Your AI Posts Just Lost Reach on Two Big Platforms

If you run a shop and you schedule a week of posts on a Sunday evening, two of the platforms you post to changed the rules underneath you in the last days of July 2026. Neither change was announced as a ban, which is why it is easy to miss, and why the effect shows up as a quiet drop in reach rather than a warning in your inbox.

Snapchat and LinkedIn moved within twenty four hours of each other. Both moved in the same direction. And both moves fit a pattern that YouTube and Google Search had already written into policy well before this week, which is what makes this worth reading rather than skimming. Four separate companies have now converged on one distinction, and the distinction is not the one most people assume.

What did Snapchat and LinkedIn actually change?

Snapchat adjusted its recommendation system so that only videos made by real people are eligible to be recommended in Spotlight. Videos that are entirely machine generated no longer get surfaced to strangers. TechCrunch reported the change on 31 July 2026, quoting Snap's stated goal of keeping Spotlight a place where people find authentic creativity from real people rather than machine output. The nuance matters more than the headline: a creator who films something and then uses AI tools to edit, cut or enhance it is untouched. That is the same line US copyright draws between a tool that processes your work and a system that generates it, which we set out in the piece on what human authorship protects in an AI assisted track. The restriction lands on clips that are AI from end to end.

LinkedIn moved the day before. It added a report option in the dropdown on every feed post labelled "seems like AI slop", which hides the post from the person who flagged it and feeds the signal back into LinkedIn's own detection. TechCrunch's report on LinkedIn's new reporting button and its automation defences quotes chief product officer Hari Srinivasan describing three parallel efforts: the report button, blocking hundreds of thousands of automated comment attempts each day, and new classifiers that pull machine written low quality material out of recommendations.

The part that will surprise anyone who has been using LinkedIn's own tools: the company is retiring "enhance your post", the button that rewrote your draft for you, and putting a proofreader in its place. The stated reason is that a proofreader keeps your voice while a rewriter replaces it. Social Media Today's coverage of LinkedIn limiting the reach of AI written posts quotes vice president Laura Lorenzetti on the reasoning, that heavy automated use of AI dilutes the value of real conversation. The same piece notes the obvious tension, which is that LinkedIn has spent two years adding AI features to the product it is now policing.

Is this actually new, or has it been coming?

It has been coming, and two of the four policies involved are over a year old. In July 2025 YouTube renamed its "repetitious content" rule to "inauthentic content" and spelled out that it covers mass produced or repetitive material. The policy sits inside the YouTube channel monetization policies, and the sentence that should stop any creator using a video generation pipeline is the one about scope: if a channel carries videos that break the rule, or if YouTube cannot clearly tell that you made the content, monetisation can be pulled from the whole channel rather than the offending uploads.

Google Search went further and earlier on the open web. Its spam policies for Google web search define scaled content abuse as generating many pages whose primary purpose is manipulating rankings rather than helping people. The definition is deliberately agnostic about how the pages were made. Automation, humans, or any mix of the two all count. Generative tools appear in the examples, but so do scraped feeds run through a synonymiser and pages stitched together from other pages.

Read those two alongside this week's two announcements and the shared logic is visible. Nobody is measuring whether a model touched your work. They are measuring whether anything in the output required you.

The four policies side by side

These were written by different companies for different products, and no single announcement puts them together. Laid out in one table, the shared shape is hard to miss.

One column in that table is doing most of the work. Look at what stays fine across all four rows and you get a single sentence: the platform is happy for a machine to help you say your thing, and unhappy for a machine to be the thing.

Should a small shop stop using AI to write posts?

No, and reading these policies as an instruction to stop would cost you time for no benefit. Not one of the four forbids using a model. Snapchat explicitly protects AI editing. LinkedIn is replacing one AI feature with a different AI feature. YouTube's rule predates the current generation of video models and was originally aimed at people uploading slideshows of stock footage.

What changes is where the AI sits in your process. Drafting with a model and then rewriting the draft with what you actually know about your customers stays safe, because the finished post carries something the model could not have produced. Pasting a model's output unedited into twelve platforms does not, because it carries nothing. Our own look at how much editing an AI draft needs before it ships lands on the same practical answer from the quality side rather than the policy side, which is a reasonable sign the answer is right.

The awkward case is the middle: a merchant who genuinely has a point of view but no time, so the model writes and they skim before posting. That is the behaviour these classifiers are built to catch, and it is worth being honest that it is common. The fix is not more editing passes. It is putting one concrete thing into each post that only you have. A number from your own sales, a customer question you got twice this week, a photo of the thing on your bench.

How would you know if your posts are already being suppressed?

You mostly would not, which is the uncomfortable part. None of these systems sends a notice. LinkedIn is the exception and only partially: Srinivasan described a private dashboard signal that warns you when your own content is reading as inauthentic because of heavy AI use. Everywhere else the symptom is the same as a hundred other causes, which is that reach drops and nobody tells you why.

