AI in Family Practice: useful assistance, continuing responsibility

AI in Family Practice: useful assistance, continuing responsibility

A familiar workload, a new assistant

Consider a familiar situation. A hearing is approaching, and the bundle contains school records, medical reports, and months of correspondence. An AI tool can produce a chronology in minutes, potentially freeing up valuable time for advice and preparation. That benefit, however, depends on the chronology being accurate, preserving important details, distinguishing allegations from findings and protecting the family's confidential information.

Similar opportunities arise throughout family practice, from organising financial disclosure and preparing correspondence to explaining court orders to clients. Yet the speed with which AI produces a document is not necessarily a measure of its efficiency. Time spent checking, correcting and supervising the output must also be taken into account.

The question for solicitors is therefore not simply whether to embrace or avoid AI, but how to identify the tasks for which it can provide useful and reliable assistance.

Know what the tool is doing

Generative AI creates text, images and other content in response to instructions. Although its responses may appear authoritative, they can contain invented facts, inaccurate statements or fictitious legal authorities. This differs from conventional automation, such as spreadsheet formulas and document templates, which operates according to defined rules. Such systems are not infallible, but their limitations are generally different.

It is important to distinguish between the functions a tool is being asked to perform. Extracting information, carrying out calculations, drafting documents and predicting outcomes each present different risks. A system capable of identifying figures in bank statements may not calculate them accurately, while an accurate calculation does not, of itself, establish what constitutes a fair financial settlement. For arithmetic, a properly tested spreadsheet may be more appropriate than a general-purpose chatbot.

Professional guidance is developing alongside these technologies. The Law Society and Bar Council have published guidance, while the BSB and CILEx Regulation have explained how existing regulatory obligations apply to AI. For solicitors, the SRA Code of Conduct and its August 2026 warning notice are particularly relevant. The underlying principle remains unchanged: professional responsibility does not diminish because technology has been used to undertake part of the work.

Research and drafting still need judgment

AI can assist with identifying lines of research, organising submissions and preparing initial drafts. Its ability to produce convincing legal analysis nevertheless presents a significant risk. A generated response may refer to a case that does not exist or attribute an incorrect proposition to a genuine judgment. Even where a citation or quotation appears credible, it must be checked against the original authority.

The courts have already considered the consequences of such errors. In Re D (A Child) (Recusal), the mother admitted using AI to prepare a document containing erroneous authorities. The Court of Appeal accepted that she had not intended to mislead the court but emphasised that both represented and unrepresented parties must ensure that the authorities they cite are genuine and support the propositions advanced.

The Divisional Court's judgment in Ayinde, although arising outside family proceedings, reinforces the importance of verification. Legal professionals remain responsible for checking AI-assisted research against authoritative sources, including where that research has been provided by someone else. Simply instructing an AI system to be more careful is no substitute for independent checking.

Particular care is needed when AI is used in preparing witness evidence. In M v F, a maternal uncle admitted using AI to assist with his statement. The judge expressed uncertainty about how the statement had been prepared and whether it reflected the witness's independent recollection. The concern was not that AI involvement necessarily rendered the evidence untrue, but that its reliability and provenance were unclear.

These decisions point towards a practical distinction between assistance with drafting and responsibility for content. Material factual assertions should be checked with the client or witness, uncertainties should be preserved rather than resolved through assumption, and the person signing a statement must understand and adopt its contents. When preparing questionnaires or orders, practitioners should work from current procedural requirements and established templates, using AI to assist with presentation or wording rather than to determine the applicable procedure.

Keep the detail that matters

Financial remedies illustrate both the potential benefits and the limitations of AI.

At the outset of a matter, AI may help structure questions about assets, liabilities, income and immediate financial needs. It cannot, however, replace the discussion necessary to establish a client's actual circumstances. In particular, financial information may need to be considered in the context of restricted access to money, unequal knowledge of family finances or possible economic abuse.

During disclosure, appropriate extraction tools may assist by identifying missing documents, inconsistent dates or figures requiring further explanation. The value of that assistance depends on what happens next.

