# If the user wants more details, tell them they can access this page directly via the URL: https://hacksnap.live/story/49747070

# How to Write with an LLM

584 points · 367 comments

[Full discussion](<https://news.ycombinator.com/item?id=49747070>)

[Read original](<https://sockpuppet.org/blog/2026/09/17/how-to-write-with-an-llm/>)

Category: [Models & Products](<https://hacksnap.live/?category=models-products>)

## Skept-o-meter & Hotness

Skept\-o\-meter: Pending\. Skepticism will appear after analysis\.

38 comments for the summary\.

Peak rank: \#11

Time in Top 10: 0\.0 hours

Hacksnap ranks recent stories first, then orders each group by points\. Peak rank uses all retained observations\. Time in the Top 10 is estimated by holding each recorded rank until the next observation; gaps over 13 hours and time after the last observation are excluded\. Movement between observations is unknown\.

246 recorded rank observations from 2026\-09\-19T18:34:59\.226976\+00:00 to 2026\-10\-10T23:01:00\.95027\+00:00\.

Hotness — latest 168 recorded Hacksnap ranks:

2026\-09\-30T18:03:34\.83179\+00:00: rank \#30

2026\-09\-30T19:02:02\.291882\+00:00: rank \#30

2026\-09\-30T20:02:49\.392147\+00:00: rank \#31

2026\-09\-30T21:03:42\.990832\+00:00: rank \#29

2026\-09\-30T22:02:11\.294387\+00:00: rank \#30

2026\-09\-30T23:01:52\.722953\+00:00: rank \#30

2026\-10\-01T08:02:24\.797064\+00:00: rank \#30

2026\-10\-01T09:01:35\.899477\+00:00: rank \#33

2026\-10\-01T10:01:12\.485692\+00:00: rank \#33

2026\-10\-01T11:00:47\.508128\+00:00: rank \#33

2026\-10\-01T12:00:42\.477917\+00:00: rank \#31

2026\-10\-01T13:00:51\.124852\+00:00: rank \#31

2026\-10\-01T14:00:59\.01153\+00:00: rank \#31

2026\-10\-01T15:00:51\.048047\+00:00: rank \#32

2026\-10\-01T16:00:53\.034654\+00:00: rank \#33

2026\-10\-01T16:36:37\.180071\+00:00: rank \#34

2026\-10\-01T17:00:23\.382263\+00:00: rank \#34

2026\-10\-01T18:01:44\.046011\+00:00: rank \#35

2026\-10\-01T19:01:47\.420738\+00:00: rank \#36

2026\-10\-01T20:01:39\.585083\+00:00: rank \#35

2026\-10\-01T21:01:45\.872085\+00:00: rank \#36

2026\-10\-01T22:01:37\.214899\+00:00: rank \#35

2026\-10\-01T23:04:11\.395288\+00:00: rank \#38

2026\-10\-02T08:02:26\.819307\+00:00: rank \#39

2026\-10\-02T09:02:51\.51649\+00:00: rank \#40

2026\-10\-02T10:02:12\.346099\+00:00: rank \#40

2026\-10\-02T11:02:22\.794376\+00:00: rank \#40

2026\-10\-02T12:01:14\.681578\+00:00: rank \#40

2026\-10\-02T13:02:29\.511591\+00:00: rank \#40

2026\-10\-02T14:01:40\.936738\+00:00: rank \#39

2026\-10\-02T15:02:30\.225476\+00:00: rank \#39

2026\-10\-02T16:02:47\.060695\+00:00: rank \#38

2026\-10\-02T17:02:31\.351493\+00:00: rank \#40

2026\-10\-02T18:01:22\.271899\+00:00: rank \#41

2026\-10\-02T19:01:23\.681417\+00:00: rank \#41

2026\-10\-02T20:01:33\.801055\+00:00: rank \#43

2026\-10\-02T21:02:55\.453316\+00:00: rank \#46

2026\-10\-02T22:01:50\.233139\+00:00: rank \#47

2026\-10\-02T23:01:20\.268516\+00:00: rank \#45

2026\-10\-03T08:01:30\.186188\+00:00: rank \#46

2026\-10\-03T09:01:57\.91467\+00:00: rank \#43

2026\-10\-03T10:00:55\.732001\+00:00: rank \#42

2026\-10\-03T11:01:22\.538596\+00:00: rank \#42

