The AI Learning Hub Journal

Redlining and First-Pass Markup

First-pass redlining, and where the lawyer stays in it1 INCOMING DRAFTThe counterparty's draft,unmarked, as received.2 COMPARED TO STANDARDChecked against the firm'sagreed positions and thefallbacks already approved.3 EDITS WITH RATIONALEEach proposed change comeswith a stated reason: whatit fixes and why it matters.4 DEVIATIONS FLAGGEDDepartures from standardare ranked by risk, notlisted alphabetically.5 LAWYER REVIEW — EVERY PROPOSED CHANGE GETS A DECISIONACCEPTthe rationale holds and theposition is one you wouldhave taken yourselfMODIFYthe direction is right, thedrafting or the strength ofit is notREJECTthe standard position doesnot apply here, or the editmisreads the dealWITH A RATIONALE PER EDITYou are checking a stated reason against your ownjudgement — fast, and it either holds or it does not.WITHOUT ONEYou must reconstruct the reasoning for every changebefore you can judge it — slower than no markup at all.The rationale is not documentation — it is the thing that makes review possible at speed
First-pass markup earns its time only when every proposed edit arrives with a reason a lawyer can accept, modify or reject

What a First-Pass Redline Actually Buys You

Applying a firm's standard positions to an incoming draft — swapping in preferred indemnity wording, adjusting a cap, inserting a missing governing-law clause — is mechanical work that consumes real hours. A model working from a playbook can produce a first-pass markup quickly, and the output is easy to inspect because a redline shows exactly what changed. That inspectability is the point. Unlike a summary, where an omission is invisible, a redline is a diff: every alteration is visible and attributable. The saving is in the typing and the lookup of standard wording, not in the decision about whether each change is right. Reviewing a machine redline is still a full review; it is just a review that starts from something rather than nothing.

  • Redlines are inspectable by construction — every change is visible as a diff
  • The saving is in mechanical application of known positions, not in deciding what position to take
  • Reviewing the markup is still a full review — it starts from a draft, it does not replace the read
  • Works best where standard wording exists and is genuinely standard across the firm

The Omission Problem

The hardest thing to catch in any redline, human or machine, is the change that was not made. A model that applies twelve playbook positions and silently misses the thirteenth produces a document that looks complete and reviewed. Nothing in the diff signals the absence. This is the same false-negative asymmetry that runs through all AI-assisted review, and in redlining it is particularly easy to miss because the presence of many careful-looking changes creates an impression of thoroughness. The countermeasure is a checklist that exists independently of the tool: the reviewer works down the firm's list of required positions and confirms each is addressed, rather than reading the markup and assessing whether it looks right.

  • A diff shows what changed; it cannot show what should have changed and did not
  • Visible thoroughness in the changes made creates false confidence about the changes missed
  • Countermeasure: check against an independent list of required positions, not against the markup itself
  • Track which positions the tool reliably misses — the gaps are usually systematic, not random

Tone, Negotiation Posture, and the Counterparty

A redline is a communication as much as a legal instrument. How aggressively you mark up a draft signals your position, and an appropriate markup for a routine supplier differs sharply from one for a strategic partner you will work with for a decade. Models default to applying the playbook uniformly, which produces markup that is technically correct and relationally tone-deaf. They also do not know what was agreed on a call, what the client has already conceded, or which points are worth spending capital on. Before any machine-generated markup leaves the firm, someone who knows the relationship should read it as a message to the counterparty and not only as a set of legal changes.

  • Markup intensity is a negotiation signal — uniform playbook application ignores the relationship
  • The model does not know what was agreed on a call or already conceded by the client
  • Read the redline as a message to the counterparty before it goes out, not just as legal edits
  • Deciding which points to spend capital on is judgement and stays with the responsible lawyer

Prefer slides, quizzes, and saved progress? Read this lesson in the library — free, no sign-up.