What Is Redlining? Structural vs AI Redlining Explained

Redlining is the mechanic of proposing tracked edits to a contract, striking through the words you want gone and inserting the words you want instead, so the other side can see exactly what changed and why. It is the actual work of negotiation: two sides do not agree a contract by talking about it in the abstract, they agree it by editing a document back and forth until both sides accept the same text. This guide (published by Adira, which sells contract software including AI-assisted redlining, so read the "where AI redlining goes wrong" section below with that in mind) walks through three levels of redlining, manual track-changes, structural redlining, and AI redlining, what each is good and bad at, and why the version you sign, not the redline trail that got you there, is what a court reads if the deal goes wrong.

What redlining actually means

A redline is a proposed edit, not an agreed one. If clause 8.2 currently caps liability at fees paid and you want it capped at twelve months' fees, you do not delete the clause and email a complaint, you redline it: strike "fees paid under this Agreement" and insert "the fees paid under this Agreement in the twelve (12) months preceding the claim," visibly, so the other side can accept, reject, or counter-redline that specific change. The name comes from marking paper drafts in red ink; Word's Track Changes, Google Docs' suggestion mode, and dedicated contract tools all do the same thing electronically, insertions underlined, deletions struck through, each change attributable to whoever made it.

The point is precision and a paper trail. A verbal "can we make the cap higher" gets lost or misremembered. A redline is unambiguous: here is the exact sentence, here is what it becomes, here is who proposed it and when.

The three levels of redlining

Manual track-changes is redlining done inside a Word document (or Google Docs), by hand, clause by clause, using the software's built-in tracked-changes feature. This is still how most contracts in India get negotiated. It works, it is free, and every lawyer knows how to read it. Its weaknesses are about discipline: someone forgets to turn tracking on before editing, someone accepts all changes before sending back "clean," a change gets buried in a 40-page document nobody scrolls to. The mechanics, including how to turn tracking on and off correctly, are here: how to redline a contract.

Structural redlining means editing the contract at the level of a clause object, not a blob of text. Instead of the document being one long string a word processor happens to display in paragraphs, each clause is a distinct, addressable unit the software understands: "this is clause 8.2, Limitation of Liability, with a defined cap value." Editing at that level means the software can show precisely which clause changed, stop an edit from silently breaking a cross-reference elsewhere in the document, and let a reviewer jump straight to "show me every clause that changed." Why treating a contract as structured data rather than flat text unlocks this is explained in full here: structured clause tree vs flat-text contracts.

AI redlining goes one step further: the tool does not just help you display a clean edit, it proposes the edit itself, checking a clause against your playbook (your standard position and fallback for that clause type) and suggesting the change that brings it into line. Paste in a vendor's indemnity clause, the tool recognises it is uncapped and proposes a cap tied to fees paid, in your house style, in seconds, real time saved on the repetitive clauses that always get the same treatment.

Redlining vs annotation vs a diff

These three get confused constantly and they are not the same thing.

Annotation is a comment or highlight layered on top of a document that does not change the underlying text, a sticky note, not an edit. Highlighting clause 8.2 and writing "check this" in the margin is annotation; the other side has to read it and make the change themselves. Marking up a PDF for free, without touching its underlying text, is exactly what Weave, Adira's free browser-based markup tool, is built for, a genuinely different job from redlining.

A diff (or "compare") is a report generated by comparing two versions of a document, after the fact, whether or not tracking was ever turned on. A diff is how you catch changes someone made without leaving Track Changes on, which happens more often, deliberately and accidentally, than most people expect. It describes what changed; it does not, by itself, invite anyone to accept or reject anything.

Redlining is the live, two-way negotiation mechanic: a proposed edit sitting inside the document, waiting for the other side's accept, reject, or counter-edit. Annotation flags a problem, a diff reports a difference, redlining is the mechanism by which two parties actually change what a contract says.

Where AI redlining goes wrong

An AI redline tool pattern-matches against your playbook and, often, against a general model of "what contracts usually say." Two failure modes matter more than the general "AI can be wrong" caveat.

First, a plausible-but-wrong edit. The proposed wording reads fluently and looks exactly like something a lawyer would write, which is precisely why it is dangerous: a redline that looks wrong gets caught, one that looks right and is subtly wrong (a cap referencing the wrong defined term, a notice period contradicting a related clause) sails through.

