The Legal AI Trust Test: Seven Checks Before You Rely on ChatGPT or Any AI in Court
Published: 31 July 2026 | Last reviewed: 31 July 2026
Using ChatGPT in court can help a litigant in person organise evidence, understand procedure and prepare clearer documents. But an AI-generated answer can appear authoritative while containing invented cases, inaccurate law or distorted facts. Before relying on any AI-assisted statement, chronology or submission, you need a reliable method for checking its accuracy, sources, confidentiality and evidential integrity.

Legal AI | Litigants in Person | Access to Justice
The Legal AI Trust Test: Seven Checks Before You Rely on ChatGPT or Any AI in Court
Artificial intelligence can help a litigant in person organise evidence, understand procedure and turn a disordered mass of documents into a clearer case. But a polished answer is not necessarily a correct answer—and the person whose name appears on the court document remains responsible for it.
If a national law firm will not trust legal AI without authoritative content, clear citations and secure handling of client material, why should a litigant in person—whose home, finances or relationship with a child may depend upon one hearing—accept anything less?
The legal profession is moving quickly towards AI-assisted research, drafting and document review. Used properly, these tools have the potential to improve access to justice. They can help people understand unfamiliar terminology, prepare chronologies, identify gaps in their evidence and express themselves more clearly.
That matters. Many litigants in person arrive at court with no lawyer, limited time and hundreds—sometimes thousands—of pages of messages, emails, orders, reports and allegations to process.
But the same technology can also invent authorities, apply the law of the wrong country, confuse an allegation with a judicial finding, omit an important fact or expose highly sensitive information. In family proceedings, those are not minor drafting errors. They can affect findings about domestic abuse, safeguarding, credibility and a child’s welfare.
Why trust—not speed—is the real issue
A July 2026 LexisNexis case study describes how Irwin Mitchell approached the introduction of generative AI across its practice. The firm identified three requirements that any legal AI system had to meet:
- wide legal coverage;
- clear citations; and
- a secure setting suitable for client material.
Crucially, the lawyers did not treat the presence of citations as proof that an answer was correct. Their process involved opening the cited sources, reading them and checking the legal position before deciding what to do next.
That distinction matters. A citation can look convincing and still be fictitious, irrelevant or incapable of supporting the proposition for which it has been used.
The question raised by the Law.com article “When Review Becomes Ratification” is therefore an important one: is a human genuinely reviewing the work, or merely approving a polished answer because it looks plausible?
Human involvement is not a safeguard if the human does not independently test the answer.
The access-to-justice opportunity is nevertheless real. Legal Futures reported on 31 July 2026 that more than 3,800 people had signed up to the CaseCraft AI small-claims platform, with claims approaching £11 million in total value. Significantly, the service reportedly uses a team of legal reviewers and solicitors to review AI-generated outputs.
This suggests a more credible direction for public-facing legal AI: technology to improve speed and accessibility, supported by genuine human judgment and visible quality controls.
The seven-point Legal AI Trust Test
Before relying on an AI-generated answer in a statement, position statement, skeleton argument, application or submission, work through all seven checks.
1. Has the AI used the correct jurisdiction?
“UK law” is not sufficiently precise. England and Wales, Scotland and Northern Ireland have separate legal systems. A tool may also produce American, Canadian or Australian law because similar legal terminology appears in those jurisdictions.
In family proceedings, the answer may also differ depending upon whether the case concerns private children law, public law proceedings, financial remedies, domestic abuse, divorce or enforcement.
2. Does it identify an authoritative source?
A legal answer should lead you back to something capable of being checked: legislation, a procedural rule, practice direction, reported judgment or official guidance.
A list of case names is not enough. Ask for the neutral citation, the court, the date and a direct link to the judgment. For procedural points, check the current Family Procedure Rules, Civil Procedure Rules or tribunal rules that actually govern your case.
3. Does the source actually support the proposition?
This is where superficial checking often fails. A case may be real but have nothing to do with the point being made. The AI may have misunderstood the judgment, taken words out of context or treated a party’s argument as the court’s decision.
Open the judgment and find the relevant paragraph. Read enough of the surrounding reasoning to understand what the court decided and why.
4. Is the law current and still applicable?
A genuine judgment can still be unsafe to rely upon. It may have been overturned on appeal, distinguished in later cases or affected by new legislation, amended rules or a revised practice direction.
Family procedure changes regularly. Forms, filing requirements, court guidance and legal-aid rules can also change without the AI’s underlying training data immediately reflecting the new position.