Three patterns are worth checking in your own recent posts, because they are the ones the classifiers were described as targeting. First, posts whose structure is identical week to week. Second, the formulaic contrast construction that generic AI writing falls into constantly, where a sentence sets up something you did not think and then corrects it. Third, comment activity you did not personally write, which LinkedIn says it is blocking at a scale of hundreds of thousands of attempts a day.

If your last ten posts could be shuffled between five different businesses without anyone noticing, that is the signal. A detection system does not need to know a model wrote it. It only needs to notice that nothing in it is specific.

What this means for your own website, not just your feed

The Google policy is the one with the longest tail for a merchant, because it applies to the pages you own rather than the posts you rent. A shop that generates a description for every product with a template and a model is not automatically in breach. A shop that generates four hundred near identical location pages or category pages to catch long tail searches probably is, under the plain wording of scaled content abuse.

The test Google states is purpose: were these pages made mainly to move rankings. A generated product description for a product you genuinely sell, sitting on a page a customer will read before buying, has an obvious answer. Four hundred pages targeting town names you do not trade in has a different one. We wrote about the mechanics of how AI content detection has changed for shops earlier this year, and the conclusion holds: the detectable thing was never the model, it was the absence of anything specific.

If you are building your storefront now, this is a reason to keep control of the pages themselves rather than accepting whatever a hosted platform generates for you. That is part of why our ecommerce website builder writes real files into your own repository instead of rendering pages from a template you cannot see or edit. When a policy changes, you can go and change what is on the page.

Which other platforms have already moved?

Snapchat and LinkedIn are the ones that moved this week, but they are not the only ones that moved this year. The same TechCrunch report on the Spotlight change lists the wider set: YouTube clarified its monetisation rules against inauthentic uploads, Meta pulled an Instagram photo editing feature built on AI, and Substack built tooling to identify newsletters written by a model. Four different product categories, four different mechanisms, one recurring judgement.

For a merchant the pattern is more useful than any individual policy, because it tells you where the next one lands. Every platform whose value depends on people wanting to read what is on it now has a commercial reason to suppress material produced faster than people can produce it. That includes the places you have not thought about yet: marketplace listings, review responses, and the email inbox, where the same economics apply and the same filters are being built.

It also tells you which platforms will not move. Paid placement is unaffected by every rule described here. If your post is an ad, nobody is checking whether a model wrote it, they are checking whether you paid. That asymmetry is worth sitting with, because it quietly changes the arithmetic between organic and paid for anyone whose organic strategy was volume.

What does "generic" mean in practice?

LinkedIn's classifiers are aimed at posts lacking a clear point of view, which sounds subjective until you look at what the descriptions of the target actually cover. Recycled thought leadership with no source. Openings engineered purely to stop a scroll. And the sentence pattern where a claim is set up only to be immediately corrected by its opposite, a construction generic AI writing produces at a rate no human writer does.

The reason that last one is so detectable is structural rather than stylistic. It is a shape a model reaches for when it has nothing specific to say and needs the sentence to feel like an insight. A person with a real opinion states it once. If you want a single edit that moves a post out of the flagged bucket, deleting every sentence of that shape is the highest yield one available, and it takes under a minute.

What changes in your week, concretely

The realistic response to all four policies fits into a short list, and none of it requires new tools.

• Stop scheduling identical posts across platforms. The uniformity is the signal, more than the writing itself. One post reworked per platform beats five copies.
• Film something. On Snapchat this is now the difference between being recommended and not. A phone clip with an AI edit on top is fully eligible; a fully generated clip is not.
• Put one unshareable fact in every post. A price you changed, a supplier delay, a question a customer asked. This is the cheapest possible defence and it also happens to be what people engage with.
• Turn off automated commenting entirely. LinkedIn is blocking it at scale and treating it as an abuse signal against the account doing it.
• Audit any bulk generated pages on your own site. If a page exists only because a keyword exists, it is the exact case Google's spam policy names.

Notice how little of that is about writing. Most of it is about supply: where the raw material for a post comes from. A merchant has an advantage here that a content farm structurally cannot copy, which is a real business producing real events every week. The policies that landed this week reward exactly that and penalise its absence.

What none of this settles

Detection is imperfect and every one of these systems will misfire. A careful writer with a plain style will occasionally be flagged, and a well disguised generated post will occasionally sail through. LinkedIn's report button hands part of the judgement to readers, which introduces its own noise, since "seems like AI slop" is also what a lot of people click when they simply disagree.

There is also the unresolved contradiction Social Media Today pointed at. LinkedIn sells AI features and demotes AI output. Snapchat ships AI editing tools and refuses to recommend AI videos. Both positions are coherent once you accept the assisted versus authored distinction, but they are going to be communicated badly, and merchants will reasonably be confused about which button in which product is safe to press.

The direction, though, is not ambiguous. Four platforms, two announced this week and two already enforced, all landed on the same rule with no coordination. For anyone selling something, the operational takeaway is smaller than the headlines suggest and more demanding than it sounds: keep using the tools, and keep being the reason the post exists.