Second, and specific to India, an AI redline trained mostly on US or UK contract patterns can propose language that is simply void here. The clearest example is a post-employment non-compete: a tool asked to "tighten" a restrictive covenant might broaden it, worldwide, indefinite, because that reads as stronger protection in a US frame. Under Indian law it makes no difference how the clause is worded. Section 27 of the Indian Contract Act, 1872 states:

"Every agreement by which any one is restrained from exercising a lawful profession, trade or business of any kind, is to that extent void." Source: Section 27, Indian Contract Act, 1872 (Indian Kanoon)

A redline that makes a post-employment restraint broader is not making the contract stronger, it is proposing a clause an Indian court will not enforce regardless of wording. A deeper walk-through of how far Section 27 reaches is here: are non-compete clauses enforceable in India. The lesson extends beyond non-competes: any AI redline needs a human who actually knows Indian contract law checking the output, not just checking that it reads well.

The Indian position: why the final signed text is what a court reads

There is a second, less obvious reason redlining precision matters, and it sits in evidence law, not contract law. Once a contract is signed, the negotiation trail behind it, the redlines, the email chains, the "we agreed on a call that..." conversations, generally cannot be used to add to, vary, or contradict what the final signed document says. Section 95 of the Bharatiya Sakshya Adhiniyam, 2023 (BSA), India's evidence law since 1 July 2024, which replaced the Indian Evidence Act, 1872, states the rule plainly:

"No evidence of any oral agreement or statement shall be admitted, as between the parties to any such instrument or their representatives in interest, for the purpose of contradicting, varying, adding to, or subtracting from, its terms." Source: Section 95, Bharatiya Sakshya Adhiniyam, 2023 (bare act text), full Act: India Code

Section 95 is the direct successor to Section 92 of the Indian Evidence Act, 1872, which said the same thing for over 150 years before the BSA renumbered it; the underlying rule is not new. The Supreme Court applied it to commercial contract interpretation in Nabha Power Ltd v Punjab State Power Corporation Ltd, (2018) 11 SCC 508, a bench of Justices R.F. Nariman and Sanjay Kishan Kaul, deciding a tariff dispute under a power purchase agreement. The Court held, citing the English case Investors Compensation Scheme v West Bromwich Building Society, that "the law excludes from the admissible background the previous negotiations of the parties and their declarations of subjective intent," and concluded that "normally a contract should be read as it reads, as per its express terms." See the full judgment on Indian Kanoon.

The practical result: if a redline you sent gets rejected, or a comment explaining "why we need this" never makes it into the final clause, an Indian court will not go looking through your email thread to figure out what you meant. It reads clause 8.2 as it stands in the signed PDF. This is why structural, precise redlining, where the accepted wording actually says what you intended, matters more than a well-argued cover email ever will. A redline that makes its point in a comment but never lands cleanly in the accepted text has, legally, achieved nothing.

Red flags when redlining a contract

Normal	Red flag	Why it matters
Track Changes stays on for every round, both sides	A "clean" version arrives with no tracked history of what changed since the last round	You cannot verify what moved without a full document compare, and something usually did
Each substantive redline carries a one-line comment explaining the business reason	Redlines arrive with no comments, just silent edits to numbers or defined terms	An unexplained change to a cap, date, or party name is the change most worth catching, and the easiest to miss
An AI-suggested redline cites the exact clause or playbook rule it is applying	An AI tool proposes a rewritten clause with no source or reasoning shown	A plausible-sounding rewrite with no citation is exactly the failure mode you cannot audit after the fact
A restrictive covenant redline is scoped to a defined period and territory	A redline "strengthens" a non-compete by widening scope or removing the time limit	Broader wording does not make a post-termination restraint more enforceable in India, Section 27 voids it regardless of scope
One master document is edited and re-sent each round, version-numbered	Parallel copies circulate by email, each side editing a different attachment	Redlines against the wrong base version create silent conflicts nobody notices until signature
The final "clean" version is diffed against the last redlined draft before signature	The clean version is trusted without a final compare	An accept-all step can silently drop or misapply a change that was actually agreed
A rejected redline stays marked rejected, reason kept in the negotiation record	A rejected point is just deleted, with no record anyone asked for it	Section 95 means a comment thread will not save you in court, so the record matters mainly for your own audit trail

A redline comment: bad versus better

Take a real limitation-of-liability clause, being negotiated by the customer's side.

