What Senior Counsel Does Before Opening Any AI Tool

Watch someone with twenty years of practice sit down with a new matter and you will notice something that looks, from the outside, like nothing is happening. No document open, no drafting, sometimes not even notes yet. What is actually happening is the part of the work that never shows up in the file, and it is the part that determines whether everything that comes after it is any good.

This matters more, not less, in an AI-assisted workflow, because a drafting tool will produce something the moment it is asked, regardless of whether the asking was any good. The habits that used to be implicit, absorbed over years of practice rather than written down anywhere, are now the entire difference between a fluent draft and a defensible one. It is worth naming them explicitly, because most of them never had to be named before.

Scope the matter before touching a document

The first habit is resisting the pull toward the document. A new matter arrives and the instinct, especially under time pressure, is to start producing something, because production feels like progress. Senior counsel resists that pull long enough to answer a smaller set of questions first: what is actually being asked for, by whom, and why now. Half of what looks like drafting work later turns out to be answering a question that was never actually the client's real question, and that only becomes visible if someone paused to check.

Separate what is known from what is assumed

Every matter arrives with a mix of confirmed facts and background assumptions that nobody has stated out loud, and the two get treated identically by default. Senior counsel makes a habit of explicitly separating them, because an assumption that turns out to be wrong invalidates everything built on top of it, while a confirmed fact does not. This sounds obvious stated plainly and is routinely skipped under deadline pressure, which is exactly when it matters most.

An AI tool cannot make this separation for you. It will treat whatever is in the prompt as given, whether it is a confirmed fact or an assumption someone typed without flagging it as one. The separation has to happen before the prompt, not inside it.

Name the risk before drafting the answer

A contract, a memo, a filing, all of them are ultimately an answer to a risk question: what could go wrong here, and how does this document manage that. Naming the risk explicitly, in a sentence, before drafting anything, forces a decision about what the document actually needs to do. Skip this step and the draft ends up organized around the topic rather than the risk, technically complete and structurally aimed at the wrong target.

The checkpoint that catches what fluency hides

Once framing is done and drafting starts, whether by hand or with a tool, a second habit takes over: a specific, deliberate check for the failure modes AI tools are known to produce, citations that read correctly but reference cases that do not exist or misstate what a real case held, clauses that sound standard but do not match the jurisdiction or the facts as scoped. This is not a general read-through. It is a targeted check, done because the draft sounds confident, not despite it.

• Scope the matter first: who, what, why now, before any document is opened
• Separate confirmed facts from unstated assumptions, out loud, not just in your head
• Name the actual risk in one sentence before drafting the response to it
• Treat a fluent, confidently written draft as a reason to check harder, not a reason to check less
• Verify every citation and case reference against the actual source, every time, not spot-checked

Why this cannot be a policy document

It is tempting to write these habits into a firm policy and consider the problem solved. A policy tells people what to do. It does not build the judgment to notice, in the middle of a busy week, that this particular matter is the one where the risk was assumed rather than confirmed. That judgment is trained the way it always was, through deliberate practice and honest feedback, and the firms getting real value from AI tools are the ones that trained it before they scaled the tools, not after.

None of this is a case against using AI in legal work. It is a case for sequencing. The tool is genuinely useful once the matter has been scoped, the assumptions separated, and the risk named, because at that point it is answering a question that has already survived scrutiny. Used before any of that, it is just producing a fluent guess, and fluent guesses are the ones that are hardest to catch precisely because they do not look like guesses at all.

Where junior lawyers get this backwards

The instinct to start producing something immediately is strongest, understandably, in the people with the least experience to fall back on when the matter feels unfamiliar. A junior associate under deadline pressure, faced with an unfamiliar fact pattern, will often open a drafting tool before scoping the matter, precisely because scoping feels like it is not making progress while drafting feels like it is. Senior counsel resists that pull not because they are more disciplined by nature, but because experience has shown them, repeatedly, that unscoped drafting produces work that has to be redone.

This is a teachable habit, not an innate trait, which is good news for anyone earlier in a career who recognizes the pull toward premature drafting in themselves. The fix is procedural: build in a mandatory scoping step before any tool gets opened, even a rough one written in five minutes, because the discipline of writing it down forces the separation of known facts from assumptions in a way that thinking about it silently does not.

This is also why the habit has to be practiced on low-stakes matters before it is needed on high-stakes ones. Nobody scopes a matter well for the first time under real deadline pressure on a case that actually matters. The lawyers who do this reliably built the habit on ordinary matters, when getting it wrong cost an afternoon of rework rather than a client relationship, and the habit was already automatic by the time the stakes went up.

None of these habits require slowing the practice down in any way a client would notice. Scoping a matter properly takes minutes, not days, and the time it saves later, in avoided rework and avoided exposure, dwarfs the time it costs up front. The habits described here are not a tax on speed. They are what makes the speed AI now offers actually usable.

Read next: A Verified Answer to the Wrong Question Is Still Wrong