AI Agents & Family Court: A Guide for Litigants in Person
Artificial intelligence is moving beyond simply answering questions. The next generation of AI tools can increasingly work through documents, use digital tools, complete multi-step tasks and produce finished outputs. For an overwhelmed litigant in person trying to manage court orders, evidence, Cafcass reports, messages, deadlines and years of chronology, that could be enormously significant. But greater capability also creates greater risk. This is the first article in my new JSH Law series exploring what AI agents could mean for family justice — practically, legally and safely.
AI Agents and the Family Court: What Litigants in Person Need to Know
There is a significant difference between asking artificial intelligence to explain how to do something and giving an AI system the tools, information and instructions to actually carry out parts of the work. That distinction matters enormously for litigants in person. Used properly, the next generation of AI tools could help people organise evidence, manage documents, identify gaps, build chronologies and reduce the crushing administrative burden of running a family court case. Used badly, they can create false law, distort evidence, expose confidential information and make an already difficult case worse.
This article is the first in a new JSH Law series about AI agents, legal work and access to justice.
I am currently completing Mercor Academy’s Introduction to Agents certification. Rather than simply collecting another certificate, I want to turn what I am learning into something practical for people navigating the family courts without a solicitor. Each article in this series will take one core idea about AI agents and ask a more important question: what does this actually mean for a litigant in person?
The short version
A traditional AI chatbot mainly gives you information. An AI agent can potentially take a goal, work through several steps, use tools and documents, interact with a digital workspace and produce a finished output.
For litigants in person, that could eventually mean far more than asking, “What is a position statement?” It could mean asking a properly configured system to organise hundreds of documents, build a dated evidence index, identify missing records and prepare a structured chronology for you to review. But the human being must remain responsible for accuracy, evidence, confidentiality and what is ultimately put before the court.
1. First: what is an AI agent?
Most people first encountered generative AI through a chatbot. You type a question. The system gives you an answer.
That can be enormously useful, but you are still doing most of the work.
Suppose you have a family court hearing coming up and 600 pages of emails, school records, court orders, Cafcass documents and messages.
You might ask a chatbot:
The AI may explain what a chronology is, suggest headings and tell you how to organise one.
Helpful.
But you still have 600 pages to work through.
An agentic system is different.
In principle, if it has been given appropriate access to the documents and the right instructions, you could give it a goal such as:
The system would not merely tell you how to build a chronology. It would attempt to perform the workflow.
That is the important distinction.
“A chatbot tells you how to do the work. An agent can potentially carry out parts of the work.”
2. Why this distinction matters so much to litigants in person
Litigants in person are often expected to perform an extraordinary number of different roles at once.
You may be trying to:
- understand court orders;
- comply with deadlines;
- organise disclosure;
- read Cafcass material;
- prepare statements;
- find relevant evidence;
- manage hundreds or thousands of messages;
- prepare a chronology;
- identify breaches of orders;
- respond to correspondence;
- research law and procedure;
- prepare hearing documents;
- look after children;
- work;
- sleep;
- and somehow remain calm enough to explain the case coherently when you eventually reach court.
That burden becomes even harder where domestic abuse, coercive control, trauma, safeguarding concerns or post-separation abuse are involved.
The administrative load is not neutral.
Being required to repeatedly reread abusive messages, reconstruct years of events and constantly revisit distressing material can itself be exhausting. Important evidence can become buried simply because there is too much of it.
This is where responsible AI has enormous potential.
Not because an AI system should decide your case.
It should not.
But because technology may be able to take some of the mechanical burden away from the human being who is already carrying everything else.
The real access-to-justice opportunity
The most valuable use of AI in family litigation may not be asking it to “tell me what the law says”. It may be allowing people to spend less time sorting documents, copying dates, checking spreadsheets and rebuilding timelines — and more time understanding the actual issues in their case.
3. What could an AI agent actually help a litigant in person do?
The technology is developing quickly, and different systems have different capabilities. But it is already useful to think in terms of workflows rather than individual prompts.
Here are some examples.
Chronology building
Instead of manually copying dates from hundreds of documents, an agent could potentially extract dated events and organise them into a structured chronology.
The human then checks it.
Evidence indexing
An agent could identify documents, dates, senders, recipients and subject matter and create an evidence index.
That can be particularly helpful where the issue is not lack of evidence but finding the right evidence quickly.
Finding patterns
In some cases, individual incidents look insignificant when viewed alone.
A pattern may only become visible when events are placed together.
For example:
- repeated cancelled contact;
- changes in explanations over time;
- conditions repeatedly being added to communication;
- school or medical information not being shared;
- important decisions being made without consultation;
- contact repeatedly being interrupted around particular events;
- different accounts of the same event appearing in different documents.
AI can be useful in locating and grouping those events.
But identifying a pattern is not the same as proving what that pattern means.
That requires judgement, context and evidence.
Checking compliance with an order
A carefully directed system might compare the terms of a child arrangements order against a contact log and identify dates where the ordered arrangements appear not to have taken place.
Again, that does not automatically establish a legal breach. There may be explanations, later orders, agreements or contextual facts the system has not seen.
But it can help identify the events that need human examination.
