AI-Generated Content Is Degrading Corporate Knowledge

The hidden cost of AI automation

Generative AI promises productivity gains, but research reveals a troubling side effect: the systematic degradation of organizational knowledge. When employees use AI tools without proper quality controls, errors compound across business processes, trust erodes, and the very systems companies rely on begin to fail.

Matthias Holweg and Thomas H. Davenport, writing in Harvard Business Review, describe this phenomenon as knowledge decay—the organizational version of "workslop," where individuals produce polished-looking but low-quality AI-generated work. When this happens at scale across interconnected processes, the consequences multiply.

How processes break down

The pattern appears across industries. In hiring, AI now touches every step: writing job descriptions, screening resumes, conducting interviews, and evaluating responses. Candidates use AI to generate keyword-optimized applications and real-time interview answers. Recruiters use AI to rank applicants. The result is a system assessing AI proficiency rather than job fit. Research confirms that while AI helps recruiters post more descriptions, they're more generic, less informative, and less likely to yield good matches.

Academic publishing faces similar pressures. Since ChatGPT's late 2022 release, the journal Organization Science has seen submission volume rise 42 percent while writing quality has declined. Some papers feature fake authors or falsely credit real researchers. Reviewers increasingly use AI to evaluate submissions, creating an AI-to-AI feedback loop divorced from human judgment.

In healthcare, roughly 40 percent of U.S. primary care physicians use AI clinical decision tools. When these outputs feed into insurance companies that also use AI for preapprovals, inaccuracies at any step cascade through the system, affecting patient care and potentially de-skilling clinicians.

Three core challenges

Holweg and Davenport identify three problems leaders must address:

Verification: Distinguishing accurate information from AI hallucinations requires labor-intensive human review that often negates productivity gains. The more capable AI becomes, the harder verification gets.

Validation: Recipients must now determine whether content represents genuine human expertise or AI generation. Clients won't pay consulting fees for AI-generated reports, lawyers face sanctions for citing AI-hallucinated case law, and students demand tuition refunds when professors rely excessively on AI.

Entropy: Information degrades as it passes through successive AI iterations, moving further from ground truth. The transformer algorithms underlying large language models are probabilistic next-word predictors with no conception of fact. Training AI on AI-generated content—"generative inbreeding"—accelerates this decay. Studies suggest half of internet content is already AI-generated.

Why it matters

Knowledge decay isn't a theoretical concern—it's actively undermining trust in hiring, research, healthcare, and other critical processes. As AI use becomes impossible to police (more than half of workers in one survey concealed their AI usage), organizations must develop explicit strategies to preserve knowledge quality rather than hoping individual employees will maintain standards.

Four protective strategies

The researchers recommend tracking the provenance of unstructured data to distinguish ground truth from generated content; restricting AI use to areas where it genuinely adds value; clearly defining what value AI provides in each application; and assessing AI's impact on entire processes rather than individual tasks.

For interorganizational processes like healthcare revenue cycles, all parties should agree on AI usage protocols to prevent systemic degradation.

These findings were first reported by Matthias Holweg and Thomas H. Davenport in Harvard Business Review.