How to Use ChatGPT as a Litigant in Person: A Practical Guide to Preparing Your Court Case

What ChatGPT can — and cannot — safely do

Set up a dedicated ChatGPT Project for the case

If you intend to use ChatGPT repeatedly during proceedings, a Project is usually more manageable than trying to run the entire case through one enormous conversation.

Projects can keep related chats, uploaded files and project-specific instructions together.

Use a neutral project name

Avoid putting a child’s full name or a highly sensitive allegation in the Project title.

Consider project-only memory

If available on your account, project-only memory limits the Project’s memory context to that Project. Chats in the Project can use other conversations within the same Project but do not draw on conversations outside it.

Project memory settings can now be changed later through the Project settings, so you do not necessarily have to make an irreversible decision at the moment the Project is created.

Add clear Project instructions

Good instructions matter more than asking AI to “act like the best barrister in the country”.

The aim is to force the system to preserve uncertainty and distinguish different kinds of information.

Before uploading documents: privacy and confidentiality

This is one of the most important parts of using AI responsibly in Family Court work.

Court files may contain:

• children’s names and dates of birth;
• addresses;
• medical information;
• school and SEND records;
• Cafcass material;
• police disclosure;
• social-care records;
• domestic-abuse allegations;
• financial information;
• information about third parties; and
• confidential or privileged communications.

Review your ChatGPT Data Controls

On a personal ChatGPT account, review Settings > Data controls.

If you turn off Improve the model for everyone, OpenAI currently says that new conversations will not be used to improve its models, although they can remain in your account history.

Use data minimisation

Give the system only what it actually needs.

Where appropriate, consider working from copies in which you have removed unnecessary:

• full names;
• dates of birth;
• addresses;
• school names;
• telephone numbers;
• email addresses;
• bank details;
• signatures;
• official identifiers; and
• irrelevant third-party information.

Use labels such as Applicant, Respondent, Child A and School where the identity itself is unnecessary to the task.

Redact properly

Drawing a black box over words in some PDF tools can leave the underlying text recoverable.

Use proper redaction functionality and retain the untouched original separately.

Family proceedings have additional restrictions

Information from proceedings concerning children can be subject to statutory, procedural and court-ordered restrictions on communication.

The fact that you possess a document does not automatically mean you can communicate it without restriction to another person or external service.

Remember retention

Files added to a ChatGPT Project may remain associated with that Project until the file or Project is deleted, subject to applicable retention arrangements.

Your ChatGPT Project should therefore never be your only copy of a case file.

Keep your master case file outside ChatGPT

Treat ChatGPT as a working environment, not as the official repository of the case.

Keep a separate, securely backed-up master file containing the documents exactly as received or filed.

A simple structure might be:

1. 00 – Case dashboard and deadlines
2. 01 – Applications and responses
3. 02 – Sealed orders
4. 03 – Your statements and exhibits
5. 04 – Other parties’ evidence
6. 05 – Cafcass and professional reports
7. 06 – Medical, school, police and agency material
8. 07 – Communications and correspondence
9. 08 – Hearing documents and bundles
10. 09 – Legal research and verified authorities
11. 10 – Drafts and superseded documents

Use consistent filenames:

Do not overwrite filed documents.

Use clear version labels such as:

Start with a document audit — not with advocacy

Do not upload a collection of files and immediately ask:

First make sure the AI and you understand what the documents actually are.

Start with the foundation material:

1. the current application;
2. the latest sealed order;
3. earlier orders that remain relevant;
4. principal witness statements;
5. the most relevant current professional report; and
6. any hearing-specific direction or notice.

Then ask for an inventory.

Check the inventory yourself before using it for further analysis.

Create five working documents that control the case

A case becomes easier to manage when you stop treating it as a pile of documents and start treating it as a set of live questions.

Prompt for an evidence matrix

Prompt for a directions tracker

Use separate chats for separate jobs

One giant conversation eventually becomes difficult to audit.

Within the Project, use clearly named workstreams such as:

• 01 – Case dashboard and orders
• 02 – Document inventory
• 03 – Chronology
• 04 – Issues and evidence
• 05 – Legal research
• 06 – Witness statement
• 07 – Position statement
• 08 – Cafcass report analysis
• 09 – Correspondence
• 10 – Hearing preparation

This makes it easier to understand where a conclusion came from and to discard a workstream that has gone wrong without contaminating the rest of the case analysis.

Legal research: use AI to find the question, not invent the answer

Legal research is one of the highest-risk uses of generative AI.

AI can generate a plausible-looking case name, neutral citation, quotation and legal principle that do not belong together — or that do not exist at all.

The courts have repeatedly warned about fabricated and inaccurate authorities.

