When the Words Come From Outside: Authorship, AI and the Contracts Your Team Signs Off On

The Discomfort of Borrowed Language

A German rapper called Vega once wrote that he disappears for a year because he only writes when it hurts. The observation lands differently when you read it alongside the current debate about AI-generated text. The sting, as the Verfassungsblog essay observes, sits inside the act of writing itself, not in some external reward. That unease is worth sitting with, because in-house legal teams and law firms are now routinely signing off on contract language that nobody on the team actually composed.

This is not a complaint. Efficiency is real and the commercial pressures on legal departments are real. But the question of who authored a document, and whose judgment shaped its clauses, matters in law in ways it does not always matter in music.

Voice as a Legal Asset

Contracts are not poetry, but they do carry a voice. The way a company defines "material adverse change," the threshold it sets for automatic renewal, the jurisdiction it defaults to when a counterparty pushes back: all of these reflect accumulated institutional judgment. That judgment is part of what clients and counterparties are actually dealing with when they engage your organisation.

When AI generates language, the risk is not that the output is wrong in some obvious way. Competent legal AI will produce grammatically sound, legally plausible clauses. The risk is subtler: the output reflects the average of what exists in the training data, not your company's specific risk appetite, your preferred indemnity architecture, or the commercial context your procurement team has spent years negotiating.

This is precisely why Adira is built to draft in a company's own voice, trained on that company's existing contracts and playbooks rather than on a generic corpus. The point is not to produce language that sounds like everyone else. It is to produce language that sounds like you, with the legal accuracy of a specialist.

Reading Contracts From Your Side

There is a second dimension to the authorship question that receives less attention: when you receive a contract drafted by someone else's AI, who is reading it on your behalf?

A counterparty's standard terms, a supplier's master services agreement, a landlord's lease: these documents are now frequently generated at speed and volume. The clause you might once have spotted as unusual, because a lawyer had read hundreds of similar documents and developed an instinct for deviation, can now slip past a tired reviewer who is simply scanning for obvious red flags.

Reading contracts from your side means something specific. It means understanding your exposures, your obligations, your termination rights, from the perspective of your role in the transaction, not from a neutral midpoint that treats all parties symmetrically. A tool that applies the same analytical lens to every contract regardless of which party you are is considerably less useful than one that has internalised your position.

Jurisdiction Is Not a Dropdown Menu

The Verfassungsblog piece is published in German and sits within a European constitutional law context. That specificity matters, and it points to something AI contract tools frequently underserve: the law is not universal.

A limitation of liability clause that is perfectly enforceable under English law may be void under German consumer protection rules. A data processing addendum drafted to satisfy GDPR may not satisfy Swiss nFADP requirements. An automatic renewal clause that is standard commercial practice in New York may trigger specific disclosure obligations in California or specific form requirements in France.

Generic AI will produce generic answers. A system that actually knows the law of the jurisdiction it is working in, that understands not just the statutory text but the way courts in that jurisdiction have interpreted standard commercial clauses, provides something categorically different. The difference between a plausible clause and an enforceable one is, in practice, the difference between a contract and an expensive dispute.

What In-House Teams Should Actually Ask

The practical takeaway for legal operations leaders and general counsel is not whether to adopt AI in contract workflows. That question is already settled for most organisations. The questions that remain are more specific and more consequential.

Does the tool draft in your institutional voice, or does it produce something that sounds like a law school textbook? Does it read incoming contracts from your side of the transaction, or does it summarise neutrally? Does it understand the jurisdiction you operate in, or does it apply English common law defaults to a French law governed agreement?

These are not marketing questions. They are due diligence questions, and they deserve the same rigour your legal team would apply to any significant vendor relationship. The discomfort of borrowed language is manageable when the borrowing is done thoughtfully. It becomes a liability when nobody checks whose voice is speaking.