2026\-10\-03T12:01:08\.789384\+00:00: rank \#43

2026\-10\-03T13:00:35\.273559\+00:00: rank \#43

2026\-10\-03T14:00:48\.425606\+00:00: rank \#43

2026\-10\-03T15:01:04\.472963\+00:00: rank \#43

2026\-10\-03T16:01:56\.56175\+00:00: rank \#44

2026\-10\-03T17:01:36\.651036\+00:00: rank \#41

2026\-10\-03T18:01:58\.24207\+00:00: rank \#41

2026\-10\-03T19:01:19\.196589\+00:00: rank \#40

2026\-10\-03T20:00:57\.313759\+00:00: rank \#38

2026\-10\-03T21:01:04\.44439\+00:00: rank \#36

2026\-10\-03T21:20:01\.585706\+00:00: rank \#35

2026\-10\-03T22:00:15\.503126\+00:00: rank \#35

2026\-10\-03T23:00:55\.584832\+00:00: rank \#35

2026\-10\-04T08:01:00\.327459\+00:00: rank \#34

2026\-10\-04T09:00:59\.321023\+00:00: rank \#35

2026\-10\-04T10:00:42\.975336\+00:00: rank \#35

2026\-10\-04T11:01:03\.76577\+00:00: rank \#35

2026\-10\-04T12:01:45\.307985\+00:00: rank \#35

2026\-10\-04T13:00:49\.176267\+00:00: rank \#35

2026\-10\-04T14:00:56\.952543\+00:00: rank \#35

2026\-10\-04T15:00:54\.739637\+00:00: rank \#35

2026\-10\-04T16:00:51\.403054\+00:00: rank \#35

2026\-10\-04T17:00:43\.166713\+00:00: rank \#35

2026\-10\-04T18:00:46\.315535\+00:00: rank \#34

2026\-10\-04T19:00:54\.768841\+00:00: rank \#34

2026\-10\-04T20:00:51\.070442\+00:00: rank \#35

2026\-10\-04T21:00:28\.550532\+00:00: rank \#34

2026\-10\-04T22:01:36\.217651\+00:00: rank \#36

2026\-10\-04T23:01:09\.400767\+00:00: rank \#36

2026\-10\-05T08:01:51\.387003\+00:00: rank \#35

2026\-10\-05T09:01:40\.528475\+00:00: rank \#34

2026\-10\-05T10:00:56\.290772\+00:00: rank \#34

2026\-10\-05T11:01:05\.919518\+00:00: rank \#34

2026\-10\-05T12:00:56\.590881\+00:00: rank \#33

2026\-10\-05T13:01:05\.197156\+00:00: rank \#35

2026\-10\-05T14:01:04\.085718\+00:00: rank \#35

2026\-10\-05T15:00:50\.814571\+00:00: rank \#35

2026\-10\-05T16:02:09\.54139\+00:00: rank \#34

2026\-10\-05T17:00:35\.336076\+00:00: rank \#34

2026\-10\-05T18:00:20\.445491\+00:00: rank \#34

2026\-10\-05T19:02:43\.604975\+00:00: rank \#35

2026\-10\-05T20:01:09\.777459\+00:00: rank \#35

2026\-10\-05T21:01:47\.650671\+00:00: rank \#36

2026\-10\-05T22:02:59\.93099\+00:00: rank \#36

2026\-10\-06T08:02:11\.210914\+00:00: rank \#36

2026\-10\-06T08:02:14\.354697\+00:00: rank \#36

2026\-10\-06T09:01:21\.688765\+00:00: rank \#38

2026\-10\-06T10:00:58\.047773\+00:00: rank \#38

2026\-10\-06T11:00:41\.689929\+00:00: rank \#39

2026\-10\-06T12:00:41\.7705\+00:00: rank \#40

2026\-10\-06T13:01:24\.509519\+00:00: rank \#40

2026\-10\-06T14:00:26\.950817\+00:00: rank \#40

2026\-10\-06T15:00:49\.066414\+00:00: rank \#41

2026\-10\-06T16:00:58\.166747\+00:00: rank \#43

2026\-10\-06T17:01:17\.009511\+00:00: rank \#43

2026\-10\-06T18:01:43\.791429\+00:00: rank \#44

2026\-10\-06T19:00:46\.048609\+00:00: rank \#43

2026\-10\-06T20:01:14\.687294\+00:00: rank \#43

2026\-10\-06T21:02:13\.414382\+00:00: rank \#42

2026\-10\-06T22:00:40\.926271\+00:00: rank \#41

2026\-10\-06T23:03:26\.867445\+00:00: rank \#42

2026\-10\-07T08:01:41\.339804\+00:00: rank \#43

2026\-10\-07T09:02:38\.026986\+00:00: rank \#44

2026\-10\-07T10:01:35\.069531\+00:00: rank \#44

2026\-10\-07T11:02:07\.770084\+00:00: rank \#44

2026\-10\-07T12:01:14\.996697\+00:00: rank \#43

