Three institutions in three industries drew the same line within days of each other. None of them coordinated, and all of them arrived late.
Read together they describe something more useful than any single policy: the filtering job is moving, and it is moving toward you.
Australia decided what counts as human
ARIA, which runs Australia's music charts, will exclude songs created mostly or entirely by AI. To chart, a track has to be substantially human made. The ABC reported that the rule takes effect this week.
What triggered it is the part worth sitting with. An AI assisted song became the most played song on Australian radio in July.
The sequence matters more than the policy. Listeners heard it, requested it, and made it number one. The institution then ruled on its eligibility afterwards.
So the chart is not protecting an audience from something it rejected. It is protecting a category definition from something the audience accepted without noticing or caring.
That distinction should reorganise how you think about provenance in your own market. Your customers are not running an authenticity check before they buy. Institutions are, and only for their own boundaries.
I covered the regulatory version of this in the AI label becoming law. What ARIA adds is a commercial rule rather than a legal one, written by a body whose entire product is a definition.
Spotify deleted 75 million tracks
The scale of the second signal makes the first look quaint. Spotify removed 75 million bulk uploads, duplicate songs and spam tracks. LinkedIn tightened its detection systems against machine generated spam.
Seventy five million is not a taste problem. It is a supply problem, and it tells you the marginal cost of producing a track has effectively reached zero.
Reddit and Stack Overflow both introduced rules restricting AI generated answers, and their stated reason is the one to underline. Engagement drops when slop spreads unchecked, as Wired reported.
That is not an ethics policy. That is a retention policy with an ethics vocabulary.
Platforms move when volume degrades the product they sell. They do not move because provenance offends them, and expecting them to enforce a standard on your behalf is a strategy with no owner.
Note also what Spotify did not do. It did not build a labelling system, or a provenance standard, or a disclosure requirement. It deleted supply.
Deletion is cheaper than classification and it will keep being the first tool platforms reach for. Which means the real risk to a legitimate publisher is not being judged harshly on quality, it is being caught in a volume sweep.
The practical read for anyone publishing at volume: the platforms are now removing supply faster than they are ranking it. Your distribution assumptions from last year are stale.
The filter moves inside the company
The third signal is analytical rather than institutional, and it lands in the same place.
Kevin Indig has been arguing that companies need what he calls slop antibodies, internal filtering for low grade machine output, on the basis that the problem is how people use the tools rather than the tools themselves.
Google's position is adjacent. Quality over authorship. Helpful, reliable and high quality is the standard, and what produced it is not the question.
Put the three together and you get one conclusion from an institution, a platform and an analyst. Nobody upstream is going to screen your output for you, so the screening has to happen before you publish.
Which is awkward, because most content teams removed exactly that gate in 2024 and 2025 to hit volume targets. The editor was the cost line that looked most optional the moment drafting became free.
That gate is now a competitive asset. Not for ethical reasons, for arithmetic ones: when supply is infinite and platforms are actively deleting it, the scarce input is the judgement that decides what not to publish.
What to do with three signals
Stop waiting for a labelling standard to arrive and settle the question. ARIA wrote a rule for charts, the EU wrote one for disclosure, Spotify wrote one for uploads, and none of them describe your category.
Instead, decide internally what your own line is, write it down, and apply it before publication rather than after a platform does it for you.
Two questions do most of the work. Would we be comfortable if a customer knew exactly how this was produced. And does this contain anything we know that a competitor could not have generated from a prompt.
The first is a reputation test. The second is the only durable one, and it connects to how human input became a ranking factor rather than a virtue.
Write the line down as an actual document rather than a shared understanding. Shared understandings do not survive the first quarter where somebody is behind on volume.
The gatekeepers are drawing lines around their own categories, in their own interests, at their own pace.
Nobody is drawing one around yours.