Document preparation
AI can help convert disorganised notes into structured material, shorten repetitive writing, identify missing dates and improve readability.
The important word is help.
It should not silently invent the facts of your case.
Case management
An agent could potentially track deadlines, create task lists, identify documents required by an order and remind you when something has not yet been completed.
For many litigants, that kind of administrative support may be more valuable than asking AI to predict who will “win”.
4. But this is where we need to be very careful
The phrase “AI agent” can sound as though the sensible thing to do is hand your case to the machine and walk away.
That would be dangerous.
Family cases involve children, allegations, memory, disputed facts, safeguarding, credibility, domestic abuse, medical information and deeply private personal data.
A system capable of taking actions is potentially more useful than a chatbot.
It is also potentially more capable of making consequential mistakes.
Do not confuse automation with accuracy
An AI system can process information quickly and still be wrong.
It can create a beautifully formatted chronology containing the wrong date.
It can produce confident legal analysis based on a case that does not exist.
The fact that the output looks professional does not make it reliable.
5. AI can invent law
This is no longer a theoretical concern.
In Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin), the High Court dealt with serious problems involving false legal authorities being placed before the court in circumstances involving actual or suspected use of generative AI.
The wider message from the judgment is extremely important:
You remain responsible for material you rely upon.
AI tools can produce apparently convincing case names, quotations, citations and legal propositions that are simply wrong.
Courts are not required to forgive inaccurate material because “AI generated it”.
A litigant in person therefore needs a verification habit.
Before relying on an AI-generated legal authority
- Confirm that the case actually exists.
- Find the judgment from a reliable legal or official source.
- Read the relevant part yourself.
- Check that the quotation actually appears in the judgment.
- Check that the case says what the AI claims it says.
- Check whether there is newer authority.
- Do not cite a case simply because an AI tool gave you a neutral citation.
If you cannot verify it, do not put it before the court as authority.
6. Witness statements need particular care
This is an area where litigants in person should be especially cautious.
A witness statement is evidence.
It is not simply another piece of persuasive writing.
The Civil Justice Council has been examining the use of AI in the preparation of court documents and has specifically identified witness statements as an area requiring careful consideration. One of the concerns is obvious: AI can subtly reshape, embellish or restructure a person’s account in ways that may alter the authentic evidence.
That matters.
There is a legitimate difference between using technology to help organise what you want to say and allowing technology to manufacture what you supposedly remember.
A useful rule
Your statement must remain your evidence, your memory and your account. AI may help you organise it. It should not supply the facts.
If AI adds an event that you cannot personally recall, strengthens an allegation beyond what you actually remember, attributes a motive you did not witness, or changes “I think” into “I know”, that can fundamentally alter the evidence.
That is not harmless editing.
7. Be exceptionally careful with confidential family-court information
Family proceedings routinely contain some of the most sensitive information imaginable.
Children’s names, dates of birth, addresses, schools, medical records, safeguarding allegations, police material, domestic abuse disclosures and psychological information may all appear within the same bundle.
Do not assume that because an AI service gives you a box to upload something, it is therefore appropriate to upload it.
Before entering confidential or personal information into an AI system, understand:
- what service you are using;
- what happens to uploaded information;
- whether data is retained;
- whether it may be used to improve or train systems;
- who can access it;
- whether you can delete it;
- and whether uploading it is consistent with your legal and procedural obligations.
The judiciary’s own AI guidance specifically emphasises confidentiality and the danger of entering private information into public AI tools.
That warning should matter to litigants too.
Family cases contain children’s information
Do not treat children’s medical, school, safeguarding or private family information as ordinary data simply because an AI tool makes uploading documents easy. The sensitivity of the information does not disappear because the interface is convenient.
8. The real skill is not “prompting”. It is directing
One of the most useful ideas behind agentic AI is that the limiting factor begins to shift.
When you are doing everything manually, the bottleneck is often execution.
There are only so many documents you can read in an evening.
There are only so many dates you can copy into a spreadsheet.
There are only so many hours in the day.
Once technology can perform more of the repetitive execution, the important human skill becomes direction.
You need to know:
- what problem you are actually trying to solve;
- what evidence the system needs;
- what it is allowed to infer;
- what it must not infer;
- what output you need;
- how the work should be structured;
- how accuracy will be checked;
- and where human judgement must take over.
That is a much more sophisticated skill than typing:
In fact, that is a poor instruction because it does not define a reliable task.
A better use of AI is usually narrower, evidence-led and capable of being checked.
9. Think like a case manager, not a passenger
There is a useful analogy here.
A good manager does not simply hand an employee a pile of papers and say:
They explain the objective, provide the right documents, identify the constraints, specify the expected output and review what comes back.
AI agents require the same discipline.
If the system produces something poor, the question is not always:
Sometimes the better questions are:
- Did it have the correct documents?
- Was the instruction precise enough?
- Did I define the scope?
- Did I distinguish fact from allegation?
- Did I require source references?
- Did I tell it not to infer missing facts?
- Did I give it a reliable way to check its work?
Those are exactly the questions litigants should be learning to ask.
