TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
In a post on MartinFowler.com, software writer Martin Fowler describes readers’ unease with what he calls “LLM voice” and advises authors not to use an LLM to draft prose they want people to read. He argues that writing also helps people develop and clarify their thinking, while recognizing uses such as translation, accessibility and routine communications.
Software writer Martin Fowler says readers can experience an “uncanny valley” when prose feels machine-generated, and advises authors to avoid using large language models to write drafts intended for human readers. In a post on MartinFowler.com, he argues that editing an AI-generated draft may not remove its characteristic voice and that writing itself helps people organize their thoughts.
Fowler describes a subjective reaction to some writing: he may suspect a post was generated by an LLM without being able to identify a reliable tell, and says his interest in reading can fade. He rejects punctuation such as em dashes as a dependable sign, and is wary of tools that claim to distinguish AI-written prose from human writing. His concern is not that such detectors necessarily give a particular result, but that checking with one does not address a reader’s reaction.
His practical recommendation is for authors to write the prose themselves, rather than ask an LLM for a draft and then revise it. Fowler says even accepting AI rephrasing suggestions can be risky, while using a model to help develop an outline may be acceptable if the author writes the actual prose. He describes one exception: a draft he knew was fully AI-generated whose final version did not strike him as having LLM voice, but says it required work from a professional editor at a major periodical.
Fowler also says his own policy is to disclose AI assistance in acknowledgments so readers can understand how it was used. He says he uses AI as a search tool for writing, then checks information against its original source; he does not use it to generate prose or develop structure. These are his stated practices and views, not findings from a study of AI writing or reader response.
Why Human Drafting Matters to Fowler
The essay connects the style of AI-assisted writing to a broader concern about how people think and communicate. Fowler says writing is a tool for working through uncertainty: composing an essay can reveal where a writer’s understanding is incomplete and help bring ideas into focus. If people hand over prose generation, he argues, they may also hand over some of that thinking and comprehension.
For readers, the immediate issue is trust and connection. A piece that feels generic or machine-produced may lose its audience even if it is accurate and has been reviewed. For writers and publishers, Fowler’s argument raises practical questions about disclosure, editing and when automation serves the purpose of a document. It is an argument based on his experience and preference; the post does not quantify how commonly readers detect AI-generated prose or measure its effects.
Products Worth Considering
Apple Watch SE 3 GPS 44mm Smartwatch
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
Amazon Kindle Paperwhite Signature Edition (12th Generation, 2024), 32 GB, Metallic Black
As an affiliate, we earn on qualifying purchases.
Where Fowler Draws the Line
Fowler does not argue that AI should never be used in writing. He identifies cases where the benefits may outweigh concerns about style, including translation for colleagues who are less comfortable writing in English and support for Jim Highsmith, a contributor to the site who uses an LLM to work around the effects of Parkinson’s disease. Fowler says he would accept an AI-influenced voice to keep receiving Highsmith’s ideas.
He also allows that AI-generated prose can make sense for a form letter or a report unlikely to be read by a person. Michael Nygard, whose comment Fowler includes, frames such material as communication with a process rather than an attempt to convey one person’s thoughts to another. Fowler’s distinction is chiefly about the purpose and audience of writing: whether it is meant to connect with people or complete a routine task.
“You’re communicating with a process rather than a human.”
— Michael Nygard, in a comment quoted by Fowler
Limits of Detecting AI Prose
The post offers personal observations and advice, not a systematic test of readers’ ability to identify AI-generated text. Fowler says he is unsure of his own ability to recognize LLM voice and does not trust detection tools to settle the question. The article provides no data on how often people share his reaction, which writing features cause it, or how much editing is needed to change that impression.
It also remains a matter of judgment when AI assistance crosses from support into replacing a writer’s voice. Fowler treats outlining differently from drafting, but does not set a universal standard for every use. His examples show that accessibility and translation needs can change the balance, while leaving unresolved how authors, editors and readers should handle disclosure in different settings.
How Writers Apply the Advice
Fowler’s post does not announce a new product, policy change or research project, and it gives no scheduled next milestone. The immediate next step it proposes is for writers to decide whether they are communicating with people or producing routine material, then choose AI assistance accordingly. For human-facing work, Fowler recommends building the prose through personal writing and practice; where AI contributes, his site’s approach is to disclose how it was used.
Key Questions
What does Fowler mean by “LLM voice”?
He uses the phrase for a style of prose that can make a reader suspect an LLM was involved and feel distanced from the writing. Fowler describes that response as subjective and does not claim there is one definitive sign.
Does Fowler recommend avoiding all AI writing tools?
No. He allows for uses such as searching for information, outlining, translation, accessibility support and routine communications. He recommends writing the prose yourself when the aim is to communicate with human readers.
Why does Fowler object to AI-written drafts?
He says editing an AI-generated draft may not remove the voice readers find alienating. He also argues that composing prose helps writers clarify their own thinking and understanding.
Does Fowler say AI-written text can never be edited successfully?
No. He describes one fully AI-generated first draft whose final version did not register to him as LLM voice. He says it required editing by a professional editor at a major periodical.
What does the post establish about AI detection?
It does not establish how accurately readers or detection tools can identify AI-written prose. Fowler expresses skepticism about both his own recognition ability and the value of detection tools, but offers no accuracy data.
Source: martinfowler.com
