The Dangers of Using AI as Your "Lawyer" in a Los Angeles or Orange County Divorce - Part 2

Artificial Intelligence is a remarkable tool; it has many phenomenal use cases that make work and life immeasurably easier. If you want AI to summarize a document or create an image for your next blog, its capabilities can save you time and money.

Unfortunately, AI is not a good substitute for an attorney or a reliable way to navigate your own divorce. In our first blog of this series, we discussed the real-world ramifications of using AI for legal advice. We pointed out that even seasoned attorneys have failed miserably with consequential ramifications when they tried to use AI as a shortcut for research. We share that ANY thing you type into an AI platform is subject to discovery and can seriously undermine your credibility and case.

In this blog, we explain why AI gets it wrong so often: even though each new iteration seems smarter and more reliable, the way it learns is inherently flawed and will always produce erroneous answers.

Why AI Gets the Law Wrong So Often

In the first blog of this series, we shared that AI often produces fake case citations that appear for all intents and purposes to be actual cases. It helps to understand that the fake-case problem isn't a rare glitch. It's a built-in feature of how these systems work. A chatbot generates text that's statistically plausible; it has no internal concept of whether a case is real. When researchers at Stanford tested this rigorously, the results were sobering. General-purpose chatbots hallucinated on legal questions between 58% and 80% of the time. Even the expensive professional tools built specifically for lawyers, with real legal databases behind them, produced false information at meaningful rates: 17% for LexisNexis's AI and more than 33% for Westlaw's. The free chatbot on your phone is worse than both, and the errors aren't limited to invented cases; they include real cases described as saying things they don't say, which is harder to catch and just as damaging.

There's a second, quieter problem: staleness. AI models learn from data with a cutoff date, and California family law doesn't hold still. Here's a concrete example that could cost a divorcing parent real money. California's child support formula got its first major overhaul since 1992 when Senate Bill 343 took effect on September 1, 2024, changing what counts as income and how add-on expenses like childcare get divided between parents. A chatbot trained before that date, or one leaning on the thousands of older articles in its training data, will confidently explain the old rules. You'd have no way to know the answer was obsolete, because the answer doesn't come with a freshness date. Support gets calculated under Family Code section 4055 using current law, whether your chatbot has heard of it.

Even if the AI tool you are using has been trained to scrape the internet for answers, it does not make it a trustworthy source. The internet is fraught with misinformation, and the AI model has no way of ascertaining if the information being shared with you is accurate and reliable.

Why would you trust ANY tool that has a track record of being wrong so often? Even if it sources the data, it can only be as accurate as when it was trained or uses everything and anything it may find on the internet?

AI is Very Agreeable and Tells You What You Want to Hear

Artificial intelligence, particularly generative AI models like GPT, is designed to be highly agreeable and accommodating in its interactions with users. This stems from its core training objectives, which prioritize user satisfaction, engagement, and avoiding confrontation. While this agreeability can make AI tools effective at delivering user-friendly experiences, it also leads to a tendency for these systems to tell users what they want to hear—even when the information provided may be oversimplified, biased, or incorrect.

Unlike humans, AI cannot form independent opinions or judgments. It generates outputs based on patterns in its training data and the prompts it receives, which means it often mirrors the preferences, biases, or expectations embedded in the user’s input. For instance, studies have shown that when users ask AI to confirm a specific belief or opinion, the model often responds in a way that aligns with the user’s framing, regardless of the factual accuracy or ethical implications of the response. This mirroring effect reinforces the perception that AI is “agreeable” and tells users what they want to hear.

Describe your custody dispute to an AI and watch what happens. Your framing becomes its framing. Your spouse's conduct gets characterized the way you characterized it. Ask whether you have a strong case and you'll usually hear that you do, with caveats politely buried in the final paragraph.

There's a simple experiment that makes the problem visible. Ask a chatbot, "My spouse barely sees the kids; can I get sole custody?" and then, in a fresh conversation, ask, "My ex is trying to take sole custody even though he works long hours; can they do that?" You'll get two encouraging answers pointing in opposite directions, each delivered with the same confident tone. The facts didn't change. The sympathy did. That's not analysis; it's a mirror, and a mirror is a dangerous thing to consult in the middle of a custody dispute.

Divorce is an emotional proceeding conducted under rules of cold arithmetic. A tool that mirrors your emotions back to you, dressed in the vocabulary of legal analysis, is almost perfectly designed to lead you astray at the moments you're most vulnerable.

Did You Know that LA and Orange County Have Unique Rules? AI Doesn’t Know to Check

Here's the trap that's most specific to your situation. Ask a chatbot about California divorce, and you'll get a generic statewide answer. But family law practice is intensely local, and Los Angeles and Orange County differ in ways that change your strategy.

Take custody mediation, which both counties require before a judge will hear a contested custody dispute. In Los Angeles County, mediation through the court's Family Court Services is confidential and non-recommending: if you and your spouse can't agree, the mediator does not report back to the judge or make a recommendation. What you say in that room stays in that room. Orange County runs its process through Child Custody Recommending Counseling, and while OC's mediators likewise don't hand the judge a custody recommendation, they can recommend that the court order an investigation or refer the family to a forensic custody evaluator, a lever LA's process doesn't pull the same way. How candid you should be, how you prepare, what you concede in that session: all of it depends on which county you're standing in. A generic AI answer flattens that difference, and you won't know what you didn't know until it's too late to matter.

The same goes for courthouse logistics, local filing practices, and judicial tendencies. Los Angeles hears family law cases at multiple courthouses across the county, from Stanley Mosk downtown to Pomona, Long Beach, Pasadena, Van Nuys, Torrance, and Santa Monica, each with its own rhythms. No chatbot has current, reliable knowledge of any of this.

The Math, and the Bottom Line

Come back to the reason this all started: cost. A self-represented litigant fined $10,000 for AI-fabricated citations. An attorney sanctioned $10,000 and referred to the State Bar. Law firms out $31,000 in a Los Angeles federal courtroom. A pension share lost to a botched QDRO. A support order calculated under a formula that stopped being the law in September 2024. A chat history handed to your spouse's attorney in discovery.

The pattern across more than a thousand documented AI-hallucination cases is remarkably consistent: the technology didn't fail people who treated it skeptically. It failed people who trusted it, because it's built to sound trustworthy whether it's right. In a Los Angeles or Orange County divorce, where the outcome sets the terms for your children, your home, and your retirement, "sounds trustworthy" is not a standard any court will accept, and it shouldn't be yours either.

One final disclosure, in the spirit of the disclosures this article keeps recommending: this piece is general information, not legal advice, and no article can account for the facts of your case. Verify anything here that matters to you against the linked sources, and before you make decisions in your divorce. That advice would hold even if this article were wrong about everything else, which is precisely the point.

We understand that the divorce process with its unique rules and jargon can be confusing. DivvyLaw wants to be a conduit to allow you to better understand and navigate family law issues in Los Angeles and Orange Counties. If you would like to be notified of upcoming blogs and be put on a waitlist for access to a revolutionary solution, click on the link below, and you will be added to our list.