Week 4: The economics of volume

Why slop gets made — ad-tech incentives, content-farm economics, and the specific mechanism by which platforms reward volume over craft

15 March 2027

Nobody sets out to make a worse video. Slop is what a system produces when the return on a video is measured in views-per-dollar-of-production-time rather than in quality-per-video, and enough of the internet’s attention is sold by the thousand views that this measurement wins whenever it’s available. This screening is about the specific machinery that makes it available, not a general complaint that “platforms reward engagement.”

The pre-AI precedent#

Content farms are not a generative-AI invention. The mid-2000s to early-2010s saw a whole industry — Demand Media and its imitators are the textbook case — built on paying freelance writers roughly fifteen to thirty dollars per few-hundred-word article, keyed to search-engine-optimised titles (“how to remove a stripped screw”), at industrial scale, because a single ad-supported pageview paid back that per-article cost many times over if enough of them ranked. Nobody involved thought “how to remove a stripped screw” was a piece of writing worth being proud of. The business case never required it to be — it required only that it be cheap enough per unit and good enough per unit to clear the search engine’s bar, at a volume large enough that the aggregate ad revenue dwarfed the aggregate cost.

The mechanism, specifically#

A platform’s ranking system generally optimises for a proxy of engagement — watch time, completion rate, shares — because that proxy correlates, imperfectly but usefully to the platform, with ad inventory sold. It does not know or care whether a video’s provenance is a director’s afternoon or a render farm’s idle GPU cycles. This is the mechanism, precisely: the ranking system has no craft term in its objective function, only an attention term, so any production method that produces more attention-worthy attempts per dollar will out-compete a slower, more careful one at the margin, video by video, without anyone at the platform making a decision that favours slop — the favouring happens automatically, as a consequence of what’s cheap enough to attempt many times.

Case study: Everline Gadget Reviews#

A specimen-shelf regular: a channel posting AI-generated “unboxing” and demo videos for gadgets that, as far as anyone has checked, were never physically unboxed by anyone. Each video costs, by a rough specimen-shelf estimate, somewhere under two dollars in generation compute and script-model calls to produce a finished ninety-second video, against a typical short-form ad rate that needs on the order of thirty to fifty thousand views to recoup that cost before the channel operator has been paid anything at all for their time. At the channel’s reported posting rate — six videos a day — that’s a daily cost low enough that a single video clearing half a million views once a month covers a month’s output outright.

What this explains about the genre#

Slop’s characteristic volume — the same channel posting dozens of near-identical videos a day — isn’t an aesthetic choice or a lack of discipline. It’s the rational strategy once the cost of an attempt falls far enough below the expected value of a hit, the same strategy a content farm ran on freelance articles a decade earlier, at a cost structure that makes today’s version harder to out-compete with craft alone.