Snap announced on 31 July that wholly AI-generated videos will no longer be recommended on Spotlight, its short-form discovery feed. The company's stated reason is refreshingly free of corporate fog: as low-quality, repetitive AI content spreads across the internet, it wants Spotlight to remain a place for original perspectives and personal storytelling. For the 971 million people who use Snapchat monthly, the intended effect is simple — less machine-assembled filler arriving unbidden.
The mechanics deserve attention, because they are cleverer than a ban and weaker than one at the same time. Fully AI-generated clips can still be posted; they simply will not be shown to anyone who does not already follow the account. On a discovery feed, that is exile in all but name. The reach was the point.
But watch the two carve-outs. First: videos enhanced or edited with Snapchat's own AI tools remain eligible for recommendation, with transparency labels attached. AI can help make the thing; it cannot be the whole thing — a line that happens to sit very comfortably with keeping creators inside Snap's editing suite. Second, and larger: the word "wholly" is carrying an extraordinary load. A great deal of AI can go into a clip before it crosses that threshold, and the threshold itself is where enforcement lives.
On enforcement, Snap concedes what everyone in the field knows: no detection system is perfect. Reliably identifying fully generated video at scale is not currently possible. Doing it properly would need provenance metadata — a tamper-resistant record of how a clip was made — adopted simultaneously by camera makers, editing software and every major platform. The industry has understood this for years and shipped nothing universal, in part because the companies best placed to build the standard are the ones selling the generators. Snap, whatever its intentions, lacks the leverage to force the issue alone.
So what is the policy worth? More than cynicism suggests. Snap says unique Spotlight contributors are up more than 120% year on year — it has an actual population of humans worth protecting, and it has now said in public that human-made content is an asset to defend rather than a cost to tolerate. Contrast Meta, which launched a wholly AI-generated video app in February and labels synthetic posts rather than demoting them. Two platforms, opposite bets, both watching the same audience: the young users most likely to walk away from a feed that feels manufactured.
For viewers, the practical upshot: expect fewer conveyor-belt clips, not none, and treat the change as a volume knob rather than a filter. For creators: the safe harbour is Snap's own tools, labelled. And for everyone else, note the date — the EU's AI-transparency rules took effect on 2 August, two days after this announcement. Being early looks like principle. Principle with good timing is still principle. Just.
