first run (an illustrative extract):

2009 (Approximate: "summer") | CLIENT and SPOUSE meet | CLIENT, SPOUSE | "we met the summer after I finished at EMPLOYER-A" | Approximate | Confirm year
12 June 2012 | Marriage | CLIENT, SPOUSE | "married 12 June 2012" | Stated | Marriage certificate on file?
March 2016 | PROPERTY-1 purchased, joint names | CLIENT, SPOUSE | "we bought the house in March" | Approximate | Year not given in notes; get transfer deed
Unknown | SPOUSE sets up BUSINESS-S | SPOUSE | "he started the company at some point" | Unknown | Ask for incorporation date

Notice what a careful reviewer catches: the March 2016 row has a month and no year, yet an over-eager tool might have written the full date as Stated. In this extract the tool flagged it correctly because the rule said not to guess. When it does not, that row is the first thing to fix, and your rule list needs tightening.

Step 5: The review step that makes it defensible

Review is not a skim. It has a defined shape so any colleague could repeat it.

1. Row against source. For each row, find the quoted phrase in the original notes. If the phrase is missing or does not support the event, delete or correct the row.
2. Dates against documents. Where you hold a document (marriage certificate, completion statement, school letter), check the date against it and upgrade Certainty from Stated to Verified in your own column.
3. Contradiction list. Read the tool's list of gaps and contradictions, then add the ones it missed. It will miss some, because it only sees the notes.
4. Allegation wording. Search the chronology for any sentence that states an allegation as fact and rewrite it as "CLIENT states".
5. Sign-off. The fee-earner initials and dates the chronology, noting it was AI-assisted. That note is your audit trail.

Keep the original notes, the redacted version and the generated draft together on the matter file. If anyone later asks how a date got into the chronology, you can show the chain.

Other chronology jobs, and one mistake to rehearse

Separating two timelines the client blends together

Clients narrate in the order they remember things, not the order things happened. An illustrative note reads: "After he lost the job we moved, though before that we'd already started the loft conversion, which is why money was tight when the baby came." Ask the tool for a second view of the same table with each row's dependency stated ("loft conversion precedes job loss precedes move"). It will often reveal a sequence the client has not noticed, which can be a useful question for the next call.

Turning a children's arrangements history into a one-page grid

A second prompt, run on the same redacted notes, can build a grid of where each child lived, with whom and for how long in each period. Constrain it the same way: no inferred dates, no judgement about suitability, quote the source. The grid is an aid for the fee-earner preparing for the discussion about arrangements, not material for anyone else.

Asset-and-timeline cross-check

Ask the tool to list events that should have generated a document (purchase, sale, loan, inheritance, pension transfer) and mark which document the notes mention. The output is a follow-up list for disclosure: "inheritance received around 2019, no document mentioned". Never paste figures from financial disclosure into this step; describe the event category only ("inheritance, sum not stated"). Identifiable financial disclosure should not go into consumer tools at all, and even in business tools it should be reduced to categories unless your firm has approved the specific product for it.

Comparing the client's account with a later statement

When the client sends a longer written account after the first meeting, run both through the same template separately, then ask for a comparison of the two tables: events in one but not the other, and dates that moved. This is a quick way to surface inconsistencies before the client has to explain them to someone else.

A one-paragraph summary for the supervising partner

Once the verified chronology exists, ask for a 150-word neutral summary for a supervising partner, built only from the verified rows and forbidden from adding anything else. A partner reading it should be able to say "this is the shape of the case" before opening the file.

A mistake worth rehearsing: the helpful correction

Suppose the notes say the couple "separated just before the lockdown, or maybe just after". A tool can return a clean row dated to a specific month and move on, having quietly resolved the client's own doubt. The row looks fine and the Source column even quotes the sentence, but the quote does not support the month. This is why the rule is not just "quote your source" but "the quote must contain the date". Add that wording to the prompt, and make the reviewer tick the date against the quote, not merely the existence of a quote.

Handling several versions of the same event

Clients and spouses often give conflicting dates for the same event. Tell the tool to keep both as separate rows with a shared label ("Separation, version 1 (CLIENT), version 2 (CLIENT's later email)") rather than choosing one. The disagreement itself is the finding, and the follow-up column should say which document would settle it, such as a tenancy start date, a bank transfer or a school record.

Five ways a chronology draft goes wrong at review

• Invented precision. A vague "around the time of the move" becomes a specific month. It shows up in review as a Stated row with no matching quote. Fix by tightening the rule and re-running.
• Merged people. Two codes get blended when two relatives share a first name in the notes. Fix by giving each person a unique code before the run (SPOUSE-SISTER, CLIENT-SISTER).
• Dropped events. Long notes make the tool skip a minor event, such as a short separation that later matters. Run long notes in sections of about 1,500 words, then merge the tables yourself.
• Smuggled judgement. The tool adds "which suggests" to a row. Delete it and keep the prohibition in the prompt.
• Leaked identifiers. A nickname or street name survives redaction. Prevent it with the stranger's read; if it happens, treat it as a data incident under your firm's policy, record it and tell whoever is responsible for data protection.

