Nothing Goes Out Under Its Name

Somebody in your office forwarded that opinion around with FYI in the subject line and nothing else in the body.

You know the one. A brief went in with case law in it that did not exist. The names looked right, the cites had the right shape, and the holdings fit the argument beautifully, which should have been the tell. Nobody caught it until a judge went looking for one of the cases and could not find it. Everybody in the building read that opinion twice, and nobody enjoyed it.

So you are not starting from zero. You have seen what this thing does when somebody in a hurry trusts it, which means the two loud stories about it land differently on you than on everybody else.

The first says it is coming for exactly your desk, and the argument is short enough to sound solid: legal work is text, this was built for text, and the words review, summarize, draft and index appear in every paralegal posting ever written. There is a slide at a CLE with a shaded band across the bottom, and somebody from your certificate program who asks, kindly, whether you have a plan.

The second arrives as a login. The firm bought a platform over the summer, an AI panel appeared inside the review tool without notice, and there is a mandatory webinar Thursday at noon. Somebody sold it to a partner over lunch on the promise that it handles first-pass review, summaries and chronologies. Nobody asked you, and you are the one who has to make it work by Monday.

Both are about a year nobody has billed yet, and neither touches this afternoon. Thursday’s transcript still has to become four usable pages. The second production came in at 4,100 pages behind an index that does not match the Bates ranges. Three clients called and nobody has called them back. Friday’s deadline is Friday whichever story turns out to be true.

Neither story is this book’s business. The transcript is.

Take the first story first. A list ends it, and nothing on the list is a matter of putting words in order.

It cannot appear. Not at a calendar call, not at a case management conference, and not at the clerk’s window at ten to five because the portal has bounced the same document three times over a caption error four people have failed to find.

It cannot be in the room either. It cannot sit through a deposition, feel which way the questioning is drifting, and have the right exhibit in the attorney’s hand before he finishes asking for the wrong one.

It has no license and no bar number, so nothing goes out under its name: it does not sign, it does not file, it does not advise. It will produce advice-shaped paragraphs on request, in a warm and reasonable voice, and that is the trap rather than the service. Said to a client in your voice, that paragraph is legal advice from nobody, and the rule it breaks has your name on it.

It cannot make the privilege call. Whether a client email to in-house counsel about a business decision, copied to three people, is privileged is a judgment whose consequences arrive eight months later in a motion. That call belongs to the attorney, and it does not belong to a chat window.

And it invents authority. Carry that one out of here if you carry nothing else. It does not garble citations, it manufactures them: complete, plausible, correctly formatted, with a holding that fits your facts a little too well. It also miscites real cases and quotes them for propositions they never reached. Every authority it hands you gets pulled and read in a real database before it goes near work product. Not skimmed. Pulled.

So much for the first story, and none of that answer turns on how good the technology gets, because not one of those things is a writing problem.

The unease underneath is still worth taking seriously; it is only aimed slightly off target. Volume reading is the part of legal support that software has been quietly eating for years, since well before anybody said the letters AI out loud. What it takes is work that follows a rule. What it leaves is exceptions, privilege, the chronology, the deposition, the client on the phone, and the person who can tell when an index is lying. Your job already lives in that pile.

Which argues for reading on. The paralegals who do well from here are quick, visibly careful, and known in the building as the person who understands where the tool fails. Chapters 7 and 9 are built for that.

Now the honest half, because a book that told you nothing was moving would be caught out by Chapter 4 and would deserve it.

Something is moving. It is not the judgment and it is not the case. It is every word written around the case.

The writing never comes up while you are studying for the certificate. Procedure, discovery, evidence, the Bluebook: that is the syllabus. Then you start, and most of the day turns out to be writing. The status update to a client who has heard nothing for six weeks. The records request to the third hospital this month, each custodian wanting its own wording. The memo to an attorney who reads the first two lines of anything. The deposition summary that has to be four usable pages by tomorrow. Nine hundred privilege-log descriptions that have to be phrased like each other. The chronology nobody has ever found six hours for. And the resume that still says you maintained the litigation docket.

None of that is judgment. Very little of it is worth what it costs you. Most of it happens after ten to five, or at home, or not at all.

Sentences are exactly what these tools are for. Not deciding what the deposition established: the four pages that let the attorney see what it established in four minutes.

They fail too, expensively, so the failures get named early. The inventing does not stop at citations. Vague dates get sharpened, so “early 2022” becomes January 2022 and then a specific false fact sitting in a working document. About four thousand pages becomes 4,127, because a number reads better than an estimate. A negation falls quietly out of the middle of a quote. A row appears in a table that fits so neatly you only catch it by counting the items in and out again.

