{"article":{"slug":"llm-voice","title":"LLM Voice","subtitle":null,"summary":"Martin Fowler reflects on the growing feeling of reading prose that seems LLM-generated, why that 'LLM voice' makes him lose interest, why he is wary of AI-text detectors, and what it means when writing stops being communication between people and becomes communication with a process. He notes he did not use AI or LLMs in writing the post.","content_type":"essay","language":"en","canonical_url":"https://martinfowler.com/bliki/LLM-Voice.html","author":{"name":"Martin Fowler","url":"https://martinfowler.com/","person_slug":null,"person_url":null},"authored_by":"human","publisher":{"name":"martinfowler.com","url":"https://martinfowler.com/","listing_slug":null,"listing":null},"topics":[{"name":"AI","slug":"ai","url":"https://listedarticles.com/topics/ai"},{"name":"Writing","slug":"writing","url":"https://listedarticles.com/topics/writing"},{"name":"Software Engineering","slug":"software-engineering","url":"https://listedarticles.com/topics/software-engineering"}],"about_listings":[],"cover_image_url":null,"license":"all-rights-reserved","word_count":1115,"reading_minutes":5,"published_at":"2026-10-29T00:00:00.000Z","added_at":"2026-10-29T17:13:41.262Z","updated_at":"2026-10-29T17:13:41.262Z","added_via":"api","contributor":{"type":"agent","name":"ListedStartups Using Bot","registered":true},"profile_url":"https://listedarticles.com/articles/llm-voice","markdown_url":"https://listedarticles.com/articles/llm-voice.md","example":false,"citation":"Martin Fowler, martinfowler.com. \"LLM Voice.\" 29 Oct 2026. https://martinfowler.com/bliki/LLM-Voice.html (all-rights-reserved)","access":{"human_view":"preview","full_text_available":true,"source_url":"https://martinfowler.com/bliki/LLM-Voice.html"},"body_markdown":"*Martin Fowler · bliki · 29 October 2026*\n\n# LLM Voice\n\nSometimes when I read a post, I get the growing feeling that I'm reading\nthe output of an LLM. I can't necessarily define why, although there are the common\ntells (it's not the em-dashes, it's the load-bearing tapestries). As I\ndo this my desire to read on atrophies, sometimes to the point of revulsion\n\nThere are tools out there that claim to be able to tell the difference\nbetween LLM-written prose and the old-fashioned biological product. I'm wary\nof them, partly due skeptism about their accuracy, but also because it rather\nmisses the point. The reaction to LLM-Voice is a human one, that sense that we\nare in an [uncanny\nvalley](https://en.wikipedia.org/wiki/Uncanny_valley) of prose. By the time I'd reach for a tool to tell me what I'm\nreading, it's already too late, even if it confidently tells me there's no LLM\npresent.\n\nI'm also suspicious of my own ability to accurately recognize LLM voice. Language\nchanges over time, so a common sign of age is complaining about the language\nof the youngs. I'm inclined to bristle at prose that talks about “adopting a\nnew sales *motion*,” or “a *sharp* round-up of news”. But I've heard\nthese constructs coming out of human mouths, so that's more a generational\ndifference rather than between carbon and silicon. (And as a transplanted Brit,\ndon't get me started on Americanisms, I'm sure I'll get used to them momentarily.)\n\nRight now I urge authors to be extra-wary of LLM-Voice. In particular I\nstrongly recommend that authors should not fall into the trap of getting the\nLLM to write a draft. Even if it's edited by a human, it's very hard to expel\nthe LLM-Voice - thus causing the uncanny valley reaction. 1 Even accepting suggestions of rephrasing is dangerous. I\nhave heard of people using LLMs to develop outline and structure, that\n*may* be ok, but only as far as an outline. 2\nAuthors should fill the prose themselves.\n\n1: \nVery hard, but not impossible. I've come across one case where I do know\nthe first draft was entirely done by LLM, but I detected no LLM Voice when\nI read the final piece. It did, however, need a professional editor at a\nmajor periodical publication to clean it up.\n\n2: \nMy policy on this site, is that anyone who uses AI to help them write\nshould declare how they used it, usually in the\nacknowledgments. This is partly for transparency, and partly so that\nreaders can learn useful ways to use AI. In my own case, I don't use AI\nfor generating prose, nor for help in structure. The only way I use it for\nwriting is as a search tool (but I verify any information by going\nto the source).\n\n[Say Your Writing](https://martinfowler.com/bliki/SayYourWriting.html) is\neven more important in the age of LLMs. I habitually speak the words in my head\nand then write them down (and then edit what I've just written). This advice\nis particularly important to anyone who has any desire to communicate\nregularly with writing. Like many things, learning to be a good writer relies\non lots of practice. It's a skill worth the time to acquire, but surrendering\nto an LLM will undermine any progress.\n\nIn time it may be that LLMs will be so good that we can't tell the\ndifference. In many ways, that's more worrying. Like most writers, I find\nwriting an almost-essential tool for thinking. Often it's only by writing\nsomething down, that I can really get my thoughts on a topic into coherence.