A one-page procedure for the family team

Write a one-page procedure covering: which tools are approved, what must be redacted, where the key is stored, the template and the prompt, the review checklist and who signs off. Store the prompt in one shared place so people do not drift into their own versions. Revisit it after the first 10 matters and note what went wrong.

A short engagement-letter clause, worded with your regulator's guidance in mind, saying that approved technology may be used to help organise your information under confidentiality controls, avoids awkward conversations later. If you also record meetings, read whether lawyers can use AI note takers in client meetings first, because recording brings its own consent questions. For cleaner inputs, tighten your AI-assisted client intake, and use a checking routine for AI outputs as the model for the review stage. If you want help mapping the whole sequence for your practice, my AI implementation consultation covers exactly that; start from the contact page.

Three numbers to track across your first ten matters

Track three numbers across your first ten matters: minutes from first meeting to a usable chronology, number of rows corrected at review, and number of follow-up questions the chronology prompted that you would otherwise have missed. If corrections are above about 15 per cent of rows, your notes or template need work. If follow-up questions are zero, the contradiction step is being skipped. If the minutes are not falling, the redaction step is the bottleneck, which is the cue to build a reusable find-and-replace list per matter type.

The aim is modest and honest: a faster, more consistent first draft that a lawyer still owns. The firms that get lasting value do not chase automation of judgement. They shorten the clerical journey from notes to a document they can think with.

Questions family teams ask before trying this

Do I need client consent before using AI on intake notes?

Check your regulator's guidance and your engagement terms. One approach is a short clause saying approved AI tools may process redacted material under confidentiality controls. Whatever you decide, tell clients plainly, keep the answer on file, and never rely on a consumer chat tool's default privacy settings as your safeguard.

Can I paste a whole financial disclosure form into the tool?

Not in raw form. Strip account numbers, addresses, employer names and dates of birth first, and summarise figures by category. If a tool has a contract that bars training on your data and gives you deletion controls, risk drops, but redaction is still the first line of defence.

How long should the review step take?

Budget roughly one minute per chronology row for a first-time check, dropping to 30 seconds once you trust the template. A 60-row chronology therefore needs about 30 to 60 minutes of fee-earner time. If checking feels slower than typing the chronology yourself, the notes are probably too messy to start from.

Should the AI chronology ever go to the court or the other side?

Not as generated. Treat it as an internal working document. Anything served, filed or sent to the client as a statement of fact should be rebuilt or fully verified by a qualified lawyer who takes responsibility for each entry.

Further reads

• How Small Law Firms Use AI to Draft Letters and Routine Documents — Draft routine letters once the chronology is done.
• AI Time Capture for Solicitors: Recover Hours You Never Billed — Recover the time this workflow saves.
• How Small Law Firms Use AI to Answer Client Status Questions — Answer client progress questions faster.
• Per-Seat or Pay-As-You-Go? Legal AI Pricing for Small Firms — Understand what legal AI tools actually cost.
• AI Implementation Plan for a Small Law Firm: The First 90 Days — Sequence this with other first-90-day changes.
• AI Readiness Checklist for Accountants, Solicitors, Consultants — Check your data handling before wider rollout.
• AI Legal Chronologies for Family Cases: Events Linked to Evidence — One dated table for a family case: extract events from messages, statements and orders, tie each row to a source and page, and keep a gaps list.
• AI for Barristers: A Chambers Checklist for Casework and BSB Rules — Turn BSB guidance into a practical chambers policy and a per-matter check before any AI-assisted work is used.
• AI for Mediators: Summaries and Agreements That Stay Neutral — A neutrality checklist for AI session notes, a balanced summary prompt, agreement drafting rules and a consent clause for every party.
• Family Law Client Intake: An AI Questionnaire and Urgency Triage — A family law intake questionnaire an AI can run, with urgency flags for safety and deadlines, a conflict-check handoff and wording for distressed callers.
• Family Law Retainer Agreements: AI Drafts, Lawyer Sign-Off — A clause-by-clause walk through an AI-drafted family law retainer: scope, fees, AI-use disclosure, and the checks a lawyer must still make before it goes out.
• Can AI Build a Medical Chronology for an Injury Case? — Turn 300 pages of client medical records into a dated, page-cited chronology, then verify it with a spot-check method before it goes near a demand or a report.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT Business pricing page (OpenAI), Claude Team pricing page (Anthropic), Microsoft 365 Copilot Business pricing page and Microsoft Learn's Copilot data, privacy and security page, all checked November 2026.