I used Claude to draft my client contracts, who owns them and am I exposed?

I run a small design studio, four people. Last winter I spent a weekend with Claude building our MSA, SOW template, and a deposit policy, iterated maybe thirty drafts, and honestly they read better than the $200 template pack I bought years ago. Eleven clients have signed them so far, zero complaints.

Then a founder friend looked horrified and hit me with three questions I couldn't answer: (1) who actually OWNS the contract text, me or the AI company, (2) can a competitor just copy my MSA word for word since AI wrote it, and (3) am I exposed in some way because no lawyer ever looked at it? Now I'm second-guessing the whole stack. Anyone actually know?

Small comfort on question 2: contract language has basically always been copied. Law firms recycle each other's clauses, template packs are remixes of other template packs, and nobody polices their MSA like a novel. Whether YOU can stop a copier was mostly theoretical even before AI.

The question that actually matters for your business is whether the thing HOLDS UP when a client refuses to pay a kill fee, not who holds copyright in the prose. I'd spend all my worry budget on question 3.

I looked into this for my Etsy shop and my understanding is AI contracts aren't legally binding, because a contract has to be prepared or at least approved by a licensed attorney to be enforceable in court. So your clients could probably walk away from those agreements whenever they want. You'd need a lawyer to re-issue them to make them real.

@diy_dana that's a myth, and worth putting to bed clearly. General information, not legal advice. There is no requirement that a lawyer draft or bless a contract for it to be enforceable. People bind themselves with handwritten one-pagers and email chains every day. If Mel's clients signed agreements with offer, acceptance, and consideration, those are real contracts regardless of what tool produced the words.

For OP's three questions: (1) ownership of the text is the least important one, and the AI provider is not going to claim your MSA. (2) As Ravi says, contract prose was never really a defensible asset. (3) is the real one, and the honest answer is that the risk isn't "AI wrote it," it's "nobody with legal training checked it against YOUR situation." The failure mode I see with generated contracts isn't bad grammar, it's terms that don't fit: an indemnity that's backwards for your side, an IP assignment that transfers work before final payment, a governing-law clause for a state you've never set foot in, no kill-fee mechanics for the way you actually bill. A generated draft plus one attorney review pass against your workflow is a genuinely solid setup. A generated draft that eleven clients signed unread by anyone is a set of small landmines that all cost nothing until one goes off.

Okay, that's simultaneously reassuring and a little terrifying, which feels accurate. Follow-up question before I book a review: during those thirty drafting iterations I pasted in a LOT of real details, client names, project budgets, one whole scope document from an active deal. Did I just breach confidentiality with my own clients by feeding their info into an AI tool? Two of my clients have NDAs with me.

General info, not legal advice. The answer lives in two documents: your NDAs and the tool's data terms. Many NDAs permit disclosure to "service providers" or "agents" under confidentiality obligations, and many don't address tools at all, so whether pasting deal terms into a third-party service was a breach depends on the actual clause language. On the tool side, the commercial tiers of the major AI products generally commit not to train on customer inputs, but that's plan-dependent and worth verifying on the live terms for the account type you actually have, not the account type you assume you have.

Going forward, the hygiene is simple: use a business tier with training disabled, redact names and numbers when the identity of the client isn't relevant to the drafting, and add an AI-tools line to your own engagement paperwork so it's disclosed rather than discovered. For the two NDA clients, read those NDAs before deciding whether anything needs to be said. There's a practical breakdown of these input/output issues in the AI output rights hub on this site.