Why AI Can’t Give You an Unbiased Answer to Your Family Law Questions

(And What That Actually Means for You)

If you’ve ever typed a divorce or custody question into ChatGPT or Claude and wondered whether the answer you got was actually neutral — good instinct. It wasn’t. And the reason why matters more than most people realize.

Let me explain what’s actually going on, because it’s not the scandal you might think it is.

AI is trained on human-generated content. Humans are not unbiased.

Every large language model learns from an enormous corpus of text — legal briefs, court opinions, forum posts, academic articles, advice columns, Reddit threads. That content reflects the people who wrote it: their assumptions, their cultural contexts, their socioeconomic backgrounds, their outcomes. If the legal system has historically disadvantaged certain parties in divorce or custody proceedings —then those patterns are baked into the training data, and by extension, into the model.

This isn’t a flaw that engineers can simply patch. It’s a structural feature of how these systems work.

AI doesn’t know your jurisdiction. Family law is intensely local.

“Unbiased” also implies “accurate,” and accuracy in family law is jurisdiction-specific in ways that AI handles poorly. Texas family law is not California family law. Travis County practice is not Williamson County practice. A custody standard that applies in one state may be completely wrong in another. AI tools are trained on broad datasets and tend to give you the most statistically common answer — which may have nothing to do with the law that governs your case.

AI has no stake in your outcome. That’s not a feature.

Here’s the part that surprises people: an AI’s “neutrality” is actually a limitation, not a virtue. When I advise a client, I am supposed to be on their side. I have an ethical obligation to advocate for them, to identify every angle that benefits their position, to anticipate what the other side will argue, and to tell them hard truths when their position is weak. That is not bias — that is representation.

AI cannot do that. It will give you a balanced overview because it has no client relationship, no obligation to you, and no authority to give you legal advice. What feels like objectivity is really just the absence of advocacy.

So, what is AI actually good for in this context?

Quite a bit, honestly — just not the things people most want it for. AI is useful for general orientation: understanding what terms mean, knowing roughly what questions to ask, getting a sense of the procedural landscape before you walk into a lawyer’s office. It can help you organize your thoughts and your timeline. It can explain concepts like community property, conservatorship, or the best-interest-of-the-child standard in plain language.

What it cannot do is apply those concepts to your facts, advise you on strategy, tell you when to settle, or predict how a particular judge in a particular county is likely to rule. For that, you need a lawyer who practices where you live and knows the courthouse you’re walking into.

AI only knows what you tell it — and you are not a neutral source.

There’s another layer to this that almost nobody talks about. AI doesn’t have access to your spouse’s text messages, your financial records, your custody history, or the context that a lawyer builds over months of working your case. It only knows the facts you choose to share — which means it is working entirely from your version of events. And when you are in the middle of a divorce or a custody dispute, your version of events is, by definition, not unbiased. You are hurt, or scared, or angry, or all three at once. You remember what you remember. You emphasize what feels important to you. When you ask AI a question framed around your perspective, it will validate the premise of that question far more readily than a good lawyer should. A lawyer’s job includes pushing back on your narrative when the facts don’t support it. AI has no mechanism — and no incentive — to do that. The result is that AI doesn’t give you an unbiased answer; it gives you your own bias reflected back at you, dressed up in authoritative language.

What you tell AI is not protected — and that matters more than you think.

This may be the most important practical warning in this entire post. When you share facts with your attorney, that conversation is protected by attorney-client privilege. It cannot be subpoenaed, it cannot be used against you in court, and your attorney cannot be compelled to disclose it. When you type those same facts into an AI chatbot, none of those protections exist. Your conversation with an AI is not privileged. Depending on the platform’s terms of service, those conversations may be stored, reviewed for training purposes, or potentially subject to discovery in litigation. In a contested divorce or custody case, what you typed into a chatbot at 11 PM when you were venting about your spouse could end up in front of a judge. You are, in effect, creating a written record of your thoughts, fears, and legal strategy with a third-party platform that has no obligation to protect you. 

The bottom line

AI isn’t biased in the conspiratorial sense — it’s not secretly rooting for one side of your divorce. But it’s also not neutral in any meaningful way. It reflects the patterns of its training data, it lacks jurisdictional precision, it has no obligation to you whatsoever, and it can only work with the one-sided picture you hand it. In family law, where the stakes involve your children, your finances, and your future, “pretty good general information based on your own retelling” is a dangerously low bar.