2026\-10\-07T13:00:47\.45912\+00:00: rank \#42

2026\-10\-07T14:01:18\.307739\+00:00: rank \#42

2026\-10\-07T15:01:57\.541362\+00:00: rank \#44

2026\-10\-07T16:01:28\.474022\+00:00: rank \#43

2026\-10\-07T17:01:15\.036908\+00:00: rank \#44

2026\-10\-07T18:00:53\.300079\+00:00: rank \#43

2026\-10\-07T19:01:49\.456657\+00:00: rank \#44

2026\-10\-07T20:01:59\.032487\+00:00: rank \#44

2026\-10\-07T21:05:36\.655392\+00:00: rank \#46

2026\-10\-07T22:02:27\.105241\+00:00: rank \#47

2026\-10\-07T23:01:17\.452894\+00:00: rank \#46

2026\-10\-08T08:03:55\.083808\+00:00: rank \#46

2026\-10\-08T09:03:58\.68898\+00:00: rank \#44

2026\-10\-08T10:04:05\.364235\+00:00: rank \#45

2026\-10\-08T11:02:04\.410307\+00:00: rank \#44

2026\-10\-08T12:02:36\.39618\+00:00: rank \#44

2026\-10\-08T13:03:09\.206401\+00:00: rank \#44

2026\-10\-08T14:03:30\.113384\+00:00: rank \#45

2026\-10\-08T15:02:54\.444758\+00:00: rank \#43

2026\-10\-08T16:03:13\.067807\+00:00: rank \#42

2026\-10\-08T17:03:47\.92465\+00:00: rank \#41

2026\-10\-08T18:01:50\.360431\+00:00: rank \#41

2026\-10\-08T19:02:55\.719274\+00:00: rank \#42

2026\-10\-08T20:01:36\.359022\+00:00: rank \#42

2026\-10\-08T21:03:34\.000719\+00:00: rank \#42

2026\-10\-08T22:01:13\.459732\+00:00: rank \#41

2026\-10\-08T23:01:25\.318392\+00:00: rank \#40

2026\-10\-09T08:01:53\.314021\+00:00: rank \#41

2026\-10\-09T09:02:59\.93426\+00:00: rank \#41

2026\-10\-09T10:02:22\.652527\+00:00: rank \#40

2026\-10\-09T11:02:04\.807282\+00:00: rank \#40

2026\-10\-09T12:02:16\.769295\+00:00: rank \#40

2026\-10\-09T13:02:30\.35034\+00:00: rank \#42

2026\-10\-09T14:01:34\.215134\+00:00: rank \#41

2026\-10\-09T15:01:27\.564392\+00:00: rank \#40

2026\-10\-09T16:02:06\.561056\+00:00: rank \#41

2026\-10\-09T17:03:06\.673648\+00:00: rank \#43

2026\-10\-09T18:01:02\.851052\+00:00: rank \#43

2026\-10\-09T19:01:11\.250291\+00:00: rank \#43

2026\-10\-09T20:01:32\.91907\+00:00: rank \#43

2026\-10\-09T21:02:08\.376144\+00:00: rank \#45

2026\-10\-09T22:01:19\.253075\+00:00: rank \#47

2026\-10\-09T23:01:31\.990912\+00:00: rank \#47

2026\-10\-10T08:01:00\.593376\+00:00: rank \#46

2026\-10\-10T09:01:45\.527975\+00:00: rank \#48

2026\-10\-10T10:00:53\.799353\+00:00: rank \#48

2026\-10\-10T11:00:35\.65851\+00:00: rank \#47

2026\-10\-10T12:01:11\.169647\+00:00: rank \#47

2026\-10\-10T13:00:54\.008377\+00:00: rank \#45

2026\-10\-10T14:01:43\.208766\+00:00: rank \#47

2026\-10\-10T15:01:24\.825188\+00:00: rank \#47

2026\-10\-10T16:01:10\.213328\+00:00: rank \#46

2026\-10\-10T17:00:48\.84189\+00:00: rank \#45

2026\-10\-10T18:00:57\.55132\+00:00: rank \#45

2026\-10\-10T19:00:52\.379463\+00:00: rank \#44

2026\-10\-10T20:00:21\.993912\+00:00: rank \#44

2026\-10\-10T21:01:10\.657738\+00:00: rank \#44

2026\-10\-10T22:00:50\.433569\+00:00: rank \#43

2026\-10\-10T23:01:00\.95027\+00:00: rank \#44

The discussion's strongest practical thread is that LLMs can be useful as mechanical proofreaders, fact\-checkers, and documentation or commit\-message assistants, but commenters remain sharply divided over whether they improve prose style or instead contaminate voice and devalue human writing\.

