Perspective 3 min read

The limits of plain English: preserving terms of art in automated drafting

General-purpose models treat legal terms of art as poor writing that needs to be simplified. We train our systems to know the difference between archaic boilerplate and necessary precision.

2026 · 11 · 01·admin

The movement towards plain English in commercial contracts has largely been a positive development. Removing “witnesseth”, “hereinafter”, and archaic preamble language makes agreements easier to read without altering their legal mechanics. However, when software is introduced to assist with contract review and drafting, this preference for simplicity can become a liability.

By default, general-purpose language models over-correct. They are trained on human preference data that rewards brevity, approachability, and conversational flow. When applied to legal documents, they often treat established terms of art as poor writing that needs to be fixed. Our perspective is different. Plain English stops where legal precision begins.

The taxonomy of indemnification

Consider the standard phrasing: “indemnify, defend, and hold harmless.” A general commercial model will frequently attempt to compress this triad into a single verb, such as “compensate” or “reimburse.” To a copy editor, this is a cleaner sentence. To a transactional lawyer, it is a catastrophic failure in drafting.

In many common-law jurisdictions, these three verbs represent distinct obligations. The duty to defend usually triggers immediately upon a third-party claim, requiring the indemnifying party to cover legal fees as they are incurred. The duty to indemnify typically only triggers upon a final judgment or settlement. To hold harmless is often interpreted as a release of liability, preventing the indemnifying party from seeking contribution from the indemnified party.

Collapsing these distinct obligations into one generic verb alters the fundamental risk allocation of the agreement. When we trained ZAAN-7B Counsel, we had to heavily penalise the model’s instinct to summarise. In Review 4.2, when a firm’s playbook requires a full indemnification suite, the absence of the word “defend” is flagged as a High severity issue. It is not treated as a stylistic quirk. The proposed inline redline in Word restores the specific triad of verbs, complete with the appropriate citation to the firm’s standard position.

The efforts staircase

Another area where linguistic simplification fails is the hierarchy of contractual efforts. To a linguist, “best efforts”, “reasonable efforts”, and “commercially reasonable efforts” are near synonyms. To a litigator, they represent a steep staircase of liability and expected expenditure.

Under Delaware law, for instance, a “best efforts” standard may require a party to take all reasonable steps to achieve a goal, potentially even at a financial loss. “Commercially reasonable efforts” allows a party to consider its own economic interests and constraints. If a firm’s playbook dictates that its client will only commit to commercially reasonable efforts for securing regulatory approvals in a Share Purchase Agreement, Review must catch any drift toward “best efforts”.

This requires exact matching of terms of art. When a user queries Recall 2.7 for precedents on efforts standards, the system relies on an exact semantic map of these phrases. It does not conflate them. If an associate uses Draft 3.9 to generate an ancillary agreement, the deal-point form forces a selection from this specific hierarchy. This ensures the generated text uses the exact phrase required by the jurisdiction and the firm’s risk profile, rather than a generic approximation.

Intentional ambiguity as a deal mechanic

The push for absolute clarity assumes that ambiguity is always a mistake. In transactional practice, ambiguity is frequently a negotiated compromise. When parties cannot agree on a strict numerical floor for a material adverse effect (MAE) clause, they rely on the established, albeit vague, case law surrounding the term. They choose ambiguity to get the deal signed, deferring the exact interpretation to a future tribunal if a dispute arises.

If a contract-intelligence platform insists on quantifying every variable or defining every broad term, it breaks the negotiation dynamic. A system that flags an undefined “material adverse effect” as an error misunderstands

See it on a contract you have already reviewed.

Send us a draft your team has already redlined and we will show you what ZAAN catches, and what it misses.