This week, Varun Anand, Clay's co-founder, announced that the sales platform has made its AI writing policy official company-wide. Engineer Sophie Alpert wrote it as a rule for the engineering team first, and other teams asked for it once they saw it work. It now covers the whole company.
If someone asks what a line means, you can't say "AI wrote that, ignore it" and call it a day.
The policy runs on four principles:
Stand behind every sentence: the writer owns every idea in a document before it goes out.
Writing is thinking: skipping the drafting process leaves the writer with a weaker grip on their own argument.
Respect the reader's time: a document should take longer to write than it takes to read.
Length is no virtue: AI can pad a document with sentences that say nothing.
The policy says verbatim AI text is fine to share, as long as it's marked as AI's own words. Everything else has to survive a question from a reviewer.
At GC AI, we're the legal AI platform built for in-house teams, used by 2,000+ legal teams worldwide. Our CEO, Cecilia Ziniti, ran legal at Anki, Bloomtech, and Replit before she started the company.
She found the gap while teaching ChatGPT classes to lawyers: they needed precision, a way to check citations against their ethical duties, and a tone built for the job, and generic AI had none of that. That gap became GC AI. Our customers deal with accountability for AI-drafted work every day, long before Clay needed a rule for their own.
Why Legal Recognizes This Rule
Everybody covering this keeps pointing out what's missing: no software that flags AI-written text, no form to declare it, no compliance sign-off. That's accurate, and it's also why the policy spread the way it did.
Most compliance rules get followed because someone enforces them. Clay's rule spread because employees wanted it, as it improved the work.
Clay stumbled into this rule through trial and error. In law, there's no room for trial and error, so the same rule had to be built on purpose from day one.
One thing Clay's policy doesn't cover: what data goes into the prompt in the first place. Samsung's engineers found that out in 2023, when they pasted proprietary source code into ChatGPT to move faster, and the company banned generative AI tools across the company twenty days later. Legal teams already have their own rules for that half of the problem.
Regulators just caught up to the writing side of the same instinct. The EU AI Act's Article 50 took effect August 2, eight days before Clay's announcement, and it requires exactly what Clay's policy asks employees to do voluntarily: mark AI-generated content so people can tell what's synthetic.
Legal reached the same place even earlier, forced there by a real case. In United States v. Heppner (S.D.N.Y. 2026), a court found no privilege in a defendant's own AI-drafted defense memos. Someone has to answer for an AI-drafted line, and a courtroom asks that question with much higher stakes than a Slack thread.
Why This Is Spreading Past Legal Teams Too
At GC AI, the first person to see a legal question these days is often a marketer. Or someone in procurement. Or a rep trying to close a deal before the quarter ends.
About a third of the people using GC AI aren't lawyers at all. They work in marketing, procurement, HR, and sales, and they're the ones who often run into a legal question first, before it ever reaches a lawyer.
That changes the lawyer's job as much as the marketer's. You stop being the bottleneck. You start being the guardrail. Clay's writing policy is the same shift, showing up in internal memos this time: when a marketer, an engineer, or a salesperson can generate a document in seconds, someone still owns every line that ships under their name.
The same standard has always applied to legal work: accuracy, real citations based on analysis, privilege, and the right voice. Writing accountability is becoming the job everywhere else, too.
Where Accountability Gets Checked
Legal teams already have a fast way to check an AI-drafted line. Exact Quote, a GC AI feature, pulls word-perfect quotes straight from the source document, so a reviewer can check a cited line against the contract or filing instead of taking someone's word for it. It is the same instinct behind Clay's reviewer question, built into the review step instead of left to memory.
Wherever AI drafts the first version now, someone still has to answer for what ships, in a Slack memo or a signed contract. Clay figured that out this week. Legal figured it out years ago.
Frequently Asked Questions
What Is Clay's AI Writing Policy?
Clay's AI writing policy is an internal company rule that engineer Sophie Alpert wrote and co-founder Varun Anand announced company-wide in August 2026. It lets employees brainstorm, draft, and proofread with AI, but it requires them to stand behind every sentence they share. A reviewer who questions a line cannot be told the AI wrote it.
Does an AI Writing Policy Belong in a Legal AI Acceptable Use Policy?
Yes. Clay's policy and a legal AI acceptable use policy both come down to the same requirement: a person reviews the output and answers for what ships under their name. A legal AI acceptable use policy adds the parts, like data classification and privilege, that go beyond what a writing-only policy like Clay's covers.
