What AI Can Teach Partners About Giving Actionable Feedback on Legal Writing

It’s 11:47 p.m. A partner marks up a brief: tighten here, awkward there, not persuasive in the margin beside what the associate thought was the strongest paragraph in the document. The associate reads the comments the next morning and learns almost nothing about writing. What she learns is how to survive this partner. The following week, a second partner rewrites the same section a different way, and she learns how to survive that one too.

Then the firm wonders why writing quality swings so wildly across teams, offices, and classes of associates.

After years of training lawyers and judges, I’ve come to think the problem isn’t that partners can’t tell strong writing from weak. They can. The problem is that their feedback on writing consists of a reaction instead of an instruction.

The translation problem

“This is too wordy” is a reaction. Here is the instruction hiding inside it: Cut a word from every sentence. Swap some long words for short ones. Keep your average sentence under 25 words. Trim prepositional strings and throat-clearing phrases like “there is,” “the fact that,” and “it is important to note that.” A writer can carry that into the next brief and the brief after that. The reaction only tells her this draft fell short. Give specific direction she can implement.

The same gap runs through every common comment. “Your structure is hard to follow” becomes:  Make three points up front and turn them into headings; open each paragraph by carrying forward something the reader just learned. “This isn’t persuasive” becomes: Tie every authority to our client’s facts within two sentences, cut the adjectives—egregious, clearly, plainly—and rebuild the point on logic alone.

The hard part of feedback is that act of translation: converting what you noticed into something the writer can do. And the reason firms struggle to lift writing at scale is that almost no one has time to perform that translation by hand, comment after comment, draft after draft.

The rewrite trap

So partners take the shortcut and rewrite the draft themselves.

In the moment, that is often the right call—explaining the fix takes longer than making it. Across a career, it is one of the most expensive habits in the building. A silent rewrite teaches an associate one thing: the partner’s version wins. It carries no principle, no structure, no judgment she can reuse. It builds dependence instead of skill, which all but guarantees you will still be rewriting her work three years from now.

The alternative costs more on Tuesday and far less by the next quarter. Name the principle, fix one paragraph as the model, and have the associate apply it to the rest. Show the standard with an example from a strong prior brief—not from your keyboard at midnight. Better still, ask rather than tell: What’s the rule here? Whose voice is this? A rewrite teaches your voice. A question teaches hers.

Inconsistency reads as chaos

Now multiply one partner by twenty. One wants force; another wants restraint. One wants the precedent discussed at length; another wants it cut to the bone. None of that is a problem—reasonable lawyers differ.

It becomes a problem the moment those preferences stay unspoken. To the associate, two partners’ contradictory edits don’t read as two defensible views. They read as nobody here knows what good is. She stops trusting any of the comments because the standard shifts with the reviewer. Midlevels become translators. Review cycles fill with people decoding personalities instead of sharpening arguments.

The firms that raise writing quality the most standardize more than partners expect—not every rhetorical choice, but the floor: structure, sentence length, headings, transitions, citation form, and the line between an edit you must make and one you would merely like to. That last distinction is invisible to the associate unless you mark it. So mark it.

The part that should surprise you

Here is where the research turns counterintuitive. Even though it wasn’t about legal writing, a recent study helps explain something law firms see every day: the same correction lands differently depending on who delivers it.

In a study published in the Journal of Information Systems, employees rated how much they trusted feedback from a human supervisor against the same feedback from an AI system, on a seven-point scale. When they expected the human to be generous, they trusted the human more—5.08 to 4.43. No surprise. But when they expected the human to be hard on them, trust in the human’s feedback fell to 2.71, while trust in the neutral evaluator held at 4.66. The person dropped below the midpoint. The neutral evaluator stayed above it.

Anyone who has delivered tough edits under deadline knows the pattern. Praise carries weight precisely because it comes from someone whose opinion shapes your career. Correction carries baggage for exactly the same reason. Strip out the hierarchy and the late-night tone, and the same critique can land cleaner.

AI feedback also appears capable of teaching well in many structured learning settings. A recent Educational Psychology meta-analysis pooled 41 studies and 4,813 students and found no statistically significant difference in learning performance between students who received AI-generated feedback and students who received human-provided feedback.

Neither finding says replace the partner. Both say something more useful: the correction and the relationship don’t have to come from the same place.

The division of labor

Most of what makes feedback work is structural. Tie it to a principle. Use the same vocabulary every time. Mark severity. Flag the recurring issues. Deliver it in the moment, without the sigh. Some partners do all of that by instinct. Most don’t have the hours.

That’s the real case for AI in legal writing—not that it outthinks your best partner, but that it stays consistent across every draft, neutral in tone, available at the point of drafting, and free of the interpersonal weight that even a careful margin note carries. Put plainly: AI doesn’t compete with your best feedback. It competes with the feedback you don’t have time to give.

What stays human is everything that matters most. How hard to push before this particular judge. Which concession buys credibility. Which issue carries the appeal. When to protect an associate’s voice instead of flattening it into your own. And, just as critically, the praise and the relationship.

In that arrangement, the partner’s job doesn’t shrink. It sharpens. Less time rebuilding sentences. More time shaping arguments and developing the lawyer who wrote them.

What the best firms are really buying

Most firms pour money into recruiting, business development, and knowledge management while treating writing as something associates absorb by apprenticeship. That held up when teams were small, clients were patient, and review cycles were slow. None of those things is true anymore.

Inconsistent writing now shows up on the income statement: longer turnarounds, more rewrite rounds, slower delegation, lower realization—quarter after quarter. Every avoidable rewrite is both a training failure and a margin leak.

A shared way of giving feedback is the rare investment that compounds. A principle taught once becomes reusable across matters, teams, offices, and the next associate the first one mentors.

Because the most expensive feedback in a law firm isn’t bad feedback. It’s feedback that gets delivered, accepted—and forgotten.

The feedback that compounds is the kind the associate can use when the partner is no longer in the room.