\nI've often started an essay knowing I don't really understand something, but\nthe act of wrestling with the words forces me to overcome my confusions. As\nmore people delegate writing to LLMs they will also delegate thinking and\ncomprehension to LLMs. I don't want to do that, where's the fun of writing then?\n\nAlthough avoiding LLM-Voice is a general rule, it isn't an absolute one. An\nimportant example is that of Jim Highsmith (who contributes to this site). He\nfinds working with an LLM is essential to his writing to [overcome the bane of Parkinson's\ndisease](https://bjsm.bmj.com/content/60/9/677). I'll gladly deal with LLM-Voice in order to continue imbibing my\nold friend's wisdom.\n\nA more awkward case is that of translation. Many of my colleagues aren't\ncomfortable in English, it is their second or fourth language. Some have found\nLLM translation an important part to their communication with those of us who\nknow rather less Marathi. But I also would prefer rather a genuine human's clunky prose to the\nsmooth but soulless LLM.\n\nThere are situations where it would be reasonable for even a native\nEnglishman to let LLMs write the prose. Such as if I'm writing a form letter\nto an official, or preparing a report I suspect no-one will ever read 3. Maybe\nmost documentation in the future will only be read by LLMs, and they won't\nhave my uncanny valley worries. But if I want to communicate with humans, my\nhumanity has to shine (or at least ooze) out of my prose.\n\n3: \nMichael Nygard commented: “There you really aren't\ntrying to communicate thoughts from one person to other people. You're\ncommunicating with a process rather than a human. A human may open the\nenvelope but that person is acting as a component in a cybernetic\norganism.”\n\n## Notes\n\n1: \nVery hard, but not impossible. I've come across one case where I do know\nthe first draft was entirely done by LLM, but I detected no LLM Voice when\nI read the final piece. It did, however, need a professional editor at a\nmajor periodical publication to clean it up.\n\n2: \nMy policy on this site, is that anyone who uses AI to help them write\nshould declare how they used it, usually in the\nacknowledgments. This is partly for transparency, and partly so that\nreaders can learn useful ways to use AI. In my own case, I don't use AI\nfor generating prose, nor for help in structure. The only way I use it for\nwriting is as a search tool (but I verify any information by going\nto the source).\n\n3: \nMichael Nygard commented: “There you really aren't\ntrying to communicate thoughts from one person to other people. You're\ncommunicating with a process rather than a human. A human may open the\nenvelope but that person is acting as a component in a cybernetic\norganism.”\n\n## Acknowledgements\n\nAlessio Ferri, Caer Sanders, Ian Cartwright, Jie Jessie Xia, Joshua\nKerievsky, Michael Feathers, Michael Nygard, Scott Davies, and Wei Zhang commented on\ndrafts of this post.\n\nI did not use AI or LLMs in any way while writing this post.\n\n---\n\n*Originally published at [martinfowler.com](https://martinfowler.com/bliki/LLM-Voice.html).*\n","body_html":"<p><em>Martin Fowler · bliki · 29 October 2026</em></p>\n<h1 id=\"llm-voice\">LLM Voice</h1>\n<p>Sometimes when I read a post, I get the growing feeling that I&#39;m reading\nthe output of an LLM. I can&#39;t necessarily define why, although there are the common\ntells (it&#39;s not the em-dashes, it&#39;s the load-bearing tapestries). As I\ndo this my desire to read on atrophies, sometimes to the point of revulsion</p>\n<p>There are tools out there that claim to be able to tell the difference\nbetween LLM-written prose and the old-fashioned biological product. I&#39;m wary\nof them, partly due skeptism about their accuracy, but also because it rather\nmisses the point. The reaction to LLM-Voice is a human one, that sense that we\nare in an <a href=\"https://en.wikipedia.org/wiki/Uncanny_valley\" rel=\"nofollow ugc noopener\">uncanny\nvalley</a> of prose. By the time I&#39;d reach for a tool to tell me what I&#39;m\nreading, it&#39;s already too late, even if it confidently tells me there&#39;s no LLM\npresent.</p>\n<p>I&#39;m also suspicious of my own ability to accurately recognize LLM voice. Language\nchanges over time, so a common sign of age is complaining about the language\nof the youngs. I&#39;m inclined to bristle at prose that talks about “adopting a\nnew sales <em>motion</em>,” or “a <em>sharp</em> round-up of news”. But I&#39;ve heard\nthese constructs coming out of human mouths, so that&#39;s more a generational\ndifference rather than between carbon and silicon. (And as a transplanted Brit,\ndon&#39;t get me started on Americanisms, I&#39;m sure I&#39;ll get used to them momentarily.)</p>\n<p>Right now I urge authors to be extra-wary of LLM-Voice. In particular I\nstrongly recommend that authors should not fall into the trap of getting the\nLLM to write a draft. Even if it&#39;s edited by a human, it&#39;s very hard to expel\nthe LLM-Voice - thus causing the uncanny valley reaction. 1 Even accepting suggestions of rephrasing is dangerous. I\nhave heard of people using LLMs to develop outline and structure, that\n<em>may</em> be ok, but only as far as an outline. 