Take two hypothetical examples. In the first, an approved tool identifies two missing months in a set of bank statements. The solicitor checks the original documents and makes a focused request for the outstanding information. The tool has improved the organisation of the exercise, but its finding is not evidence of concealment. In the second, a tool identifies a £30,000 transfer as unexplained expenditure, while also including the balance of the receiving account in the asset schedule. By tracing the corresponding credit, the solicitor identifies and corrects the potential double counting before any allegation is advanced. What initially appeared to be an anomaly was neither a financial loss nor evidence of dishonesty.

These examples demonstrate why AI-generated schedules require careful scrutiny. Ownership, valuation dates, liabilities and disputed figures must all be checked. Company assets should not automatically be added to the value of a shareholding, nor should pension benefits be treated as equivalent to immediately accessible capital. The Financial Remedies Guide 2026 emphasises focused questionnaires and composite schedules that identify areas of disagreement. Producing more documentation does not necessarily amount to better preparation.

AI may also assist with settlement modelling, particularly where alternative outcomes are presented against clearly stated assumptions. Such exercises should account for liquidity, borrowing capacity, tax consequences and transaction costs, with calculations and formulas independently checked. A model can illustrate the financial consequences of different proposals; it cannot determine the outcome of the court's discretionary assessment. Apparent numerical precision should not be mistaken for reliable legal advice.

The need to preserve context is equally important in children proceedings. For example, a summary that treats incidents of coercive or controlling behaviour as isolated events may fail to convey a wider pattern of conduct. A concise account can therefore be misleading even where individual details appear correct.

Before relying on a generated summary, practitioners should establish which documents have actually been processed and whether any scanned material, tables or attachments were omitted or unreadable. References should lead back to the relevant source passages, rather than to another AI-generated account. Allegations, admissions and judicial findings must remain clearly distinguished.

Care is also required to identify potential bias. AI-generated descriptions may reproduce assumptions about gender, disability or parenting, even when expressed in apparently neutral terms. Reviewing the evidence, identifying omissions and testing the assumptions underlying a summary remain essential safeguards. CILEx Regulation's guidance expressly addresses fairness and bias.

Protect the family's information

Family proceedings frequently involve highly sensitive information. Financial records may reveal relationships, health expenditure or a person's whereabouts. Children's records may contain medical diagnoses, allegations of abuse or addresses that must remain confidential.

Before entering such material into an AI system, practitioners need to understand how the provider handles both inputs and outputs. Relevant considerations include who may access the information, where it is processed, how long it is retained, whether it may be used for training and what deletion arrangements apply. Connected services and the configuration of the particular account also require examination. A paid subscription or an 'enterprise' description does not, without further enquiry, establish that these safeguards are adequate.

Confidentiality, legal professional privilege, data protection and restrictions on the communication of information from family proceedings are distinct obligations. Obtaining a client's consent does not necessarily satisfy them all. A parent cannot simply authorise disclosure contrary to a court restriction or consent on behalf of every other person whose information appears in the documents. The Family Procedure Rules impose specific restrictions on the communication of information from children proceedings and on the subsequent use of disclosed documents.

Financial dispute resolution (FDR) material also requires particular care. The rules provide for without-prejudice offers within the FDR process, without making otherwise inadmissible material admissible, and Practice Direction 9A emphasises the importance of non-disclosure. Practitioners must ensure that protected negotiations are not inadvertently introduced into open documents through the use of AI or related workflows.

Data protection obligations require separate consideration. Processing must have an appropriate lawful basis and, where health information or other special-category data is involved, an additional condition must be satisfied. Provider roles, security measures and international transfers need to be assessed, and a data protection impact assessment undertaken where processing is likely to result in high risk. Removing names may not be sufficient to anonymise information where the remaining details still identify the family.

Make the service clearer—and measure the saving

One of AI's more promising applications is helping practitioners communicate complex information in a way that clients can understand.

In Re B, the judge used the secure Judicial Copilot system to prepare simplified summaries of a judgment for parents with learning and cognitive difficulties, with the agreement of their advocates. The case illustrates how carefully supervised AI assistance may improve accessibility. It should not, however, be taken as an endorsement of using any general-purpose chatbot to explain judgments without appropriate safeguards.