10. A practical example: from chaos to an evidence schedule
Imagine you believe the other parent has repeatedly failed to facilitate ordered telephone contact.
You have:
- a child arrangements order;
- WhatsApp messages;
- emails;
- a handwritten contact diary;
- Cafcass correspondence;
- and perhaps several years of material.
The wrong approach would be to ask AI:
That instruction starts with a conclusion and asks the technology to build a case around it.
That is exactly how confirmation bias can be amplified.
A much safer instruction would be:
That produces evidence which can then be analysed.
It does not ask the AI to decide the case.
That distinction — between extracting evidence and manufacturing conclusions — is critical.
Before asking AI to work on your family court case
- Define the precise task.
- Give it only the information needed for that task.
- Tell it to distinguish evidence, allegation and inference.
- Require source documents and dates wherever possible.
- Tell it not to invent missing information.
- Verify every important factual proposition.
- Independently verify every legal authority.
- Protect confidential and children’s information.
- Do not allow AI to create memories or evidence for a witness statement.
- Remember that you remain responsible for what you file or say in court.
11. Why I think this matters for access to justice
The discussion about AI and law is often framed around whether technology will replace lawyers.
I think that misses one of the most immediate questions.
What about the people who do not currently have a lawyer to replace?
Thousands of people enter the family courts trying to understand procedure, organise evidence and present profoundly important cases without continuous legal representation.
For them, even modest technological assistance can matter.
The Government is now openly exploring how responsible AI adoption could make legal services faster, more affordable and more accessible. Its 2026 Legal Services AI Growth Lab expressly identifies improved access to legal services and access to justice as potential benefits of legal AI.
That is important.
But access to AI is not the same thing as access to justice.
A tool is only useful if people understand its limits.
Giving an overwhelmed litigant access to a powerful AI system without teaching them how to verify it, protect their data, distinguish evidence from inference and recognise fabricated law does not solve the access-to-justice problem.
It may simply create a new one.
“The future is not simply litigants using more AI. It must be litigants learning how to use AI without surrendering judgement, evidence or responsibility.”
12. The human part becomes more important, not less
The more capable AI becomes, the more important it is to understand what only the human being can properly decide.
AI can help sort information.
It cannot experience your child’s behaviour.
It can identify apparently inconsistent accounts.
It cannot automatically determine why somebody gave them.
It can locate references to fear, contact or safeguarding within documents.
It cannot make the court’s welfare determination.
It can help you see your evidence more clearly.
It should not replace your judgement about what is important, fair, proportionate or child-focused.
And ultimately, judges make judicial decisions.
Technology should help us make the evidence easier to understand.
It should not pretend to be the judge.
Conclusion: the bottleneck is changing
The first major lesson from the development of AI agents is not really about technology.
It is about how work is organised.
Until now, many litigants in person have been limited by their own capacity to process information.
You may know that something important happened repeatedly, but not have the hours required to locate every example.
You may have the evidence but struggle to turn it into something a judge can absorb quickly.
You may spend so much time on administrative work that very little energy remains for strategy.
Agentic AI may begin to change that.
But the objective should never be to hand over responsibility for your case.
The objective is to use technology to reduce unnecessary labour so that more human attention can be directed towards the things that matter: the evidence, the children, the safeguarding issues, the law, the choices being made and the outcome being sought.
That is where I believe AI has real potential in access to justice.
Not replacing the human being.
Giving the human being more capacity to think.
Coming next in this series
In Part 2, I will look more closely at what actually makes an AI system an “agent”, the idea of an agentic spectrum, and how to work out whether a tool is merely answering you or genuinely acting on your behalf.
Sources and further reading
- Courts and Tribunals Judiciary, Artificial Intelligence (AI) – Judicial Guidance, updated October 2025.
- Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin).
- Civil Justice Council, Use of AI in Preparing Court Documents, consultation update, June 2026.
- Solicitors Regulation Authority, Misuse of AI – Warning Notice, 17 August 2026.
- UK Government, Advisory AI Growth Lab: Legal Services, August 2026.
- Mercor Academy, Introduction to Agents, studied September 2026.
Need help turning a difficult family court case into something clear and manageable?
JSH Law provides practical, evidence-led support for litigants in person navigating family proceedings.
I can help you organise evidence, build chronologies and schedules, review Cafcass material, prepare statements and hearing documents, identify gaps and contradictions, structure appeal or application paperwork and make an overwhelming case easier to understand.
The purpose is not to manufacture a case or tell you what you want to hear. It is to help you identify the evidence, understand the issues and present your position clearly, calmly and effectively.
You can book a 15-minute initial telephone consultation below.
The initial consultation is an opportunity to identify the issue, understand what support may be useful and discuss next steps. It is not legal advice and does not create a solicitor-client relationship.
Follow JSH Law
Follow JSH Law for practical family court guidance, evidence-led case preparation, safeguarding commentary, access-to-justice analysis and the rest of this AI & Family Justice series.

© 2026 JSH Law Ltd. All rights reserved.


© 2026 JSH Law Ltd. All rights reserved.
© 2026 JSH Law Ltd. All rights reserved.
© 2026 JSH Law Ltd. All rights reserved.


