"Did you use artificial intelligence for this?" is quickly becoming the least useful question a legal leader can ask their team. Not because the answer doesn't matter, but because, increasingly, the answer is going to be yes, which tells you almost nothing.
As Factor's 2026 GenAI in Legal Benchmarking report found, legal teams have moved quickly on AI access, with more than 82% of teams reporting broad access and over half of respondents saying AI is used regularly across their team.
AI is no longer something your team might be experimenting with on the margins. It is already woven into everyday work. Associates and in-house lawyers are using it to summarize, compare, rewrite, brainstorm, structure and draft — sometimes lightly, sometimes heavily. Often, in a blended way, that makes it impossible to point to a paragraph and say with confidence whether it came from the lawyer, the AI or both.
What has not kept pace is trust. In Factor's benchmarking research, only around 22% of legal teams reported high trust in AI outputs, highlighting the next challenge for legal leadership. The question is no longer, "Did you use AI?" The question is, "How did you do this, and why should I trust the result?"
The variable is the process, not the tool.
One of the easiest mistakes a leader can make is treating AI use as the risk. AI-enabled work can be sloppy, just as human work can be sloppy. But both human and AI-enabled work can also be excellent. The variable is not the tool, it is the process.
In my work training over 4,000 lawyers and legal leaders on AI, I've seen the instinct to police AI use backfire when AI is framed as suspicious by default. This framing encourages shadow use and pushes people to hide how they worked instead of talking openly about it. Crucially, it gives supervisors less visibility into the thing that actually matters: how the work was done.
Focusing only on whether someone used AI is a bit like asking whether they wrote the memo at home or in the office. It may tell you something about the setting, but it tells you very little about the quality. Good supervision in the age of AI starts by rejecting that binary.
Surface-level quality is now cheap.
For years, supervisors could use polish as a rough proxy for rigor. If a memo was clear, structured and well written, that often suggested that the thinking behind it was sound — not always, but often enough. That shortcut no longer works.
AI can generate polished, coherent writing in seconds and produce something that looks like midlevel or senior work, even when the underlying thinking is still very junior. This changes the signal that supervisors are relying on. A polished paragraph no longer proves a strong process, but simply that the person prompting the tool knew how to generate gloss.
That's why leaders need to be careful about mistaking fluent output for reliable work product. Clear writing still matters, but it is no longer evidence, on its own, that the person has done the reading, tested the reasoning, checked the authority or thought through the edge cases. Surface-level quality is now cheap. Trust is not.
Blended workflows are the new normal.
Another reason the AI use question breaks down is that the answer is rarely clean or linear.
In practice, strong AI use often looks like this: Someone brainstorms with AI, drafts manually, asks AI to restructure a section, rewrites it themselves, uses AI again to tighten the language, gets feedback from a supervisor and then revises once more. That is not AI work or human work, but a blended work product.
Supervisors should not waste time trying to identify which sentence was written by whom. From the finished document alone, you often cannot tell, but, more importantly, that is not the right supervisory question to ask.
The better question is whether AI was used effectively, responsibly and in a way that supports a trustworthy result. That is a different standard, and a much more useful one.
Verification is no longer embedded in the workflow.
In traditional workflows, a lot of verification happened naturally on the way to the answer. A lawyer has to find the authority, read the cases, compare the facts, test the reasoning and build the argument step by step. The process itself created checkpoints. AI changes this, and that is the part that most leaders feel before they can name it.
Now it is possible to jump straight from a question to a plausible answer. The output may look complete, and it may even be correct, but the intermediate verification steps are no longer guaranteed to have happened. This becomes an issue for leaders who are charged with supervising the work.
If someone hands you a neat, plausible summary or memo, you cannot assume the usual checks occurred along the way. You need to deliberately surface the process that sat behind the output. This does not mean turning every review into an interrogation, but rather knowing what to ask.
Supervise the process, not the tool.
The simplest way to think about this is through two buckets: inputs and outputs.
On the input side, you want to understand what went into the tool. What did you ask it to do? What sources or documents did you give it? Did it have the right jurisdiction, governing law, client posture or business objective? Was it working from primary material or general knowledge? What constraints did you set?
On the output side, you want to understand how the result was checked. Did you verify that the cited sources were real? Did you confirm that the answer actually fits our facts and posture? Did you test for what might be missing, overstated or assumed? Are any of the conclusions stronger than the support allows?
Those questions do two things at once: They improve quality and create a coaching moment. They help supervisors move from fixing the draft to understanding the workflow that produced it. That is where the real leverage is.
Because once you can see the process, you can improve it. You can spot whether the issue was bad context, weak source material, a poor prompt, lack of verification or overreliance on AI for a task that needed more human judgment. Without that visibility, you are left reacting only to whatever errors happen to surface in the final draft.
This is what legal leaders need to do now.
The broader leadership point is this: If AI is already part of how legal work gets done, supervising AI-enabled work is no longer optional, but a core management capability.
This means that leaders need to do more than approve tools. They need to create a shared foundation for what good AI use looks like on their team. That foundation should cover three things: what AI is good at, where it struggles and how to get better results from it. Without that baseline, leaders are trying to supervise a process that they have never clearly defined.
They also need to make expectations explicit. Give people a language for talking about process, not just output, and normalize the idea that responsible AI use is not about hiding the tool, but making the workflow visible enough to trust.
Clear guardrails matter too: where AI is appropriate, which tools are approved, what data can go where, and how AI use should be disclosed or recorded.
The future of supervising legal work will not be about spotting whether AI was involved, but judging whether the process was sound, whether the checks were appropriate and whether the output is good enough to act on.
That is the bar now: not "Did you use AI?" but "Can I see how this was done, and do I trust it enough to move forward?" This is the question that legal leaders need to get good at asking.
Originally published in Law360 Pulse. To find out more about our ‘Supervising in the Age of AI’ workshop, contact sense@factor.law.
