Why Substantive Writing Should Remain Human-Authored, According to Erich Grunewald

The Fluency-Substance Gap

According to Erich Grunewald’s personal blog, large language models have mastered the surface features of compelling prose—grammar, vocabulary, coherence, stylistic variation—to a degree that fools most readers on a first pass. Yet this fluency masks a fundamental limitation: the ability to sound authoritative is not the same as having something true or original to say.

Grunewald argues this distinction becomes critical when the stakes are substantive. A marketing email or templated customer-service response can tolerate LLM authorship because the bar for those tasks is functional clarity, not originality or accountability. But journalism, opinion, technical documentation, and research occupy a different category. They require the author to stake a claim, reason through evidence, and remain answerable if the claim is wrong. An LLM cannot do any of these things.

Why AI Struggles With Reasoning and Originality

According to Grunewald, the gap between fluency and substance reflects how language models actually work. They are pattern-matching engines trained on human text; they do not reason from first principles or generate genuinely novel insights. When confronted with a prompt asking for original analysis, they recombine familiar arguments, often in ways that feel coherent but lack depth or critical rigor.

Grunewald emphasizes that this is not a temporary limitation destined to vanish with scale or fine-tuning. The problem runs deeper: an LLM has no skin in the game. It cannot be wrong in any sense that matters—it has no reputation, no career, no stake in whether readers act on its advice. Human writers, by contrast, earn credibility through years of getting things right and paying a cost when they get them wrong.

The Accountability Problem

Substantive writing typically implies someone is willing to sign their name to a claim. Grunewald notes that delegating this work to an AI tool creates a moral hazard: the writer remains responsible for the content but has outsourced the reasoning to a system incapable of reasoning. If the content is wrong, whose fault is it? The AI has no accountability; the human author does. This asymmetry makes AI-generated substantive work problematic for anyone serious about epistemic integrity.

This is especially acute in domains where readers rely on the author’s judgment—journalism, scientific communication, legal analysis, medical advice. Outsourcing these to LLMs is not just a shortcut; it is an abdication of the responsibility that substantive writing carries.

Why This Matters

Teams and individual writers deciding whether to adopt AI for content creation should ask a concrete question: Is this content a competitive differentiator, or is it a commodity? If your value proposition depends on original insights, rigorous reasoning, or hard-won expertise—hire a human writer and keep the work in-house. If the content is templated, routine, and interchangeable (FAQs, boilerplate product descriptions, standardized reports), AI tools are a reasonable efficiency play.

More broadly, organizations that outsource substantive writing to LLMs are betting that readers care only about surface fluency. For some audiences and contexts, that bet may pay off. But for any domain where trust, originality, or accountability matter—publishing, research, policy, professional services—the bet is likely to fail. Readers and regulators eventually notice the difference between plausible prose and defensible claims.