A pattern in client meetings that did not exist five years ago.
A client comes in with a problem. They explain the problem. I begin to give my analysis. Within the first few minutes of my analysis, they interrupt with something like: “but ChatGPT said…” or “but I asked Claude about this and it told me…” Sometimes they have screenshots. Sometimes they have a printed-out conversation. Often they have, in their head, a fully-formed alternative version of the legal situation that the AI told them was true.
I am no longer having a conversation with the client. I am having a three-way negotiation: me, the client, and the AI whose opinion they have already received.
This is the new structure of legal consultation. And it requires skills that most lawyers, including me, did not develop in law school.
The composite pressure
What makes this conversation harder than the conversations of a few years ago is that the AI’s opinion does not arrive alone. It arrives as part of a composite of authority.
The client has typically also consulted:
- A friend or colleague who “had a similar situation”
- An article they read online (often itself AI-generated)
- A previous lawyer’s brief opinion from years ago, on a tangentially related matter
- Their own intuition, sharpened by anxiety
The AI’s contribution to this composite is not the only source, but it is often the most confidently-stated one. When the client comes to me, they are not relying on the AI alone—they are arriving with a blended consensus of multiple authorities, of which the AI is the most articulate and the most reinforcing. The AI has not just given them an opinion; it has organized and amplified the other opinions, knitting them together into something that feels coherent.
I am now arguing against this composite, not against the AI alone. When I tell the client that their position has problems, I am not contradicting just one source. I am contradicting a self-reinforcing coalition of sources that the client has spent the last week building in their head.
This is what makes pushback harder than it used to be. The client doesn’t just have a wrong opinion; they have a system of mutually-confirming opinions, and the system feels, from the inside, like a comprehensive understanding of their situation.
The tactical problem
Pushback against a single bad opinion is straightforward. Pushback against a composite is not.
If I tell the client “your friend doesn’t know the law on this,” they can fall back on the AI and the article. If I tell them “the AI got this wrong,” they can fall back on the friend and the article. Whatever I attack, the composite has redundancy. The structure is designed—not deliberately, but emergently—to resist contradiction.
I have learned a few tactical responses, which I want to share because I have not seen them written about elsewhere.
Acknowledge before contradicting. “I understand why ChatGPT gave you that answer—if your situation were a straightforward debt collection case, that’s the right analysis. Here’s why your situation isn’t a debt collection case.” The acknowledgment shows the client that I have considered the AI’s framing, which makes them more willing to consider mine.
Isolate the AI from the rest of the composite. “Your friend’s experience and the article are reflecting one set of cases. The AI is generalizing from that same set. Here’s the category your situation actually falls into, and here’s why none of those sources were tracking it.” This separates the AI’s opinion from the supporting context, weakening the composite without attacking any individual source.
Make the client an investigator. “I’d encourage you to go back to the AI and ask it the opposite question—‘what’s the strongest argument that I do owe this money?’ See what it says.” When the client does this, the AI obligingly produces a counter-analysis. The client now has, from the same source, two contradictory analyses. The AI’s authority is undermined by its own willingness to argue both sides. This is more persuasive than anything I could say.
Concede partially where I can. “The AI is right about the basic principle, just wrong about which principle applies.” This is often literally true and lets the client feel that the AI wasn’t completely wrong, which makes them less defensive about abandoning the AI’s conclusion.
None of these tactics are sophisticated. They are the basic moves of any persuasion situation. But they have become necessary in client meetings in a way they weren’t five years ago, and most lawyers haven’t consciously adopted them yet.
When pushback fails
Sometimes none of these tactics work. The client is too invested in the composite, too anxious about the underlying situation, or too unwilling to accept that their preferred outcome may not be available. In these cases, the conversation can fail.
What I have learned is that failed conversations are not always failures of communication. Sometimes the client is genuinely choosing to believe the AI over me, and there is nothing I can do about it.
When this happens, I have a choice. I can keep arguing, escalate my tone, produce more citations, try harder. This rarely works and usually makes things worse. Or I can step back and tell the client, clearly and without heat: “I have given you my best analysis. You have decided to rely on a different analysis. I respect that decision, but I cannot represent you on the basis of an analysis I believe to be wrong. If you want to proceed with the AI’s framing, you should find a different lawyer. If you want to proceed with mine, I’m here.”
This is uncomfortable to say. It feels like losing a client. But I have come to believe it’s the right move. A client who is determined to act on the AI’s framing will not be served by a lawyer who pretends to agree with that framing. They will pay for representation they cannot use. And when the strategy fails—which it usually will—they will blame the lawyer, not the AI.
The clients I lose this way, I have come to think, are clients I would have eventually lost anyway, after spending more time on them. The clients I keep are clients who actually want my analysis. The selection is healthier than the alternative.
What this skill actually is
I want to name the skill I have been describing, because I don’t think it has a good name yet.
It isn’t “legal analysis.” That’s the underlying competency, but it’s not the part that’s new.
It isn’t “client communication.” Lawyers have always needed that.
It’s something more specific: the ability to deconstruct a composite of authorities that includes AI-generated opinions, isolate the inaccurate parts, and lead the client through a process of revising their belief without making them feel humiliated or attacked.
This is not what law schools teach. It is not what bar exams test. It is not what continuing legal education programs are organized around. It is, however, increasingly the skill that determines whether you can keep clients. Lawyers who develop it will thrive. Lawyers who don’t will find their client relationships gradually deteriorating, without being quite able to say why.
The composite pressure isn’t going away. It is going to get more intense. AI tools are improving, which means client-side AI use will become more sophisticated. The composites will be richer, more internally coherent, harder to deconstruct. The skills required to manage these conversations will become more demanding.
If you are a lawyer, this is the area where I would invest my professional development. Not in learning to use AI better—that matters too, but it’s not the highest-leverage skill. The highest-leverage skill is learning to be the person in the room who can lead the client through the AI-shaped fog they arrive carrying.
A closing observation
There is one more thing I want to say about these conversations, because it has surprised me.
The clients who eventually accept my analysis—the ones who let me deconstruct the composite—often trust me more afterward than they would have if they had come in blank. The act of working through the conflict, of being shown why their initial belief was incomplete, creates a stronger bond than a smooth consultation would have created.
I think this is because the client experiences the conversation as evidence that I am willing to push back on something they wanted to believe. Most professionals are agreeable. The AI is, by construction, agreeable. A lawyer who pushes back, calmly and persuasively, against what the client wants to hear is a rare experience. The clients who recognize this become long-term clients, often referring others.
So the three-way negotiation isn’t only a problem. It’s also an opportunity. The clients who survive the deconstruction are clients I am happier to work with than the clients I had five years ago, because we have already done the hard work of establishing what they can expect from me.
The AI in the room is, in this sense, performing a kind of involuntary screening function. The clients who can’t get past the AI’s framing are clients I shouldn’t be representing anyway. The clients who can are the ones worth keeping.
I am not sure the legal profession was ready for this. But it’s the situation we’re in.
Part of an ongoing series on lawyer-client dynamics in the AI era. Related: why clients are shopping for the answer they already want and why clients pay for certainty, not answers.
If you’ve developed your own tactics for managing AI-shaped client conversations, email [email protected]. I’m building a library of these.