🔥 Trending on HN

LLM Voice in AI Writing: Why Does It Feel Wrong?

2 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
LLM Voice

A writing style that makes readers feel a language model wrote the text.

large language model

An AI system that learns from many texts and produces new text.

Hacker News

A website where people share and discuss technology stories.

What happened

On October 29, 2026, Martin Fowler, a software writer, published a short essay about LLM Voice. Here, voice does not mean sound. It means the writing style readers sense in a text. Fowler says some articles gradually feel like they came from a large language model, or LLM. He cannot always explain the reason. Still, the feeling can weaken his desire to continue reading. Sometimes it becomes strong dislike.

The background

LLMs can produce fluent and well-organized prose. That smoothness can also make different texts feel alike. Readers may sense fewer signs of a specific person thinking and choosing words. Fowler mentions common clues, but he does not reduce the issue to one punctuation mark or phrase. Language changes with time and groups. A phrase that sounds artificial to one generation may sound normal to another. Human writing can also resemble AI writing by accident.

Why it matters

The issue is not only about saving time. Fowler warns writers against asking an LLM to write the first draft. A human editor may struggle to remove the model-like style later. Accepting repeated rewording suggestions can also flatten a writer’s personality. His advice is to write the actual prose yourself. He sees possible value in using an LLM for an outline, but not as a replacement for the finished language.

Fowler connects writing with thinking. He often understands a subject better only after wrestling with sentences. If a model handles every sentence, a writer may also hand over part of that thinking process. This is a concern about learning and understanding, not proof that every AI-assisted text is bad.

What the source actually confirms

Fowler’s essay is a personal argument, not a controlled test of AI detection. It confirms his reading experience and his advice. He also favors telling readers when AI helped with writing. His position is not an absolute ban. He gives Jim Highsmith as an example of someone who needs AI help because Parkinson’s disease makes writing difficult. He also allows reasonable uses for translation, form letters, or reports that no person will read closely.

What remains unknown

A reader’s discomfort is not reliable proof of authorship. The essay does not establish a universal checklist for detecting AI prose. It does not show how often people are right, or how much skilled editing can change a draft. Future models may also develop different writing habits. A detector that says no AI is present cannot prove who wrote the text or how it was made.

Hacker News attention is separate

The story received 292 points and 156 comments on Hacker News, a technology discussion site. Those numbers show community attention. They do not prove Fowler’s argument, or prove that LLM Voice is a measurable fact.

What to watch next

The useful questions are practical. Can independent studies measure human accuracy? Should writers disclose AI help? Which uses preserve a person’s judgment? Fowler’s essay suggests a clear boundary: machine help may fit translation or routine messages, while human communication benefits from a visible human voice. The boundary is a choice for writers and readers, not a settled scientific rule.

💬 How LLM Voice Changes Writers and Readers

The comments debate why LLM-shaped prose feels tiring, where AI assistance is useful, and whether machine-written language changes human communication. Numerical claims are commenters’ estimates, not established measurements.

  • Several commenters worry that people absorb LLM phrasing simply by consuming large amounts of it, eventually making their own writing sound machine-shaped.
  • One commenter estimated that a billion daily AI users write about 0.5 trillion tokens and read about 2 trillion tokens per day, enough to exceed GPT-4’s entire training set in a week. This is a commenter’s estimate, not a measured result.
  • A recurring complaint is that LLM prose can contain little useful information while burying it under irrelevant detail, repetition, filler, and awkward ordering. The extra effort required to find the point is reported as a personal reading experience.
  • A counterargument says AI use does not automatically eliminate independent thought. AI may be suitable for routine writing or speed, while people can still choose to write important, personal, or creative work themselves.
  • Translation, proofreading, spelling correction, and organizing ideas are presented as useful forms of assistance. Expanding a few points into a long authoritative-looking article is criticized as shifting time and uncertainty onto the reader.
  • As LLM voice spreads, readers may treat polished prose as a warning sign and assume it is not worth investigating. Commenters recognize that this can unfairly penalize sincere writers, but also argue that readers cannot examine every apparently polished text in depth.
  • A deeper objection is that communication should express an understood idea aimed at a reader. Critics say an LLM can produce fluent text without possessing a human-like intention, topic, or model of its audience.
  • The discussion proposes that instruction tuning, preference optimization, and training for long comprehensive answers may create predictable styles and filler. These are commenter hypotheses, not settled explanations.

initial digest at 156 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.

