Let Lawyers Be Lawyers, Let AI Search for the Cases

Legal research AI is built to hand you an answer — and that is genuinely useful for some work, especially if its low stake. But most of the time, that is not what the work requires. Not every day. Not every case. But when a critical issue of strategy, analysis, or judgment turns on a knotty question of law, they aren't going to just ask a junior lawyer for the answer. They are going to dive in and read the key cases themselves. For 20 years as a litigator, first as an associate and then as a partner, that was the request I received from senior lawyers or made to junior lawyers. Why? It's because reading the cases — the key, relevant cases — is not grunt work. It isn't something you want to push off on a less experienced colleague. To the contrary, it is one of the places where you can bring all your experience and judgment to bear to determine your best way forward. It's while reading cases that you have thoughts like 'none of these courts have ever really faced facts like ours', 'our issue is a loser unless we can reframe it this way', 'we may survive this motion, but we are going to need to be ready to explain X to the jury'. It is, in short, one of the times where your judgment and experience as a lawyer really matter and you are really bringing value to your clients. Which is why it's strange to me that legal AI seems to be gravitating towards taking that work away from lawyers. Most AI research tools are designed to produce an answer: a grounded, footnoted response to a question you typed. That is genuinely useful for some work, especially if its low stake. But for things that matter, for the legal issues on which a tactical or strategic decision hinges, it is, by definition, taking away your ability to be creative and analytic about how to use the underlying cases in your legal argument. Put another way, an AI summary of a set of cases is, by design, a reasonable, well-worded, articulate—and decidedly average —distillation of the answer you want. And for the most part, that is not what you want. You don't want the AI answer; you want the AI to help you get to your answer . For that, you need to read the cases. Let me give a few examples: A client calls about a new matter, or an old matter takes a turn, and you have an hour before the call. What you need is not a memo. You need to have read the six or seven opinions that actually govern the question, so that when the client asks whether anyone has ever won on this, you can answer from the cases instead of from a summary of the cases. The difference shows up immediately in conversation. A lawyer who has read the opinions can handle the follow-up question. A lawyer working from a summary has to say they will look into it. This is the request that answer-generating tools handle worst. You have a theory. It might be creative, it might be a stretch, and what you want to know is whether any court has ever gone there, and if so, what the facts looked like. A tool that gives you an answer will give you the general rule. The general rule is usually the thing you are trying to get around. What you need is the outlier: the district court that took the argument seriously, the concurrence that left a door open, the case with facts strange enough to be useful. Those do not survive summarization. They have to be read. Opposing counsel cites 11 cases for a proposition. Some of them stand for it. Some are cited for a sentence buried in a footnote. You need to know which is which, and whether any court has gone the other way. That is verification work, and it goes faster when you start from the opinions rather than from someone's description of the opinions. It is also the moment where a mischaracterized citation does the most damage, in either direction. A demand letter, a discovery dispute letter, a settlement proposal. You are making a point that would land harder with a citation behind it. You are not writing a brief and you are not spending three hours. You want one good case, correctly characterized, in the next 20 minutes. Small task, constant occurrence, and historically awkward to staff. Too small to hand off, too slow to do yourself between other work. Here's what's ironic. While current AI tools are tuned to tell you the answer — the thing they are decidedly average at (not bad; just average), they are actually capable of being quite good at the thing you really want: 'find me the cases.' Or in some cases, find me the cases and point me to the relevant sections. I know AI is capable of doing that because, along with many other lawyers I know, I am using AI to do just that: find me the cases. (Full disclosure, I also know AI can do that, because I've actually built an AI tool designed to do exactly that — and only that, and it works really well.) Of course, if you go about it this way, the reading is left to you. Intentionally. That part has not been automated and should not be. Reading opinions is how you develop a feel for a body of law, and it is where the arguments come from that the other side did not think of. Handing the AI the annoying part, sorting hundreds of opinions down to the 30 that matter, is a good trade. Handing AI the reading is not. Sam Davidoff is the founder of Align, a litigation technology company and the maker of Align Research . He spent 20 years as a litigation partner at Williams & Connolly. From Paper to Pixels: How Litigators Can Successfully Adopt Digital Binders Three Binders Every Litigator Needs Litigation Binder Checklist Preparing For Witness Preparation The Ultimate Deposition Outline: 3 Tips for Better Litigation Prep Image © iStockPhoto.com.