5. Has it separated evidence, allegation, inference and finding?
This is essential in family cases. An AI-generated chronology or statement may turn:
- an allegation into an established fact;
- a suspicion into a conclusion;
- something reported by a child into a verbatim quotation;
- a professional’s observation into a diagnosis;
- an interim judicial comment into a final finding; or
- a disputed incident into “coercive control” without analysing the wider pattern and evidence.
The court needs to know what happened, who says it happened, what evidence supports it, whether it is disputed and whether any finding has already been made.
6. Is the information being handled securely?
Do not assume that a public or consumer AI service is confidential merely because you are using it through a private account. Check the provider’s current terms, privacy controls, retention arrangements and whether uploaded information may be used to operate or improve its systems.
In UK v Secretary of State for the Home Department [2026] UKUT 81 (IAC), the Upper Tribunal gave a serious warning about placing confidential documents into open-source AI tools. In April 2026, the Chancellor of the High Court stressed that confidentiality cannot simply be assumed when public AI systems are used.
This is particularly important in family proceedings, where documents can contain:
- children’s names, dates of birth and addresses;
- medical, school and social-care records;
- domestic-abuse allegations;
- police disclosure;
- Cafcass or local-authority material;
- financial information; and
- information about people who have never consented to its use.
7. Has there been meaningful human review?
Meaningful review is not reading the document once and deciding that it sounds professional. It means checking the draft against the evidence, opening every authority and deciding independently whether each significant proposition is accurate, necessary and fair.
It should also include a strategic question: does the document actually help the court decide the issues, or has AI merely produced a longer and more combative version of the case?
Why family proceedings require particular care
A small money claim may turn largely upon a contract, an invoice and whether payment was made. Family proceedings can require the court to evaluate relationships, patterns of behaviour, risk, credibility, trauma, a child’s wishes and feelings and the welfare consequences of competing proposals.
AI can organise the material relevant to those questions. It cannot safely decide which parent is telling the truth, diagnose coercive control or determine what arrangements are in a child’s best interests.
It may miss the significance of conduct that appears minor in isolation but becomes important when seen as part of a repeated pattern. It may also exaggerate isolated disagreements into allegations of abuse because it has been prompted to “make the strongest possible case”.
The JSH Law view
AI should help a litigant make the court’s job easier. It should not be used to manufacture certainty, intensify allegations or replace the careful welfare analysis the case requires.
A practical example
Suppose a parent asks an AI tool to prepare a position statement from several years of messages. The resulting draft says:
“The father has persistently alienated the child from the mother.”
That sentence may be unsafe. “Parental alienation” is not established merely because contact has reduced or a child is resistant. The draft may have ignored other possible explanations, failed to distinguish fact from submission and used a contested label instead of identifying the actual behaviour and its welfare impact.
A more evidence-led formulation might be:
“The mother is concerned that the child’s relationship with her has deteriorated following the repeated cancellation of contact on the dates set out in the attached schedule. She asks the court to consider the reasons for that deterioration, the impact upon the child and whether further assessment or protective directions are required.”
The second version identifies the concern without presenting a disputed conclusion as a proven fact.
What meaningful human review actually involves
If AI has helped prepare an important court document, the final review should be structured and recorded. The following process is a sensible minimum.
- Read the original documents. Do not review only the AI’s summary of them.
- Verify every date. Check dates against orders, messages, emails and official records.
- Check every quotation. Confirm it is exact, complete and not taken out of context.
- Open every authority. Confirm the case or rule exists and supports the stated proposition.
- Mark disputed facts. Separate allegations from admissions and judicial findings.
- Identify AI inferences. Remove conclusions that do not have a clear evidential foundation.
- Check proportionality. Remove repetition, rhetoric and material that does not help determine the issues.
- Restore the litigant’s own evidence. A witness statement must remain the witness’s truthful account—not the AI’s preferred narrative.
- Keep a verification note. Record the sources checked and material corrections made.
If the reviewer begins with the assumption that the AI draft is probably right, review can quietly become ratification. The correct starting point is that every important factual and legal proposition must earn its place in the final document.
Safer and higher-risk uses of legal AI
| Potentially useful with checks | Higher-risk use requiring particular caution |
|---|---|
| Explaining legal terminology in plain English | Giving a definitive opinion on the likely outcome |
| Creating a chronology template | Selecting which incidents are true or most important without human checking |
| Identifying missing dates or document categories | Inferring events that are not recorded in the evidence |
| Improving headings, grammar and readability | Rewriting a witness’s evidence until it no longer sounds or reads as their account |
| Suggesting search terms for legal research | Relying on an AI-generated case summary without reading the judgment |
| Testing whether a document is easy to follow | Allowing AI to determine welfare, credibility or safeguarding conclusions |
Using AI to prepare a court document
Current judicial guidance recognises that AI chatbots are being used by unrepresented litigants and may be the only assistance some people receive. It also recognises the access-to-justice value of clearer and more coherent documents.