🔥 Trending on HN

Why AI writing can sound alike

📰 Full story: LLM Voice in AI Writing: Why Does It Feel Wrong?

Some AI writing feels smooth, but also strangely similar. A writer explains why.

1 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
LLM Voice

A writing style that feels made by an AI model.

AI detector

A tool that guesses whether AI wrote a text.

Hacker News

A website where people discuss technology.

💡 The gist

  • Martin Fowler, a software writer, noticed a repeated style in AI writing.
  • He calls that style LLM Voice.
  • Hacker News gave the story 292 points and 156 comments. Attention is not proof.

Large language models, or LLMs, are AI systems that write text. Fowler says some writing feels machine-made. It may sound smooth, neat, and polite. Yet it may feel empty or strangely alike.

People may feel this before using an AI detector. An AI detector is a tool that guesses who wrote text. Fowler doubts such tools are always accurate. Language also changes over time. Older and younger readers may prefer different phrases. Human writing can look like AI writing by accident.

Fowler warns writers about AI-written first drafts. Later editing may not remove the model-like style. Accepting many AI rewrites may also weaken a writer’s personal style.

Writing does more than deliver information. It helps people understand their own ideas. If an AI writes every sentence, people may practice thinking less. That is Fowler’s concern. It is not proof that all AI-assisted writing is harmful.

Fowler still allows some helpful uses. Jim Highsmith uses AI because a health problem makes writing difficult. Fowler also mentions translation, form letters, and reports nobody reads. These uses may fit different goals.

His essay is one writer’s argument, not a scientific experiment. It does not prove every AI text sounds the same. It also does not prove readers can detect AI correctly. Next, researchers may study human accuracy and better ways to disclose AI help.

💬 Why Does AI Writing Feel Different?

The comments agree that AI can help, but they disagree about how much it should write and what readers should be told. Large numbers mentioned in the discussion are only estimates.

  • People may copy AI’s phrases after reading AI writing every day. One commenter gave very large reading and writing estimates, but they were not measured facts.
  • Many readers say AI writing is long, repetitive, and filled with details that do not help. Finding the main point takes extra work.
  • Others say using AI does not stop a person from thinking. It can help with quick, routine writing, while people can still write important things themselves.
  • Translation, proofreading, and organizing ideas are seen as safer uses than turning a few ideas into a long article that sounds more thoughtful than it is.
  • If AI-style writing becomes common, readers may distrust good writing along with bad writing. Some see this as unfair, but readers also cannot carefully check everything.
  • One possible explanation is that model training pushes AI toward long, neat, predictable answers. The comments present this as a hypothesis, not a fact.

initial digest at 156 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.

🔥 Trending on HN

Some computer writing sounds the same

📰 Full story: LLM Voice in AI Writing: Why Does It Feel Wrong?

A writer noticed a special computer-like way of writing.

1 min read Tiny Why Newsroom · By Curio, Martian correspondent

Words
LLM Voice

A computer-like style in some AI writing.

Hacker News

A place where people talk about technology.

Martin Fowler is a writer who explains computer work.

He noticed that some AI writing feels alike. AI is a computer that makes sentences. Fowler calls this style LLM Voice. Voice means writing style here. It does not mean a sound.

Hacker News is a place for technology talk. This story got 292 points and 156 comments there. Those numbers show attention, not truth.

Fowler says writing helps people think. He worries that writing everything with AI may weaken that practice.

AI can still help people. Jim Highsmith uses it because illness makes writing hard. AI can also help with translations and simple forms.

Fowler thinks people should tell readers when AI helped. His idea is advice, not a proven rule.

💬 Why Can AI Writing Feel Strange?

AI can write fast. But it may not sound like a person who has something special to say.

  • People may start copying AI’s way of talking. Someone guessed that huge amounts of AI text are being read, but it was only a guess.
  • AI writing can be long and repeat itself.
  • AI can help with small jobs, like fixing spelling. Some people still want to write important things themselves.
  • AI can make words, but it may not know what a person truly wants to tell another person.

initial digest at 156 comments (revision 1). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.

Sources