Writing is probably the worst use case of AI (www.sitepoint.com)

An opinion piece argues that writing—particularly blog writing—is one of the weakest and most harmful use cases for AI. The author concedes LLMs are genuinely useful for tasks like meta descriptions, summaries and alt text, but warns that using AI to generate whole articles produces homogenized, low-value content. Because most LLMs are trained on the same internet corpus (and some models have special access to sources like Reddit), their outputs tend to converge: many sites publishing AI-derived paragraphs create near-duplicate coverage that adds no new insight. That “snake-eat-snake” feedback loop risks degrading the quality of the web — the very data future models will train on. The piece highlights practical and technical consequences: ad-driven blogs lose traffic to chatbots, shrinking budgets and incentivizing more AI-generated posts; content detectors remain unreliable with high false positives; text lacks the metadata that can help trace AI images, so detection is hard to scale; and major platforms like Google reversed an anti-AI-content stance (policy shifted from 2022 to 2023) in favor of rewarding quality regardless of origin. The takeaway for the AI/ML community is clear: smarter models don’t solve the socio-technical problem of content value and dataset contamination. Human editing and original reporting retain importance, and blindly outsourcing writing to LLMs risks eroding both web quality and future training data.