December 7, 2025

LLMs Make Legal Advice Lossycheap chatbot summaries take important choices away

With many-megapixel cameras in our pockets and networks so fast that streaming hiccups piss off pre-teens, it’s easy to forget how frail digital imagery used to be. When storage cost real money and the Internet was slow, we took photos at much lower resolution. Then we used formats like JPEG to strip detail in clever ways that preserved overall impressions, for the sake of speedy sharing. But if you zoomed in on an image, or even just looked closer, you’d often find blurry blobs, jagged edges, and cryptic artifacts where nuance should have been. Such were the losses of so-called lossy compression.

In the real world, there was no enhance button. Sometimes a photo just wouldn’t show you what you needed to see. So you had to search around for a better image and wait for it to download. Or you had to go back to the source.

Something similar is happening with advice to some of my clients. Instead of reading what I write for them, they’re pasting it into chatbots for summaries, then reading those summaries instead. It’s fast. It’s cheap. It’s at least notionally private. They don’t have to tell me they’re doing it.

More often than they might hope, I can tell.

Abstraction Control

When I give advice, I am nearly always summarizing something from the get-go. Unless we are really in the weeds of some specific risk assessment or legal theory, I rarely copy-paste statutes, regulations, or court opinions at my clients. Rather, it’s my job to give the shortest, cleanest generalizations of the rules that work for the decision they need to make—abstractions of the law that leak off into irrelevant space, not all over their dreams and plans.

I pick the level of each abstraction. That is often one of the most important decisions I make in service of a client on a problem. How high-level can we talk? Where do we need to drill down from there?

LLMs give my clients a silent veto over these decisions. Prevailing trends toward ever sparser chat- and text-style writing, plus the ever crushing weight of time pressure, push them in one direction: fewer words, fewer details, leakier abstractions. In other words, lossier pictures of the law, more riven with smudge-blobby background, corrupted lines, and arbitrary aberrations. Cheap automatic summaries of expensive professional précis of complex, often unsynthesized source material.

Based on prior, pre-LLM experience with expertise, I suspect using a chatbot to do this looks and intuitively feels different from using an LLM to add information. Few remain surprised now when prompting an LLM too hard summons out inscrutable gobbledygook littered with allusions or even citations to nonexistent authorities, thanks to the headline-grabbing lawyers who’ve fallen in this hole by the pageful. But I’m not aware of any similar click-grabbing parables for the dangers of chatbot summarization, unless you squint and count the poor souls who’ve convinced they’re geniuses or deities. Given the abstractness involved—we’re talking about a general pattern with the process of generalization here—I wonder if there could be.

Meanwhile, I should admit some concern that summaries not of my choice will nonetheless be attributed to me. I worry that when my clients “trade e-mails with a lawyer”, the system they’ve put in between us becomes invisible. I wonder how much of my audience has had the experience of belaboring a piece of writing, then feeding it to a chatbot for a summary and reading the result.

For Instance

Having made that rather lofty generalization-of-generalities point, I feel a sudden urge to touch ground again. Perhaps that’s the kind of instinct that’s helping keep some of us sane these days.

Take the basic rule of copyright ownership that authors own the rights in their works. That’s a fair general statement of the law on the books. It’s also a dangerous thing to tell a client involved in working business. In business, the “rule”—the normal thing that usually happens—is that legal rights end up with employers and clients. Sometimes the law itself gives it to them. Sometimes they get there through contracts most everyone signs.

Even that more practical heuristic lays traps for unwary clients when exceptions matter. Is the client dealing with contracts doing one of the kinds of jobs that can be “work made for hire” under the statutory definition? Is telling the client about “ownership” without mentioning some creators’ rights to terminate old deals after 35 years inviting a fatal assumption, or omitting pointless pad on the word count? Do we need to talk exceptions just under current law, for new work, or do we need to consider old rules for older works, as well?

Not every legal question points to a decision tree as knotted as this one. But this one’s not an exceptionally crooked specimen, either. The law can do far worse, then flock a few branches with uncertainty.

