For somebody considering divorce, asking an AI chatbot a question can be much easier than telephoning a solicitor. It is available at any time, there is no charge each time another question occurs to you, and you can describe a complicated situation and receive an apparently clear answer within seconds.
Many of the questions put to AI are probably much the same as those asked during a first appointment with a family solicitor. Will everything be divided equally? Can one parent stop the other seeing the children? Does inherited property have to be shared? How is child maintenance calculated? What happens to a pension, a business or a house that one person owned before the marriage?
People are also using AI for more than straightforward questions. They may ask it what is likely to happen in their case, paste in a letter from their former partner or solicitor, or use it to draft a reply.
The difficulty is that family law is highly dependent on individual circumstances. A fact that appears unimportant to the person asking the question can sometimes change the advice considerably.
A fluent answer is not necessarily a reliable one
Generative AI is very good at producing convincing language. That is not the same as exercising legal judgement.
This distinction is easy to overlook because the answer itself may look authoritative. It can be clearly written, logically arranged and delivered with complete confidence, even when something within it is wrong.
One well-known problem is the AI ‘hallucination’. An AI system can invent a case, quotation or legal principle. It can also get something wrong in a less obvious way, such as referring to a genuine case but misrepresenting what it decided.
There is a further problem with information becoming outdated. Family law procedure, legislation and guidance change. Unless the AI being used has access to reliable current information, there is no guarantee that the position it describes is still the position today.
Answers can also shift during the course of a conversation. Ask a short question and the answer might be a definite yes. Add another fact and it becomes possibly. Rephrase the original question and the conclusion may change again.
Sometimes that change is entirely justified because the additional fact matters. The user, however, may have no way of knowing whether the answer changed for a sound legal reason or simply because the chatbot interpreted the new prompt differently.
The hidden facts a chatbot may never uncover
This is perhaps a bigger problem than an obviously incorrect answer. Part of a family solicitor’s job is working out what matters. Clients do not necessarily know that themselves, particularly when they are dealing with divorce or separation for the first time.
Someone might ask, for example, “Can I leave the family home?” A solicitor asking further questions could discover that there is domestic abuse, financial control, mortgage arrears or an immediate concern involving a child. The conversation may therefore need to go in a completely different direction.
AI can ask follow-up questions, but ultimately it is working with the information given to it. If the person asking the question does not realise that a particular fact is important, there is a good chance that fact will never make it into the conversation.
That matters particularly where somebody or their children may be at risk.
Jurisdiction may be an issue
Another danger is receiving an answer that is broadly correct law, but for the wrong country. England and Wales, Scotland and Northern Ireland have distinct legal systems. Online material also contains a huge amount of information about family law in countries such as the United States, Canada and Australia. Terminology and legal concepts can therefore become mixed together in an AI response.
International families present an even greater challenge. A couple might have married abroad, own property in several countries or currently live in different jurisdictions. There may already be proceedings taking place elsewhere.
In those circumstances, seemingly basic details can be important. Where does each person live? Where are they domiciled? Where was an existing order made? Have proceedings already started?
A general answer about divorce law is of limited use if the first question should really have been which court has jurisdiction to deal with the case.
Using AI to write letters and legal documents
Drafting is an obvious use for AI and, for relatively routine correspondence, it can be useful. Someone can provide a collection of notes and ask for them to be turned into a clear email. AI can also remove repetition or help a person explain something they are struggling to put into words.
Obviously a letter can be beautifully written and still be a bad letter to send. It might concede a point unnecessarily, make an allegation stronger than the evidence supports or turn a relatively minor disagreement into a more confrontational exchange. The person using AI may concentrate on whether the letter sounds professional rather than what it actually says.
The stakes become higher with court forms, witness statements and proposed orders.
For example, an AI-generated witness statement might turn an assumption into a fact or tidy away an apparent inconsistency that actually needs explaining. It may also produce language that the person signing the statement would never normally use.
There are different problems with orders. An apparently sensible financial order might fail to deal properly with the mechanics of selling a property, implementing a pension arrangement or dismissing future financial claims. An agreement between two former spouses also does not necessarily achieve financial finality simply because it has been written down in formal-looking language.
Ultimately, the person signing or filing a document remains responsible for it. An invented case or inaccurate quotation does not become acceptable because it came from an AI system. Where a document could affect someone’s home, pension, finances or relationship with their children, professional review before it is signed or sent can be particularly important.
Privacy, confidentiality and the family’s digital footprint
There is another issue that is easily forgotten when using a chatbot: the amount of personal information involved in family law. A person might be tempted to upload bank statements, correspondence from their former partner, medical information or documents containing details about their children. Some of that information relates not just to the person using the AI service, but to other people who have not chosen to share it.
Users should not assume that telling a chatbot something is equivalent to telling their solicitor. Communications with a legal adviser may attract legal professional privilege. A conversation with a general AI service does not acquire the same protection simply because the subject being discussed is a legal problem.
Removing names and obvious identifying information can reduce some of the risk, although it does not necessarily make a document anonymous. An unusual combination of facts can sometimes identify the people concerned.
Before uploading documents from a family case, it is worth asking whether the AI actually needs to see them at all. If it does, the user should understand the privacy terms and controls of the particular service being used.
Where AI can still be genuinely useful
None of this means that AI has no place in helping somebody understand a family law problem. There is a considerable difference between using it as a source of information and relying on it to decide what you should do.
AI can be useful for explaining unfamiliar terminology, helping somebody put events into chronological order or suggesting questions to take to a solicitor. Someone preparing for an initial appointment could use it to organise a list of assets or turn several pages of notes into a shorter chronology, provided they then check that nothing important has been altered or omitted.
It can also provide a starting point for understanding subjects such as mediation, the general divorce process or pension sharing. Information that matters to a particular case can then be checked against an authoritative source or discussed with a solicitor.
Perhaps the most useful role for AI is therefore preparation rather than advice. It can help somebody arrive at a legal appointment with their thoughts organised and their questions clearer.
What it cannot reliably do is know which question the person has forgotten to ask.