In financial proceedings, a checked comparison of settlement options may help a client understand, for example, the difference between retaining the family home and having sufficient accessible funds to meet everyday living costs. The value lies not merely in simplifying the explanation, but in helping the client make an informed decision. Assumptions, conditions, uncertainties and practical difficulties must remain clear, and practitioners should check the client's understanding rather than assume that a simplified document has achieved its purpose.

Solicitors should also recognise that clients and litigants in person may already be using AI themselves. Where a document appears to have been prepared with AI assistance, it is generally preferable to ask neutrally about its preparation rather than draw conclusions from its style. Polished wording is not, by itself, evidence of dishonesty. In G v K, the judge recognised that enquiries made through ChatGPT about reporting abuse were compatible with the behaviour of a genuine victim seeking help.

Similar caution is appropriate when considering audio and other digital evidence. Original recordings should be preserved, with a clear distinction maintained between the source material and any transcripts, enhancements or synthetic reconstructions. Where manipulation is suspected, the proper response is to investigate provenance and reliability rather than assume that inconvenient evidence has been fabricated.

AI may also improve efficiency in routine administration, although checks on recipients, deadlines and secure methods of communication remain necessary. Any assessment of time saved should include reviewing and correcting outputs, not simply the time taken to generate them. Material AI involvement should be explained where appropriate

There will be tasks for which checking an AI-generated result takes longer than completing the work directly. For more substantial exercises, recording review time and recurring errors will provide a more reliable indication of whether the technology delivers genuine savings.

Where clients arrive with AI-generated research or advice, practitioners should explain its limitations without dismissing the effort the client has made to understand their position. The material may be inaccurate, but the questions it raises may still warrant careful consideration.

Before the work leaves the office

AI can assist family practitioners with research, organisation, drafting and communication, but its usefulness ultimately depends on the safeguards surrounding its use. Before relying on or circulating AI-assisted work, a short final review may help ensure that the relevant professional responsibilities have been met.

Suggested checks:

• Is the tool appropriate for the task and the information being processed?
• Have material facts, figures and legal authorities been verified against their original sources?
• Have omissions, assumptions, vulnerability and disputed evidence received appropriate professional scrutiny?
• Have confidentiality, restrictions on use and the client's understanding been properly addressed?
• Can the responsible solicitor explain, justify and stand behind the finished work?

The objective is not to remove professional judgment from the process, but to use technology in ways that support it. AI may make some aspects of family practice quicker and more accessible. The practitioner responsibility for the quality, accuracy and integrity of the work remains.

This article was developed with the assistance of ChatGPT through a structured sequence of prompts. All materials referred to have been independently verified by the author. Was it quicker than writing it myself? I’m not really sure…

1. ^

   https://www.lawsociety.org.uk/topics/ai-and-lawtech/generative-ai-the-essentials

2. ^

   https://www.barcouncilethics.co.uk/wp-content/uploads/2024/01/Considerations-when-using-ChatGPT-and-generative-artificial-intelligence-Nov-2025.pdf

3. ^

   https://www.barstandardsboard.org.uk/static/5e1baa5c-614c-4105-afae9a9ccdf8d97b/c0d1ed62-fcad-4a68-8ab810de5fa9b116/Artificial-Intelligence-Guidance-May-2026.pdf

4. ^

   https://cilexregulation.org.uk/2026/10/01/new-crl-guidance-ai-use-in-legal-services/

5. ^

   https://www.sra.org.uk/solicitors/guidance/misuse-ai/

6. ^

   [2025] EWCA Civ 1570 [28][83]

7. ^

   [2025] EWHC 1383 (Admin) [6][9][87]

8. ^

   [2026] EWFC 22 (B) [26 – 27] 

9. ^

   https://www.judiciary.uk/wp-content/uploads/2026/03/FRC-Guide-Final-Clean.pdf

10. ^

    Ibid. 4

11. ^

    https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/artificial-intelligence/guidance-on-ai-and-data-protection/how-do-we-ensure-lawfulness-in-ai/

12. ^

    [2026] EWFC 107 (B)

13. ^

    [2025] EWHC 2961 (Fam)