More Companies Are Launching AI Writing Policies as the Slop War Heats Up

There is a new kind of document circulating on LinkedIn, and it has started to behave like a genre. A company publishes a short set of principles about how its people are allowed to use AI to write. A marketing or engineering leader posts it. It gets thousands of reactions. Another company reads it and ships its own.

This week it was Heidi Health's turn. The Melbourne clinical-AI company, which automates documentation and form-filling so clinicians can focus on patients and now runs more than two million consults a week across 116 countries, published an AI writing policy built on six principles. Among them: I think, therefore I write; AI is your thought partner, not your writer; transparent attribution is essential. The person who posted it, Heidi Health's global head of communications Emma Mackenzie, made no claim to having invented the form. "We're not the first company to do this, and we won't be the last," she wrote. That sentence is the tell. The AI writing policy now arrives with an awareness that it is one of many.

The slop that pays for it

The demand is not mysterious. Last September, researchers at BetterUp Labs and Stanford's Social Media Lab gave the problem a name: workslop, a deliberate play on "AI slop," defined as AI-generated work that masquerades as good work but lacks the substance to move a task forward. In a survey of 1,150 full-time US desk workers, 40% reported receiving workslop in the previous month. Each incident cost the recipient close to two hours to sort out, which the researchers costed at roughly $186 per employee per month, or more than $9 million a year for a 10,000-person company. The damage was social as well as financial: 42% said they trusted a colleague less after receiving low-quality AI output from them.

Most of that slop was manufactured on purpose. Through 2025, the loudest corporate story about AI was the mandate: use it more, use it faster, prove you used it. The bill is now arriving. Shopify's Tobi Lütke, one of the movement's most quoted champions, recently warned that staff who pass along unexamined AI output are lobbing "slop grenades" that leave colleagues to clean up the mess. Duolingo, another AI-first standard-bearer, found that around 20% of the AI-generated stories it produced was pure workslop that had to be filtered out. The writing policy is what the second act looks like: the same companies discovering that "more AI" and "more usable work" are not the same instruction.

Four companies, one document

The template was set by Clay. Its policy was written by engineer Sophie Alpert for the engineering team, adopted by other departments once they saw it work, and made company-wide by co-founder Varun Anand in August. His post drew nearly 10,000 reactions and more than 700 reposts. Its four principles have become the load-bearing structure that later versions borrow from: stand behind every sentence, writing is thinking, spend more time writing than the reader spends reading, length is not a virtue.

The borrowing is visible. Amsterdam's Polarsteps, the travel app with roughly 100 employees, released internal AI writing principles earlier this year; CEO Clare Jones lets staff brainstorm and draft with AI but requires them to apply their own thinking, disclose when they've used it, and make sure the final copy reflects their own reasoning and voice. Jones's reasoning for writing anything down at all is the most honest line in the whole trend: simply asking people to stop had not worked, and a formal policy gives everyone a shared reference point instead of a leader repeating themselves. The Berlin- and New York-based HR software firm Leapsome rolled out its own version in mid-August for around 130 staff, prioritizing clarity and requiring staff to stand behind every statement. Tellingly, some Leapsome employees asked the company to go further and restrict AI in peer-to-peer messages and marketing copy.

Read four of these back to back and the convergence is the point. Disclosure, ownership, "writing is thinking," AI as assistant rather than author: the genre already has conventions.

The people writing them are selling AI

The recurring irony is that the companies drawing these lines are, disproportionately, AI companies. Clay's platform helps thousands of companies research accounts and personalize outreach at scale, with AI doing much of the work. Heidi Health's entire product generates clinical documents. It exists to write the note so the doctor doesn't have to.

The line they're all drawing falls in the same place: between the product and the person. Heidi Health's policy carves out AI as a genuine equalizer, pointing at people writing in a second language, or anyone frozen by a blank page, while refusing to let the tool stand in for a colleague's actual thinking. Polarsteps still has professional in-house writers produce its consumer-facing copy. The machine can assemble, draft and personalize; what it cannot do is be the author of record for something a human is supposed to have thought about. Whether that distinction survives the commercial pressure to ship more, faster, is the open question hanging over all of it. Each of these companies has at least committed the line to writing, which is more than most of the industry has done.

The policy is the marketing

For a brand audience, the most useful thing to notice is where these documents live. They are not buried in an employee handbook or routed through legal. They are published on LinkedIn, by marketing and comms and engineering leaders, built to travel. Heidi Health's was written by Emma Mackenzie. Clay's became one of the most-shared brand artifacts the company produced this year.

That makes the AI writing policy a new kind of marketing object. In a market where every competitor's site, deck and outbound email is converging toward the same frictionless AI register, publishing a credible standard for not sounding like that is a positioning move. It says: a human wrote this, and a human will answer for it. The risk, of course, is the same one that shadows any values statement: it's cheaper to publish the principles than to live by them, and a document about resisting slop can itself be a very polished piece of slop. The companies that will get credit for this are the ones whose published writing, six months from now, still reads like a person had to sit down and think. The rest will have added one more artifact to the pile the policy was supposed to shrink.