It is a totally normal part of a functioning lawyer’s day to mentally scale ladders of abstraction like this, picking and choosing where to stand the client, what to point at, and what to leave out. After all, every irrelevant nuance, variation, and aberration included costs money to write up, eats client time to read, and drags the convo closer to the limit, where they get overwhelmed with new info and everyone has to double back.

Getting good at neither coddling nor firehosing clients is a significant part of the difference between being a good law student and being a good lawyer. If you learn all the rules, exceptions, and citations and stop there, you are at best fully qualified to practice law badly. Practicing law is often partly teaching law, but unless the job is law professor, rarely are we called upon to teach anybody everything on a topic.

We’ve all seen the kind of run-on, over-inclusive, largely unhelpful advice that comes from lawyers who “dump” the law as they learned it, or in the raw form they recently dug it out of the books, without any tailoring to their reader or the problem. This is what a lot of law-student and new-lawyer writing looks like.

Overshooting on detail is apparently a natural tendency for competent, aspiring minds learning the field. Now we’ve given giving clients the tools to overcompensate by undershooting. We’ve automated it, hyped it, and subsidized it.

Word Choice

I have also seen a second, related danger. I’m less confident I can spot this when it’s happening, but I’m developing a hunch that’s more often than I first thought.

To summarize a summary is not just to compress an already compressed summary, but also, often, to reword it. Choice of words can be important in what I do. Not inevitably in the cartoonish way those with little experience of the law tend to suppose, where failing to say exactly the right words produces no magic. In a more serious way affecting how much advice my clients need, longer-term.

Many of my clients, maybe most, have heard me say that it’s my job to learn their language, not their job to learn mine. I’ve found this to be an effective way of signaling that it’s on me to write and speak so my clients understand, and on them to complain when they don’t and not feel sheepish about it. It’s a framing tool like the flipped form concept, but for correspondence, not just drafted operative terms.

Still, there are occasionally tidbits of jargon that just make sense to teach and to learn. To help clients communicate with others besides me. To help them search out supplementary materials. To make them seem savvy in more experienced company.

Sometimes these are words or phrases directly from primary sources, like “derivative work” in the Copyright Act. Sometimes they’re live concepts in law or business practice, like “right of first refusal”. And sometimes they’re well distributed technical terms, like “forking” in software. It’s important to be able to explain each of these in Real English. But there’s separate, identifiable value in knowing which shorthands get said and heard.

When a client uses a chatbot to effectively rewrite my guidance, I lose control of how worthwhile terms get introduced. I lose control of how they’re sprinkled through the text later, to develop fluency. Both content and presentation end up on invisible autopilot. There is no bright line between those in the law.

I’ve seen this bite in particular where a client takes a summary without telling me, advances the conversation, and then misunderstands a later message. All of a sudden, it turns out they weren’t following along on all the levels I’d assumed. They weren’t getting everything they paid for, either. If I was going deeper or longer than suited their needs, that should have been said, and I should have adjusted, potentially reducing their bill. As they fell deeper behind on the lingo then anticipated, I can only expect they hit their limit sooner, inducing yet more summarization on the sly.

Still

None of this is to say it’s all bad. There is a pressure coming here that I feel I ought to welcome.

I blog in sections. My e-mails often come with bold-faced headings. I do a lot of outlining. Increasingly chat-, text-, and tweet-driven workstyles are veering ever further away from my preferred dose and concentration of prose. And yes, I’m getting older.

I see some strong if somewhat hand-wavy arguments that slower, longer, and careful writing better suit at least some of the kinds of work clients pay me to do. But in the same way it’s on me to write so clients understand, there’s a strong case it should be on me to write more like they would prefer to read, too.

Fortunately, while the dramas of this new lawyer-client dynamic interest me well enough, I don’t think they point to any kind of newfangled remedy. We ought to be on the level. If clients are in positions that drive them to interpose a chatbot between us, out of ebullient techno optimism or simple time crunch, I should be making it clear that’s something they should unhesitatingly tell me. I should be prompting them to say so.

Your thoughts and feedback are always welcome by e-mail.

back to top — edit on GitHub — revision history