The safer workflow is:

1. identify the exact legal question;
2. ask AI to identify potential primary sources;
3. open the source yourself;
4. check that it exists;
5. read the relevant provision or paragraph;
6. check that it supports the proposition;
7. check whether it remains current; and
8. record the source and the date checked.

For a fuller verification framework, see the JSH Law Seven-Point Legal AI Trust Test .

Using AI with a witness statement

This requires particular care because a witness statement is evidence.

AI can help with:

• structure;
• numbered paragraphs;
• grammar;
• removing repetition;
• putting events into intelligible sequence; and
• identifying matters that need clarification.

It must not:

• add a fact you did not give it;
• create a precise date from an approximate memory;
• invent a quotation;
• turn suspicion into direct knowledge;
• strengthen an allegation;
• remove genuine uncertainty;
• diagnose another person;
• state another person’s motive as fact; or
• change the substance of your evidence.

The Civil Justice Council has specifically identified the risk that AI can reshape or embellish witness evidence in ways that are not immediately obvious.

Using ChatGPT for a position statement

This is often a more suitable AI-assisted task because a position statement is written submission rather than witness evidence.

AI can help you reduce long emotional notes into a clearer structure.

Give it:

• the type of hearing;
• the latest sealed order;
• the purpose of the hearing;
• the live issues;
• material developments;
• the evidence relied upon;
• the other party’s position stated fairly;
• the orders sought; and
• the applicable page limit.

You can then ask AI to test it rather than praise it:

Using AI to build a chronology

Chronology work is one of AI’s most useful applications — if you control the source material.

Then manually compare the chronology against the source documents.

Your working chronology can contain additional analytical columns. A chronology eventually prepared for the court must comply with the relevant directions and current Practice Direction 27A.

Using AI to analyse a Cafcass report

Do not prompt:

That framing invites confirmation rather than analysis.

A safer exercise is:

This gives you something much more useful than a general attack: a paragraph-by-paragraph map of the actual disagreement.

Using ChatGPT for hearing preparation

AI can be particularly useful in the final days before a hearing.

Ask it to help prepare:

• a one-page hearing plan;
• a short introduction;
• the three most important points;
• bundle references for those points;
• a numbered list of orders sought;
• likely questions from the judge;
• the strongest points against your position;
• questions for a witness where appropriate; and
• a practical attendance checklist.

Keep an offline copy of the documents and speaking notes you actually need.

The JSH Six-Question Check before asking AI to draft

Before generating a document, make sure you can answer six questions yourself.

If you cannot answer those questions, AI is likely to produce a polished version of an unclear case rather than solve the underlying problem.

Three checks every AI-assisted court document should pass

1. Factual and evidential check

• Is every important factual statement accurate?
• Is an allegation labelled as an allegation?
• Are findings distinguished from assertions?
• Can important propositions be traced to source material?
• Are names, dates, quotations and page references correct?
• Has AI changed the meaning or degree of certainty?

2. Legal and procedural check

• Does every cited authority exist?
• Have you opened and read the relevant part?
• Is the legislation, rule or Practice Direction current?
• Does the latest court order impose a specific requirement?
• Is the filing deadline correct?
• Is the page limit correct?
• Is any statement of truth or prescribed wording correct?

3. Strategic and human check

• Does the document answer the question the court must decide?
• Is it proportionate?
• Has irrelevant history been removed?
• Does it acknowledge material evidence that goes against your position?
• Is the order sought precise?
• Can you explain and defend every paragraph?
• Does the evidence still sound like your evidence?

Ten mistakes to avoid

1. Uploading everything before deciding whether it is necessary or appropriate.
2. Using one enormous chat for the whole case.
3. Asking AI to prove that you are right.
4. Allowing allegations to become facts.
5. Using AI-generated case citations without opening the judgment.
6. Letting AI strengthen witness evidence.
7. Relying on AI to calculate a critical deadline.
8. Replacing original evidence with an AI summary.
9. Using AI-generated labels or diagnoses as though they were established evidence.
10. Filing anything you have not personally read, checked and understood.

Used well, AI can improve access to justice

The value of AI for litigants in person is real.

It can make an incomprehensible order easier to understand. It can reveal that a file is missing an important document. It can turn five years of notes into a usable chronology. It can help a distressed parent replace angry language with a focused explanation of the child’s welfare.

It can also expose weaknesses before the judge does.

None of that requires pretending that AI is a lawyer or that its answers are infallible.

The safest model is human judgment supported by AI — not human judgment replaced by AI.

Using AI to prepare your case? Check the evidence before filing.

AI may help structure a chronology or first draft, but it cannot verify facts, authenticate exhibits or replace your responsibility for what you submit. JSH Law can help organise source documents and prepare a clear evidence-led case file.

Evidence and chronology support · Book a consultation