Harvard Business Review says generative AI “worksop” feedback loops are worsening company decisions

Generative AI was sold as a quality accelerant. Harvard Business Review says it can turn into the opposite, when the very outputs teams generate start corrupting the inputs they later trust.

This month, HBR published two articles describing a feedback loop where AI-generated low-quality output degrades the information companies rely on to make decisions. The issue is not that AI always produces errors. It is that, in organizations that pushed hardest to adopt generative AI, the system can quietly start feeding on its own weak material, so that downstream work gets worse over time.

Why does this matter so much for executives? Because modern decision-making is increasingly dependent on internal information pipelines. If those pipelines are fed by documents, summaries, analyses, and draft content that are not reliable, leaders do not just get a few bad emails. They get a progressively noisier basis for prioritization, strategy, and execution. In a feedback loop described by HBR, low-quality outputs do not stay isolated. They spread into the information that teams use next, which then becomes the “source data” for subsequent AI and human work.

This is the uncomfortable twist: the technology was supposed to prevent bottlenecks and reduce friction in knowledge work, not intensify it. Generative AI can produce text quickly and at scale, so organizations often use it to accelerate drafting, summarizing, and ideation. But speed is not the same as truth. When AI is used repeatedly across processes, a small quality decline can compound. That is what HBR’s description of the feedback loop emphasizes: AI-generated low-quality output degrades information companies rely on for decisions, creating a self-reinforcing cycle.

There is also a governance angle that boards and audit committees will recognize. When output quality degrades silently, accountability gets harder. A decision made on top of questionable internal information can look “reasonable” because it arrives in a polished format. The problem is not only factual accuracy. It is the reliability of the information layer the organization treats as authoritative. If the underlying inputs become less trustworthy, then every downstream process, from leadership dashboards to project planning to customer messaging, is working with a weaker foundation.

For context, the adoption pattern matters. HBR’s framing points at companies that pushed hardest to adopt generative AI, which suggests a particular incentive structure: the competitive pressure to modernize quickly, the desire to capture productivity gains early, and the internal momentum that comes when early pilots turn into broad deployment. When adoption outruns the discipline needed to maintain information quality, feedback loops become more likely. The more an organization leans on AI outputs as raw material, the more exposure it has to a scenario where low-quality content becomes part of the system’s next iteration.

Regulatory background also makes this harder to ignore, even if HBR’s articles focus on the feedback loop itself. Across markets, regulators have increasingly emphasized the need for trustworthy AI and risk management, including requirements that push organizations toward clearer documentation, oversight, and controls. Even without tying the feedback loop to a specific regulation in the source, the direction of travel is the same: enterprises cannot treat generative AI as a plug-and-play text engine. They need governance structures that address how outputs affect the integrity of the decision inputs.

The second-order implication is that the “cost” of generative AI may not show up where leaders expect. If leaders measure success primarily in adoption rate, content volume, or time saved, a quality decline can hide in plain sight. HBR’s warning implies a different metric set: not just how much AI content is produced, but how well it preserves the informational quality leaders rely on for decisions.

Strategically, this is a wake-up call for every executive who is using generative AI to drive faster business processes. If HBR is right about the feedback loop, then the winning move is not stopping AI. It is treating information integrity as a first-class product requirement. Otherwise, companies could find that the system meant to speed up work is actually rotting the very decision-making machinery that keeps the business on course.