So the workflows in this book all split the job at the same seam. The facts are yours end to end: dates, names, page:line cites, Bates ranges, quotes, authority, anything carrying a digit, checked against the file, the transcript, the database and the docketing system. Wording and shape are its side. Where the seam is easy to blur, a WATCH OUT box names the check. It costs about a minute, and that minute is the entire price of the thing.

One thing never crosses in either direction: deadlines. Not response dates, not court days, not a quick sanity check on a date somebody else computed. Your firm’s rules-based docketing system is the authority, and the chat window is not in that conversation.

The commercial part, once. Nobody in a law office notices your effort; they notice turnaround. The summary that lands the day after the transcript does. The update that reaches the client before the client reaches for the phone. None of that is new. What changed is the price of a first draft, which used to be an evening and is now ten minutes plus the checking you already do.

Take a Tuesday out of the month before a discovery cutoff. Nothing dramatic happens. That is what makes it typical.

The transcript landed Monday night, two hundred and some pages, and the attorney said whenever you get to it, no rush, which means Thursday. You give it an hour at half past eight and it goes well.

At half past nine the second production arrives, 4,100 pages, and the index that came with it does not match the Bates ranges. You spend until noon working out what is actually in there and which months are missing.

The afternoon belongs to the phone and the doorway. A client asking whether anything is happening. A custodian who needs the authorization again in a different format. Then the attorney, who wants the exhibit index re-sorted by witness before a nine o’clock meeting, and it is already three thirty.

So the summary gets written at home, across two evenings, from your own notes with your own page:line cites in them. Five hours, and it is good. Genuinely good: the attorney lifts two of the admissions straight into a motion six weeks later.

The expensive part of that week is the part nobody logs. Nine other clients heard nothing from anybody. On Thursday one of them calls, and the call runs forty minutes, and none of it is about anything that has actually gone wrong. It is about six weeks of silence. The chronology still does not exist either, so eight months from now somebody asks when the second notice went out and it costs a person another forty minutes.

None of that is a writing problem. You produced something excellent at ten at night after a bad day. It is a supply problem. Doing any one of those things properly takes about three hours, the week wanted nine of them, and Tuesday does not stretch.

So, plainly, the contents.

A working manual for using an everyday chat tool on the writing that surrounds your cases. ChatGPT, Claude, Gemini: whichever your firm’s policy allows. First drafts of routine documents, built into your firm’s own forms. Discovery shells with the objections framework blocked in for the attorney. The consistency sweep that catches the two dates that do not match. The status-update kit that ends the “any news?” calls. Deposition summaries from your notes, production logs, exhibit indexes, privilege-log phrasing, chronologies, workback checklists off verified dates. Published opinions broken down, legalese decoded, Bluebook mechanics. Your own career, too: the resume, the certification plan, the raise conversation. The 150 prompts are numbered, grouped by job, and listed again at the back for the days somebody is at your desk waiting.

What it is not: a research tool, and not an authority on anything — not a statute, not a rule, not local practice, not your judge’s standing order on AI, which you check per judge and per case. It does not compute deadlines. It does not give legal advice, and neither do you. It is not a place client information goes: no names, no case numbers, no fact pattern specific enough to be recognized. Fictional facts and [BRACKETS] go in; real details go on afterward, in your firm’s systems. Where the tool gets near something that could cost somebody a license, a RULES CORNER box shows where the fence runs. Chapter 10 collects them. Nothing here goes near a live matter until you have read that chapter and your firm’s own policy, which outranks this book.

None of it is technical. A phone or a laptop, an email address that is not the firm’s, thirty minutes; Chapter 2 walks you through it. Chapter 3 covers the single skill involved, and you already practice it: write a records request that comes back right on the first pass, or take an assignment and produce what the attorney actually meant, and you can brief one of these tools. That is not a compliment. It is the same discipline, aimed at a different reader.

Before you start, the size of the claim.

Nothing here changes your life, and no book ever changed anybody’s title. What it moves is the hour after five, and after a month or so, most of the week’s evenings. The page across from this one shows where the hours come from and totals about ten and a half. The number is honest, and it is not a gift: expect the first week to cost more than it returns. By the third week it is paying you back.

Most readers come out of a book like this using two of its workflows, and two was the design. Usually the deposition summary frame and the status-update kit. Work them weekly for a month, until reaching for them stops being a decision, leave Chapter 7 for a quieter stretch, take back four or five hours, and let the rest go. Nothing here awards points for reaching Chapter 12.

Your judgment was never the thing at risk. Somebody with a license still has to make the privilege call, and somebody with hands still has to pull the case and confirm it says what the draft claims. What is actually on the table is less than either story promised and worth more than both: the summary that lands the day after the transcript, the client who hears from you before anybody has to ask, and a chronology that exists.

Give Chapter 2 half an hour, and start on the next transcript that lands.