Clay, an AI sales and go-to-market platform, has introduced a company-wide AI writing policy for employees who use AI to draft text that will be seen by others.
Co-founder Varun Anand said the policy was originally written by Sophie Alpert on the engineering team. It first applied only to engineering, then expanded into a formal company-wide policy after other teams found it useful.
Every idea and every sentence must be owned by the author
Clay reduces the policy to four core principles. The first is: 「You must stand behind every idea and every sentence.」
In practice, that means any document carrying an employee’s name should reflect that person’s actual judgment, line by line. The policy says that if a reviewer asks what a sentence means, the author cannot respond by saying AI wrote it and they simply left it in.
Clay frames this as more than a question of AI hallucinations or factual accuracy. If the writer has not fully read, understood, and agreed with AI-generated text, yet still asks coworkers to spend time reading it, the company says that amounts to wasting the reader’s time.
For that reason, Clay is not asking employees to prove a document was written by a human. It is asking the person who submits the document to take responsibility for the content, whether AI was used or not.
Writing is part of thinking
The second principle is straightforward: 「Writing is thinking.」
Clay says the value of writing does not stop at the finished document. When someone decides which information matters, what deserves emphasis, how arguments should be arranged, and how to explain a complex issue clearly, that person is also working through the issue itself.
Under that view, internal documents such as Project Specs, progress reports, and Retrospectives can serve as a form of 「proof of thought.」 If the whole drafting process is outsourced to AI, the end result may still look polished, but the more important thinking process may have been skipped.
Anand summarized the point this way: spending time writing leads to deeper understanding of a topic, while bypassing that process entirely can leave someone with a weaker grasp of the problem.
Authoring time should exceed reading time
The third principle may be the most debated part of the policy: 「More time should be spent authoring a document than consuming it.」
Clay’s argument is simple. The time spent creating a document should, in principle, exceed the time required for someone else to read it. A company document is usually written once and read many times.
The policy gives a concrete example. If an employee spends 30 seconds entering a prompt, asks ChatGPT to generate a document that takes 10 minutes to read, and then sends it to 20 coworkers, the writer may have boosted personal output, but the team as a whole may have taken on 200 minutes of reading cost.
Clay criticizes that approach directly. If someone uses a very short prompt to produce a long document and then expects others to read it, the company says that does not respect their time. If coworkers want ChatGPT’s answer, they can ask ChatGPT themselves.
On the other hand, if the author spends extra time cutting filler and tightening the structure, reducing a 10-minute read to 3 minutes, that cost is paid once by the author while every later reader benefits.
Longer is not better
The fourth principle is: 「Longer is not better.」 Clay cites a well-known line from the French mathematician Pascal: 「I have made this letter longer because I have not had the time to make it shorter.」
Clay argues that generative AI amplifies this problem. In the past, writing a long document at least required time, which imposed a natural cost on the author. Now a single prompt can quickly produce a document with solid structure and a professional tone, even if much of it carries little information.
Anand said AI can easily produce very long documents, and one common way it does so is by filling them with sentences that effectively say very little, burying the points that matter.
Clay goes so far as to offer a blunt suggestion: if a long document comes from a very short prompt, it may be better to share the prompt itself with everyone.