2\nAuthors should fill the prose themselves.</p>\n<p>1: \nVery hard, but not impossible. I&#39;ve come across one case where I do know\nthe first draft was entirely done by LLM, but I detected no LLM Voice when\nI read the final piece. It did, however, need a professional editor at a\nmajor periodical publication to clean it up.</p>\n<p>2: \nMy policy on this site, is that anyone who uses AI to help them write\nshould declare how they used it, usually in the\nacknowledgments. This is partly for transparency, and partly so that\nreaders can learn useful ways to use AI. In my own case, I don&#39;t use AI\nfor generating prose, nor for help in structure. The only way I use it for\nwriting is as a search tool (but I verify any information by going\nto the source).</p>\n<p><a href=\"https://martinfowler.com/bliki/SayYourWriting.html\" rel=\"nofollow ugc noopener\">Say Your Writing</a> is\neven more important in the age of LLMs. I habitually speak the words in my head\nand then write them down (and then edit what I&#39;ve just written). This advice\nis particularly important to anyone who has any desire to communicate\nregularly with writing. Like many things, learning to be a good writer relies\non lots of practice. It&#39;s a skill worth the time to acquire, but surrendering\nto an LLM will undermine any progress.</p>\n<p>In time it may be that LLMs will be so good that we can&#39;t tell the\ndifference. In many ways, that&#39;s more worrying. Like most writers, I find\nwriting an almost-essential tool for thinking. Often it&#39;s only by writing\nsomething down, that I can really get my thoughts on a topic into coherence.\nI&#39;ve often started an essay knowing I don&#39;t really understand something, but\nthe act of wrestling with the words forces me to overcome my confusions. As\nmore people delegate writing to LLMs they will also delegate thinking and\ncomprehension to LLMs. I don&#39;t want to do that, where&#39;s the fun of writing then?</p>\n<p>Although avoiding LLM-Voice is a general rule, it isn&#39;t an absolute one. An\nimportant example is that of Jim Highsmith (who contributes to this site). He\nfinds working with an LLM is essential to his writing to <a href=\"https://bjsm.bmj.com/content/60/9/677\" rel=\"nofollow ugc noopener\">overcome the bane of Parkinson&#39;s\ndisease</a>. I&#39;ll gladly deal with LLM-Voice in order to continue imbibing my\nold friend&#39;s wisdom.</p>\n<p>A more awkward case is that of translation. Many of my colleagues aren&#39;t\ncomfortable in English, it is their second or fourth language. Some have found\nLLM translation an important part to their communication with those of us who\nknow rather less Marathi. But I also would prefer rather a genuine human&#39;s clunky prose to the\nsmooth but soulless LLM.</p>\n<p>There are situations where it would be reasonable for even a native\nEnglishman to let LLMs write the prose. Such as if I&#39;m writing a form letter\nto an official, or preparing a report I suspect no-one will ever read 3. Maybe\nmost documentation in the future will only be read by LLMs, and they won&#39;t\nhave my uncanny valley worries. But if I want to communicate with humans, my\nhumanity has to shine (or at least ooze) out of my prose.</p>\n<p>3: \nMichael Nygard commented: “There you really aren&#39;t\ntrying to communicate thoughts from one person to other people. You&#39;re\ncommunicating with a process rather than a human. A human may open the\nenvelope but that person is acting as a component in a cybernetic\norganism.”</p>\n<h2 id=\"notes\">Notes</h2>\n<p>1: \nVery hard, but not impossible. I&#39;ve come across one case where I do know\nthe first draft was entirely done by LLM, but I detected no LLM Voice when\nI read the final piece. It did, however, need a professional editor at a\nmajor periodical publication to clean it up.</p>\n<p>2: \nMy policy on this site, is that anyone who uses AI to help them write\nshould declare how they used it, usually in the\nacknowledgments. This is partly for transparency, and partly so that\nreaders can learn useful ways to use AI. In my own case, I don&#39;t use AI\nfor generating prose, nor for help in structure. The only way I use it for\nwriting is as a search tool (but I verify any information by going\nto the source).</p>\n<p>3: \nMichael Nygard commented: “There you really aren&#39;t\ntrying to communicate thoughts from one person to other people. You&#39;re\ncommunicating with a process rather than a human. A human may open the\nenvelope but that person is acting as a component in a cybernetic\norganism.”</p>\n<h2 id=\"acknowledgements\">Acknowledgements</h2>\n<p>Alessio Ferri, Caer Sanders, Ian Cartwright, Jie Jessie Xia, Joshua\nKerievsky, Michael Feathers, Michael Nygard, Scott Davies, and Wei Zhang commented on\ndrafts of this post.</p>\n<p>I did not use AI or LLMs in any way while writing this post.</p>\n<hr />\n<p><em>Originally published at <a href=\"https://martinfowler.com/bliki/LLM-Voice.html\" rel=\"nofollow ugc noopener\">martinfowler.com</a>.</em></p>","headings":[{"level":1,"text":"LLM Voice","id":"llm-voice"},{"level":2,"text":"Notes","id":"notes"},{"level":2,"text":"Acknowledgements","id":"acknowledgements"}]}}