YouTube claims users spend 70% less time on “borderline” content, like conspiracies, since it tweaked recommendations in January but fails to offer more details
and in the great Silicon Valley tradition it lacks the underlying figures or anything comparative https://www.washingtonpost.com/ ... See also Mediagazer
Context & Ripple Effects
The claim lands on an engine whose outsized power was documented back in 2018, when reporting showed YouTube's [[a:926527|recommendations engine drives more than 70% of viewing time and often surfaces divisive or false videos]] users never searched for. Since then the platform has been walking the dial backward: first a UK experiment targeting a 50% reduction in borderline views from recommendations, now a global figure of 70% less watch time since January.
What makes the number hard to weigh is that YouTube has spent two years quietly rebuilding how it measures itself — sources described new internal metrics including total time spent and the murkier "quality watch time" — while publishing no underlying figures or comparisons behind today's claim.
First-order effects
- Creators of borderline content — conspiracies and adjacent material — face a direct reach collapse, since the recommendation system that once amplified them is now suppressing their share of viewing time by YouTube's own account.
- Advertisers and policymakers get a headline stat they cannot audit: with no baseline, methodology, or comparative data released, the 70% figure functions as a press statement rather than a verifiable metric.
Second-order effects
- Independent researchers, who already find YouTube hard to study because video resists bulk analysis, are pushed further toward reliance on platform self-reporting — the late-2019 study showing the algorithm steering users to mainstream videos being one of the few outside checks available.
- Rival platforms face a template: YouTube is demonstrating that a recommendation tweak plus a self-published percentage can serve as the public answer to moderation pressure, raising the bar for what Facebook and Twitter are expected to disclose.
Third-order effects
- If the pattern holds, platform governance settles into a regime where the operator defines the metric — "borderline," "quality watch time," eventually violative views, which YouTube later began quantifying per 10K views — and regulators and researchers must argue over definitions rather than data.
- That structure favors incumbents: whoever owns the recommendation system and its telemetry controls both the intervention and the evidence of its success, making external auditing capacity the scarce resource in platform accountability.
The trend: Recommendation platforms are shifting from optimizing raw watch time toward curated "quality" engagement, with the operator itself supplying — and defining — the numbers that judge the shift.