Does a feed that shows AI posts to fewer people keep AI and human creators in balance after the first year?

This explores whether a recommendation algorithm that shows AI-made posts to fewer people can keep the balance between AI and human creators over the long run, beyond the first year. The short answer is that the corpus can't settle the long-run question. No study here tracks this kind of exposure adjustment for more than twelve months. What the corpus does have is the starting mechanism and several reasons to doubt it scales by itself.

The mechanism comes from a large Chinese short-video platform. Across about 179,000 matched pairs of posts, the algorithm gave AI-generated videos less exposure than comparable human-made ones Can algorithmic distribution prevent AI content from overwhelming creator diversity?. On the same platform, AI creators still matched human creators' total engagement. They did it by posting much more, even though viewers watched fewer of their videos all the way through Can AI creators match human creators through posting volume alone?. The algorithm's penalty applies to each post, but creators can make up for it by posting more. If making AI content keeps getting cheaper, a fixed per-post discount gets outpaced, unless the platform keeps making it steeper. The excerpt available here also leaves out the exposure results and robustness checks. Treat the moderating effect as a hypothesis, not a finding.

The growth curves show how fast the pressure can build. AI-attributed posts on Medium and Quora rose from about 2% to about 38% in under two years, while Reddit barely moved, from 1.3% to 2.5% Is AI-generated content rising faster on some platforms?. By mid-2025, roughly a third of new websites were AI-generated or AI-assisted, and the range of ideas across them had narrowed How much of the internet is AI-generated now?. The gap between Reddit and Medium suggests that platform design and community norms shape supply as much as ranking does. That puts the 'does it hold' question back on each platform rather than on algorithms in general.

A quieter problem is that downranking AI content only works if the system can tell what is AI-made. People detect AI content at about chance levels Can people reliably spot content made by AI?. Writers who use AI assistance edit its paragraphs only 23% of the time, and lightly when they do Do writers actually edit AI-generated text before publishing?. As mixed human-AI content becomes normal, the line the algorithm relies on gets blurrier. The 'epistemic hyperinflation' argument goes further: the tools that judge AI content are increasingly AI themselves, so evaluation may never catch up with production Can AI generate knowledge faster than humans can evaluate it?.

The less obvious point is that even if exposure balancing works on raw numbers, it may not protect what matters. AI posts can win engagement while building no lasting reputation for any real person. Over time, that wears down social media's role of surfacing credible human voices, and monetization carries on regardless Does AI content displace human influencers on social media?. So the useful long-run question is less whether the algorithm can keep AI's share down and more whether the platform still does the job it was built for. The corpus has much stronger evidence for that worry than for the algorithmic fix.