Ask a firm where its institutional knowledge lives and you will get an answer about people, experience, and judgment.
The honest answer is the form file.
Everything a firm has learned about how to do a thing correctly — which certificate a court requires, how the caption is laid out, which paragraph the insurer always fights about, what the signature block must contain — is encoded in the documents it copies when it does that thing again. Not in a manual. In Motion to Compel - FORM.docx, last meaningfully reviewed by somebody who has since retired.
That has two consequences, one good and one bad, and AI sharpens both.
The bad one first: templates propagate errors perfectly
A mistake in a template is not a mistake that happened once. It is a mistake that will happen every time anyone uses that template, silently, until somebody notices — and the whole point of a template is that nobody reads it carefully, because reading it carefully is the work it exists to save.
I have written elsewhere in this section about discovering that a Minnesota rule I had cited for years had been abrogated in 2020 and its text relocated. The interesting part was not the discovery. It was what came next: the old citation was sitting in a form file. It had been reused, correctly, dozens of times — correctly in the sense that each user faithfully reproduced what the template said, which was wrong.
That is the general shape of the problem. Templates are where stale law goes to survive. A rule changes; the world updates; the template does not, because a template is not a document anyone is responsible for maintaining. It is a document everyone is responsible for copying.
The usual suspects, in rough order of how often I have found them:
- Rule and statute citations that have been renumbered, amended, or abrogated.
- Dollar thresholds — jurisdictional amounts, filing fees, statutory caps, exemption figures — that adjust and were captured once.
- Deadline recitals stated as a fixed number of days in the body of a document, which then contradicts the actual current rule.
- Court names, division names, and addresses after a reorganization.
- The firm’s own name, entity type, address, and bar information, in every signature block, after any change.
- Formatting requirements — spacing, font, margins, certificate language — after a local rule amendment.
Every one of those is invisible in ordinary use, because the document looks right. It has always looked right.
Now the good one: a template is a machine-readable specification
Here is the reframe that makes this a practice-management opportunity rather than just a hazard.
A well-maintained template is the single most valuable artifact you can hand an AI assistant, because it is a specification of correctness that a model does not have to infer. Everything the firm knows about the correct form of this document is already in it.
Which produces the most important operational rule I have on this subject, and it took an embarrassing number of bad outputs to learn:
Generation must populate the template. It must never rebuild the document.
Ask a model to “draft a motion to compel formatted for Hennepin County District Court” and it will produce something that looks like one. The caption will be plausible. The spacing will be approximately right. The certificate will be approximately right, which in a filing context means wrong.
Hand it the firm’s actual template file, tell it to copy that file and populate the fields, and the entire class of formatting error disappears — not because the model got better, but because the correct answer was supplied rather than reconstructed. The model’s job becomes filling in the parties, facts, and argument, which is what it is genuinely good at.
The general principle, which applies well beyond documents: where a correct artifact already exists, supply it. Do not ask a language model to reproduce it from memory. This is the same rule as never citing a case from memory and never computing a deadline from memory, applied to form.
Template hygiene, as an actual practice
If the form file is the knowledge base, it deserves the treatment a knowledge base gets.
1. Give every template an owner and a review date. Put both in the document — a footer field or a first-page comment block that says who owns it and when it was last verified. A template nobody owns is a template nobody updates.
2. Review on a schedule, not on discovery. Annually is plenty for most. The review is not “does this look right” — it is: pull the current text of every rule and statute cited, confirm every dollar figure, confirm the court’s current requirements, confirm the firm’s own information.
3. Keep them in version control, or at least in dated versions. You want to be able to answer “what did this template say when we filed that motion in March.” If a defect is found, that question determines the scope of the problem.
4. When you find a defect, search for its siblings. This is the step people skip and it is the one that matters. A wrong citation found in one template is almost certainly in three others, plus a checklist, plus a brief bank. Fixing the instance in front of you fixes one document. Searching fixes the pattern.
5. Audit the whole set for citations, mechanically, once a year. Extract every rule and statute citation from every template, then verify each against the primary source. This is a genuinely good use of automation: it is mechanical, it is boring, it is exactly the kind of thing humans skip, and the failure mode of skipping it is invisible. Run the same four-pass discipline you would run on a brief.
6. Strip placeholders properly. A specific and recurring humiliation: template placeholders are often colored or highlighted so they are visible while drafting. When populated by a script rather than a human, the text changes and the formatting does not — and a document goes out with the client’s name in red. Reset character formatting when populating, and render the final document and look at it before it goes anywhere.
The audit that finds what you did not know
A concrete exercise worth doing once, which reliably finds things.
Take every template in the firm. Extract every citation, every dollar figure, and every stated deadline. Put them in one table with the template each came from. Then verify each against the current primary source.
Two outputs, both useful. The first is a list of defects to fix. The second is more interesting: a map of what your firm believes. Seeing every rule your templates rely on, in one list, is a surprisingly clear picture of the practice’s actual doctrinal footprint — and of which parts of it nobody has checked in a decade.
The first time I did a version of this, the useful finding was not any single error. It was how many of the citations traced back to the same handful of source documents, copied forward for years. The firm’s knowledge base had a much narrower base than its output volume suggested.
Where this connects
Three threads from elsewhere in this section converge here, which is why I think templates are underrated as a practice-management surface.
Never reproduce from memory what you can retrieve. Cases, rules, deadlines, and now form. Same principle, four applications.
Structure supplied is error prevented. The formatting specialist in our chain is instructed to copy the template file and populate it, never to rebuild a caption by hand. That instruction exists because the alternative produced a document that looked right and was not.
Errors that recur are worth fixing structurally. A wrong citation caught in a brief is a save. The same wrong citation caught in the template it came from is a fix. Always ask where the error was stored.
A firm’s forms are the most-used and least-examined documents it owns. Treating them as a maintained knowledge base rather than a drawer of old files is unglamorous, cheap, and one of the higher-return hours available.
General commentary on practice management. Not legal advice. No client document is described; all examples are invented. On verifying rules and citations against primary sources rather than from memory, see the related pieces in this section.