Almost Never Use AI to Write Anything Substantive
A Plea: Why You Should Almost Never Use AI for Substantive Writing
By Erich Grunewald | August 06, 2026
I believe there is a strong case to be made against using Artificial Intelligence to perform the actual act of writing—the physical and mental process of putting words on a page—whenever the goal is to communicate a substantive idea. This applies to:
- Research reports and memos
- Thoughtful emails
- Blog posts and novels
- Any text intended to convey an analysis or a complex argument
This stance holds true even if you provide the AI with exhaustive bullet points, dictated notes, or significant context, and even if you spend time editing the resulting output.
The Core Argument
My position is based on three primary pillars:
- Cognitive Integration: The act of writing is fundamentally inseparable from the act of thinking.
- Subtle Inaccuracy: AI-generated text tends to be vague or incorrect in ways that are difficult for a human to detect.
- Interpersonal Ethics: Presenting AI-written work as your own is
efficientmisleading and rude.
🛠️ Clarifications and "Throat Clearing"
Before diving deeper, it is important to define where AI does fit into the workflow. I am not advocating for a total ban on LLMs, but rather a specific boundary.
AI Usage Guidelines
| Category | Permitted Use (Green Light) | Discouraged Use (Red Light) |
|---|---|---|
| Research | Data analysis, searching, transcription | Drafting the final analysis |
| Ideation | Brainstorming, feedback on drafts | Generating the core argument |
| Refinement | Line editing, tightening prose | Writing the first substantive draft |
| Execution | Formatting, organizing notes | Synthesizing thoughts into a narrative |
Important Caveat: Using AI for copy-editing is acceptable only if a human deliberately accepts or rejects every single change.
The Trade-off
It is undeniable that AI is faster and requires significantly less effort. However, for the practice of writing to be "bad" overall despite these perks, the disadvantages must be severe. I believe they are.
Note on Future Tech: This critique applies to current and near-future models. If a model eventually arrives that is capable of truly substantive writing, it will likely be because it is performing the entire research and thinking process end-to-end, rendering the "writing" step a mere formality.
🧠 The Writing Process Is the Thinking Process
The goal of research is to develop accurate beliefs about critical questions. In my view, the most effective way to achieve this is through the act of writing.
As Paul Graham has noted:
"Writing about something, even something you know well, usually shows you that you didn’t know it as well as you thought... Half the ideas that end up in an essay will be ones you thought of while you were writing it."
Clara Collier echoed this sentiment on Patrick McKenzie’s podcast, explaining that in substantive writing, there is no moment where she isn't actively thinking and evolving her perspective. She noted that struggling with a transition often reveals a fundamental flaw in the logic of the outline itself.
The Cognitive Loop
If you feed an outline into an LLM, the machine will not stop to question if the outline is logically unsound. It simply fills in the blanks.
When we force ourselves to structure an argument into paragraphs, we expose the "holes." We discover contradictions and realize where our research was insufficient. An ethical writer acknowledges counterarguments; an AI simply smooths over the gaps.
Mathematically, we can view the relationship as:
To produce a quality piece, someone must hold the entire architecture—the evidence, the arguments, and the synthesis—in their mind. AI allows the writer to bypass this essential mental labor.
⚠️ The Danger of "Invisible" Errors
When a human writes, they make thousands of micro-decisions regarding phrasing to ensure precision. AI, conversely, often produces text that looks correct but is subtly vague or wrong.
For instance, if I prompt a model like Claude Fable to write a paragraph on AI chip smuggling, it might produce the following:
Since October 2022, the United States has restricted exports of advanced AI chips to China, aiming to slow Chinese progress at the frontier of AI.
But export controls are only as strong as their enforcement, and AI chips are compact, extraordinarily valuable, and sold through sprawling global supply chains — properties that make them attractive and feasible to smuggle.
Diversion networks route controlled GPUs through intermediaries in places like Singapore, Malaysia, and the UAE, using shell companies and falsified paperwork to move them into China.
Estimates of the scale vary widely, from tens of thousands to perhaps more than a hundred thousand high-end chips per year, and even the lower end represents enough compute to matter for training and deploying capable AI systems.
Meanwhile, the Bureau of Industry and Security, tasked with policing a multi-hundred-billion-dollar trade, operates with a budget and headcount better suited to a far smaller problem.
While this looks professional, the lack of specific, human-verified nuance can lead to a superficial understanding of the topic.
🖼️ Visual References
Final Checklist for Writers
- Have I used AI to think for me, or to help me organize my thoughts?
- Have I personally verified every claim in the text?
- Am I being honest with my reader about the origin of this prose?
- Did I embrace the struggle of the first draft to ensure my logic is sound?
