What Changed in Social Platform Policy This Year, and What Did Not

Every few months a wave of posts announces that a platform has “changed the algorithm”, usually accompanied by advice that happens to align with whatever the author sells. Almost none of it is traceable to anything a platform published.

This is a periodic look at the difference: what actually changed in documented policy, and what is inference presented as fact.

The distinction that matters

Platforms publish two very different kinds of information about themselves, and they get conflated constantly.

Policy — terms of service, community guidelines, monetisation rules. These are published, versioned, and enforceable. When they change, the change is documented and you can read it.

Ranking behaviour — how a recommendation system weighs signals. This is described in general terms and never specified. Meta’s Transparency Center, TikTok’s recommendation documentation and YouTube’s How YouTube Works all describe inputs without weightings.

So a claim like “Instagram now prioritises carousels” is not a policy statement. It is either an observation from someone’s own data — which is a sample of one account in one niche — or it is invented. Neither is a change you can verify, and both get reported the same way.

What is actually documented

The changes worth tracking are the ones in the first category, because they are checkable and because they have consequences you can plan around.

Enforcement of inauthentic engagement remains a stated policy everywhere. Every major platform prohibits buying engagement in its terms. This has not changed and there is no indication it will. What varies is enforcement intensity, which is not published.

Monetisation programmes continue to tighten on traffic quality. YouTube’s Partner Programme has long treated inauthentic traffic as a review concern, and this is the area where the gap between “prohibited” and “actually enforced” is narrowest — because there is revenue attached and an advertiser on the other side who cares. For anyone earning from a channel, this is materially different exposure than on platforms where the downside is a worse engagement ratio. We cover this in YouTube views explained.

Transparency reporting keeps expanding. More platforms publish more data about content removals and account actions. This is the best available source on how enforcement actually behaves, and almost nobody in this industry reads it.

What did not change, despite the claims

Three things get announced as new every year and are not.

“Engagement now matters more than followers.” This has been true for as long as recommendation feeds have existed. It gets rediscovered annually because it is a useful thing to say when selling engagement services.

“The algorithm is punishing X.” Punishment implies a rule being applied. What is usually happening is that a format performs worse in a specific niche, which is a distribution outcome rather than a policy. The distinction matters because one is appealable and the other is not.

“Buying followers is now safe / newly dangerous.” Neither. The terms have prohibited it throughout, and enforcement thresholds have never been published. Anyone claiming to know the current threshold is guessing with a number attached. Does buying members get you restricted? works through what is and is not knowable here.

Where the confusion comes from

It is worth being fair about why this happens, because it is not all bad faith.

Creators genuinely do experience sudden changes in their numbers. Reach drops by half from one week to the next with no change in what they posted. Something real happened, the platform explained nothing, and in the absence of an explanation people construct one.

Most of those experiences have prosaic causes that are invisible from inside a single account:

Sampling variance. Distribution is decided per post from a small initial test audience. Small samples are noisy, and a run of three weak posts can be nothing more than three unlucky draws. On TikTok in particular the variance is extreme enough that judging any single post is close to meaningless — TikTok growth basics covers why batches are the only readable unit.

Audience drift. The audience a system associates with an account changes gradually as the account’s content changes. Nobody notices the drift; everyone notices the day the posts stop landing.

Seasonality and competition. The same post competes against a different set of posts in December than in June.

Genuine but unannounced changes. Platforms do adjust their systems continuously and are under no obligation to say so. Some of the reported changes are real. The problem is that the reports cannot distinguish these from the first three causes, and neither can anyone outside the company.

So the honest position is not “nothing changes” — things change constantly. It is that a change observed from one account cannot be attributed, and confident attribution is the part that is invented.

What is worth reading instead

Four sources that are checkable, in descending order of usefulness:

1. The platform’s own policy pages, with their revision history where it is published. Dry, and the only authoritative account of what is prohibited.
2. Transparency reports. Enforcement volumes, categories and appeal outcomes. This is the closest thing to data about how enforcement actually behaves, and it is almost never cited in growth writing.
3. Developer and creator documentation. Rate limits, API changes and feature rollouts are documented because developers need them to be, and they often reveal product direction earlier than announcements do.
4. Your own analytics, read in batches. The only source that is actually about your account. Ten posts in a format tell you something; one tells you nothing.

How to read a claim about platform behaviour

A short test that filters most of it:

• Is there a link to the platform’s own documentation? If not, it is inference, and it should be labelled as such.
• Is the sample stated? “We saw this across 12 accounts in one niche” is useful. “We noticed” is not.
• Does the conclusion happen to recommend the author’s product? Not disqualifying, but it should raise the evidentiary bar.
• Is a specific number given for something unpublished? Exact weightings and enforcement thresholds are not public. A precise figure for either is a fabrication, however confidently delivered.

The one change that would matter

Worth naming, because it is the thing to watch for rather than the thing that gets reported.

The market in purchased engagement depends on a specific asymmetry: platforms prohibit it in policy but enforce it inconsistently and invisibly, so buyers can treat the risk as theoretical. Everything about how this industry operates — the vague guarantees, the untraceable supply chains, the refill windows — is shaped by that gap between the stated rule and the observed consequence.

What would actually change the market is not an algorithm adjustment. It is enforcement becoming visible and predictable: published thresholds, stated consequences, or a systematic removal programme that people could observe operating. There is no particular sign of that happening, and there are reasonable arguments why a platform would not want it — publishing a threshold publishes the specification for staying under it.

Until then, the practical position stays where it has been. The rules say one thing, enforcement is unpredictable, and the reliably predictable cost of buying engagement is not a penalty but a broken ratio. That is covered in the Instagram followers guide and, for the platform where it works differently, in does buying members get you restricted?.

What this means practically

The stable ground has not moved. Recommendation systems weigh signals that are expensive for a viewer to produce — completion, rewatches, saves, shares — because those are the signals that are hard to fake. Cheap signals stay cheap and stay weakly weighted. That structural fact is why the market in purchased engagement keeps selling the wrong things, and it does not depend on any particular year’s changes.

If you want the durable version rather than the news version, which signals matter and the social media growth guide cover the mechanics that have held constant throughout.

We will update this page when something genuinely documented changes. If you spot a policy change we have missed, the contact page is the fastest route.