Why can’t I just use an LLM like ChatGPT, Claude or Gemini to get my own results?

Family law in England and Wales isn’t a fixed formula. It is shaped by decades of decided cases that tell a court how to apply Section 25 of the Matrimonial Causes Act 1973 to real facts, and those cases do not all carry equal weight. Some are binding, others merely persuasive; some are still good law, others have been overtaken; a few are directly on point, most are not.

A general chatbot has no dependable, current view of which authorities are the leading ones for your situation. It may reach for a case it happens to remember, cite one that has since been distinguished or overruled, or lean on a decision that simply isn’t relevant to your facts. It can sound authoritative while pointing at the wrong law entirely. Knowing which precedent a court would actually give weight to is a large part of the skill, and it is precisely what a general model cannot reliably do.

A fair assessment under Section 25 weighs many factors together: needs, the sharing principle, the welfare of any children, pensions, the length of the marriage, and each party’s resources and contributions. Those factors interact, and leaving one out can change the whole picture.

In a free-form chat, you supply whatever comes to mind and the model answers on that basis. If you mention the house but forget the pensions, or set out your own contributions but not your spouse’s needs, the answer tilts accordingly. It will not reliably notice what is missing or ask for it. The real risk is not a wrong sum; it is a confident answer built on a partial, imbalanced version of your situation, with no signal that anything important was left out.

General models are trained up to a point in time and then frozen. Case law is not. The way courts approach pensions, conduct, and the sharing of assets built up before or after a marriage continues to develop. A chatbot working from a training snapshot can be quietly out of date, and it will not tell you so. SmartJudge instead works from the official archive of judgments published by The National Archives, refreshed regularly, so the reasoning reflects the law as it stands rather than as it once was.

The hardest part is that a general model’s mistakes don’t look like mistakes. It can invent a case name, misremember a statutory provision, or apply a principle to the wrong facts, all in fluent, assured prose. Nothing in an ordinary chat checks whether a cited case actually exists or says what it is claimed to say.

SmartJudge runs a citation guardrail that checks case references against a verified list before they appear in your report. That doesn’t make it infallible, but it removes one of the most dangerous failure modes of using a chatbot cold.

To be clear, SmartJudge is powered by large language models too. The difference is everything built around them:

• It always works through the Section 25 framework, so the same factors are weighed every time, in balance, rather than whatever you happened to type.
• It asks structured questions, so the important context is gathered rather than left to chance.
• It draws on current England and Wales case law from The National Archives, not a frozen training snapshot.
• It checks every citation against a verified list before it appears.

That is the gap between improvising an answer and producing a structured, grounded one. It is still an AI assessment, and still not a substitute for advice on your own circumstances; you can read an honest account of its limits on our Limits of AI page. But it is a far more reliable starting point than a general chatbot on its own.