LinkedIn has a new button. Tap it under a post that reads like it came out of a prompt, and the platform logs it, feeds it into a classifier, and starts training its systems to recognize the next one. Hari Srinivasan, the company’s chief product officer, announced it on July 30 alongside a smaller change: LinkedIn is retiring the “enhance your post” feature that ran your writing through AI, replacing it with a plain proofreading tool that isn’t supposed to touch your voice. Srinivasan’s reasoning for why people leaned on the AI feature in the first place was almost sympathetic. People felt more confident running a post through it before hitting publish. The fix, in his telling, isn’t to shame that instinct out of people. It’s to stop nudging them toward it in the first place.
It’s not alone. All four platforms moved within a fortnight. Substack partnered with Pangram, a company that specializes in detecting AI-generated text, letting readers scan any post, note, or comment of 100 words or more. YouTube revised its payment rules to spell out three kinds of video it won’t pay creators for. On July 31, Snap announced it would stop recommending “wholly AI-generated videos” in Snapchat’s Spotlight feed, though AI-“enhanced or edited” clips remain fair game for the algorithm.
Despite sharing a similar headline, these items are all quite different, and those differences matter. Pangram’s scan works on text, not video. It also only applies to posts published after July 21. In the announcement, Chris Best stated plainly:
“Platforms that reward fakeness will create a race to the bottom.”
YouTube’s approach, spelled out by trust-and-safety chief Matt Halprin in a Creator Insider video, divides disallowed content into three buckets: generic and repetitive AI filler, material built to be “off-putting” or distressing for views, and AI personas dispensing health or financial advice. According to LinkedIn, the platform is now catching hundreds of thousands of automated comment attempts a day, and claims billions of other automation attempts blocked in the last couple of months alone, though that figure comes from LinkedIn itself, not an outside audit. Snap’s version is the narrowest of the four: a recommendation change, not a content rule, one that leaves its own AI editing tools completely untouched.
There’s a fifth platform that belongs in this story, and it’s the one that didn’t answer. Meta runs Facebook, Instagram, and Threads. When the BBC asked the company directly about AI-slop crackdowns, Meta didn’t offer a policy. It pointed reporters back to an earnings call instead. In it, Mark Zuckerberg described social media entering a “third phase” built on AI-generated content:
“as AI makes it easier to create and remix content, we’re going to add yet another huge corpus of content.”
He said nothing about restricting anything, not even a passing mention of slop. So “platforms are finally cracking down” holds for four companies and, as far as the public record shows, gets silence from a fifth, one that’s currently in the business of adding to the pile the other four say they’re trying to clean up.
YouTube itself is not entirely clean either. Gizmodo’s coverage of the policy update flagged the obvious complication: the crackdown is “made worse by the fact that the company itself seems to be investing in generative AI.” Restricting monetization for AI filler and building more AI tools isn’t necessarily contradictory in a strict sense. They simply sit oddly next to each other in the same press cycle. Google, YouTube’s parent, is not exactly playing down its AI ambitions elsewhere. A company can build the tools and police the output at the same time. It’s a stranger position than the announcements make it sound, however.
We need the reader-facing numbers to make sense of why any of this matters beyond corporate messaging. A YouGov survey run for Pangram, the same detection company Substack just partnered with (worth flagging, since Pangram isn’t a disinterested source here), found that 69% of Snapchat users say they’ve seen AI-generated content they believed was misleading. That number is specific to Snapchat. It didn’t come from the BBC’s own reporting on this story, and it lines up with what the app’s own users are apparently telling pollsters right now. Across platforms generally, two out of three people surveyed said the same. Trust drops further once you ask people to compare directly: asked whether they trust AI-generated content or human-generated content more, 69% pick human, full stop. Only 8% said the opposite. People aren’t confused about whether a problem exists. What they can’t tell is whether four companies’ announcements will change anything they actually see in their feed on Monday.
That question isn’t the kind a company statement is built to answer, and none of the four tried. Four platforms wrote new rules this month. One platform, asked directly, talked about its earnings instead.