However, the guidance makes the underlying responsibility clear: a litigant remains responsible for material put before the court. If AI appears to have been used, a judge may ask what checks were undertaken.
There is not currently a universal rule requiring every responsible use of AI to be declared. The position is context-dependent. If asked, a litigant should answer honestly and be able to explain how the document was checked.
Particular care is required where a document contains a statement of truth. Under Part 17 of the Family Procedure Rules , a statement of truth verifies the factual material in the document. AI cannot hold the required belief and cannot accept responsibility for an inaccurate statement. The person signing must understand and honestly believe the facts being verified.
Before filing: the 60-second final check
- Is every important date correct?
- Is every quotation exact?
- Does every cited case exist?
- Have I read the relevant part of every authority?
- Have I labelled disputed allegations fairly?
- Is the document in my own voice and consistent with my evidence?
- Can I explain how the final document was checked?
Trusted legal AI must not become a privilege reserved for large firms
Large firms can purchase specialist systems, legal databases, secure document environments, training and professional supervision. Litigants in person are more likely to rely upon a free public tool while dealing with the most serious dispute of their lives.
That creates a new access-to-justice divide. It is not enough to tell unrepresented people that they may use AI and then leave them to discover hallucinations, confidentiality risks and evidential distinctions by trial and error.
The answer is not to ban AI. Used carefully, it can make legal information more accessible and help people present their cases more coherently. The answer is to build a trusted pathway:
- AI for organisation, explanation and first drafts;
- authoritative sources for the law;
- secure systems for sensitive documents; and
- genuine human judgment before anything is relied upon.
Trusted legal AI is not the chatbot that produces the smoothest answer. It is a verifiable process involving sources, citations, privacy, jurisdiction, evidence, an audit trail and a human being willing to take responsibility for the result.
Frequently asked questions
Can I use ChatGPT to help prepare family-court documents?
AI can assist with structure, plain-English explanation, chronology templates and early drafting. The final document must be checked against the original evidence and applicable law. Do not use it to invent facts, determine credibility or replace welfare and safeguarding analysis.
Do I have to tell the court that I used AI?
There is no universal disclosure requirement applying to every responsible use. The answer may depend upon the context and any directions made in your case. If the judge asks whether AI was used or what checks were undertaken, answer honestly and explain your verification process.
Can AI find cases for me?
It can suggest cases and research terms, but never assume a citation is genuine. Locate the judgment through an official or recognised legal source, read the relevant paragraphs and check whether the decision remains authoritative.
Can I upload my court bundle?
Do not upload an unredacted bundle to a public or unverified tool. A bundle may contain confidential and highly sensitive information about children and third parties. Check the tool’s security, contractual protections, retention arrangements and privacy settings before processing any case material.
Can AI prepare a chronology?
Yes, but every entry must be checked against the source document. Look specifically for omitted incidents, duplicated events, incorrect dates, altered quotations and AI-generated interpretations presented as facts.
Can AI write my witness statement?
It can help organise a draft, but the evidence must remain yours. You must understand, adopt and honestly believe the factual contents. Remove language that exaggerates your evidence, attributes motives you cannot prove or no longer reflects how you would give your account to the court.
Have you used AI to prepare an important court document?
JSH Law can help you turn AI-assisted work into a clear, evidence-led and court-appropriate document. Support can include checking chronologies against source documents, distinguishing allegation from finding, reviewing statements and position statements, organising evidence, checking cited authorities and preparing for a hearing.
Sources and further reading
- LexisNexis: Irwin Mitchell and Lexis+ with Protégé
- Ayinde v London Borough of Haringey; Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin)
- Artificial Intelligence: Judicial Guidance, October 2025
- Civil Justice Council: Use of AI for Preparing Court Documents
- Chancellor of the High Court: Legal Professional Privilege in the Age of AI
- Legal Futures: Small-claims platform aims to expand
About the author
Jessica Susan Hill writes about family justice, domestic abuse, safeguarding, evidence, access to justice and the responsible use of legal technology. Her work focuses on helping litigants in person understand the legal process and present their cases clearly, fairly and effectively.