Original clause: "Vendor's total liability under this Agreement shall not exceed the fees paid by Customer in the preceding twelve (12) months."

Bad redline: The customer strikes "twelve (12)" and inserts "twenty-four (24)" with no comment, in a document where a separate audit-rights clause still references "the preceding twelve (12) months." No explanation is given for the change, and the related clause is left untouched.

What is wrong: a silent numeric change is the hardest thing for a busy reviewer to catch on a second pass, and leaving a related clause on the old figure creates an inconsistency nobody notices until it matters.

Better redline: The customer strikes "twelve (12)" and inserts "twenty-four (24)," and adds a comment: "Aligning the liability lookback with our standard 24-month audit window in clause 11.3, so both periods match. Happy to discuss if this affects pricing." The redline also flags clause 11.3 for the vendor's attention instead of leaving it silently misaligned.

What changed: the edit is now easy to spot and easy to negotiate on the stated commercial ground, and the related clause is not left contradicting the one that just moved. This is what "structural" redlining protects against, a change in one place breaking a promise made in another.

How redlining fits with your playbook and clause library

Redlining does not happen in isolation. What you redline to is your playbook, your pre-agreed ideal and fallback position for each clause type; what you redline from is often a clause pulled from your clause library, approved standard wording you swap in rather than draft fresh. Structural redlining makes both usable at scale: on flat text, "find every uncapped indemnity across our last 200 vendor contracts" is a manual search project; on a structured clause tree, it is a query. Why that data model matters, beyond redlining, is here: structured clause tree vs flat-text contracts.

US and global contrast

The mechanic is identical everywhere Word and Track Changes are the standard drafting tool, US and UK lawyers call the same thing "redlining" or "blacklining" and negotiate the same way. What differs is the evidentiary stakes. US law recognises a broadly similar parol evidence rule, extrinsic negotiation evidence generally cannot vary a fully integrated written contract there either, but the exceptions get argued more flexibly in practice than the narrower proviso list under Section 95 of the BSA allows for. The safer assumption for any contract governed by Indian law is the harder one: what is not in the final signed clause, functionally, was not agreed.

FAQ

Is redlining the same as marking up a PDF with comments? No. Marking up a PDF with highlights and comments is annotation, it flags something for the other side to fix themselves, but does not change the contract text. Redlining is a tracked edit to the actual words, ready to be accepted or rejected.

Can I trust an AI-generated redline without checking it myself? No. Treat it as a fast first draft of the edit, not the edit itself. Check that it cites the source rule it is applying, check the wording against your actual position, and for any Indian-law issue like a restrictive covenant, IP assignment, or stamping, verify it separately; a tool trained on global patterns can propose language that looks stronger but is unenforceable here.

If the other side deletes something from my redline without telling me, will I notice? Not reliably, by eye. Run a document compare (a diff) against your last sent version every round, and do not trust an "accept all" or a "no changes" claim. A full walk-through is here: how to check what the other side changed.

If we agreed something by email but it never made it into the signed contract, does the email still count? Generally no. Under Section 95 of the Bharatiya Sakshya Adhiniyam, 2023, an Indian court reading a signed contract will not normally accept email or negotiation evidence to add a term the final document does not contain, subject to narrow exceptions like fraud. Get anything you actually agreed into the signed text.

What does Adira do with AI redlining specifically? It checks a clause against your stored playbook and proposes a structural edit in your house style, aimed at cutting time spent on standard, repetitive clauses. It still requires a human sign-off round, and Adira does not train its models on customer contract content. Adira's paid plans (Practice $89 to $109 per seat per month, Firm $179 to $219, Enterprise on custom pricing, 7-day trial, as published on adiralaw.com, last verified 4 September 2026) are one option among several redlining tools on the market today.

This guide explains what redlining is and how manual, structural, and AI-assisted redlining differ. It does not tell you whether a specific redline is safe to accept, whether an AI-suggested edit is enforceable, or whether your negotiation record would help you in a dispute, that depends on the exact clause and facts this page cannot see. For that, get a lawyer to look at the actual document before you sign.