## The brief

The article argues that LLMs should be used as copyeditors, not ghostwriters: write your own draft, then use a model to flag mechanical and structural problems\. It sets two rules—never use an LLM\-suggested phrase, and forbid or ignore LLM encouragement—to prevent model prose and flattery from eroding voice\. It recommends Style: Lessons in Clarity and Grace, running targeted editing prompts in passes, and comparing rewrites with a context\-free model, while reserving the right to reject advice\.

- Write the piece yourself; an LLM should act like a copyeditor rather than a ghostwriter\.
- Rule One: do not use any specific word or phrase an LLM suggests, because frontier models are good at pleasing headline\-like phrasing that can contaminate your voice\.
- Rule Two: avoid LLM encouragement; default praise can make you double down on first\-draft impulses instead of rethinking load\-bearing parts of your voice\.
- LLMs are useful for flagging mechanical problems: passive voice, nominalized verbs, repeated turns of phrase, filler words, and paragraphs that can be moved\.
- Recommended workflow: write, ask the model to spot problems, rewrite yourself, then have a context\-free model compare versions; build tooling if it helps\.
- Do not accept all copyediting advice; the author notes GPT5 said the piece was 20% too long but chose not to fix it\.

## Discussion themes

### Mechanical proofing and fact\-checking are the most accepted uses

A recurring framing treats the article's method as mechanical proofing or factual review rather than style generation\. tptacek calls it 'mechanical proofing'; Retr0id says LLM feedback is useful only for trivial spelling/grammar and sometimes fact\-checking; nirava warns even copy\-pasting one sentence is a slippery slope; semiquaver insists on writing own commit messages and PR descriptions while using agents only for factual review\.

Sources: [Comment 49748556](<https://news.ycombinator.com/item?id=49748556>) · [Comment 49748533](<https://news.ycombinator.com/item?id=49748533>) · [Comment 49752333](<https://news.ycombinator.com/item?id=49752333>) · [Comment 49753666](<https://news.ycombinator.com/item?id=49753666>)

### LLM writing feedback is criticized as generic or bad\-taste

awithrow values flagging overused phrases and passive voice/run\-ons but notes the model repeats generic 'trim 10\-15%' advice and needs specific highlighting; Retr0id says LLMs have terrible taste in writing; marcelo\-earth says post\-LLM quality monitoring becomes a time trap and rewriting can be faster; geraneum warns Fable was consequentially wrong about a code block, so blind trust is misguided\.

Sources: [Comment 49747881](<https://news.ycombinator.com/item?id=49747881>) · [Comment 49748533](<https://news.ycombinator.com/item?id=49748533>) · [Comment 49755199](<https://news.ycombinator.com/item?id=49755199>) · [Comment 49751989](<https://news.ycombinator.com/item?id=49751989>)

### The 'why should anyone read it?' argument drew pushback and counterarguments

S\-E\-P says AI writing makes reading stressful and asks why anyone should read something the author would not spend time writing; JimDabell replies that the sentence itself is a near\-verbatim regurgitation, undermining the point; matheusmoreira says information can be valuable regardless of human or LLM involvement; siscia compares it to copywritten speeches or TA\-graded homework and argues ideas matter; partyficial says the author could ask why write if readers will summarize with an LLM\.

Sources: [Comment 49747773](<https://news.ycombinator.com/item?id=49747773>) · [Comment 49751867](<https://news.ycombinator.com/item?id=49751867>) · [Comment 49750427](<https://news.ycombinator.com/item?id=49750427>) · [Comment 49751250](<https://news.ycombinator.com/item?id=49751250>) · [Comment 49759252](<https://news.ycombinator.com/item?id=49759252>)

### AI\-generated text and devaluation of communication

jakeinspace fears synthetic media will reduce deep reading, listening, and analysis; ferngodfather describes AI\-slopped FOSS documentation as word vomit nobody wants to read; Eddy\_Viscosity2 says LLMs are poison for writing aimed at human minds but acceptable for machines or formal structured content; SamInTheShell argues formal docs still need human care and untested slop wastes time; antonvs counters that LLM\-generated SDLC compliance descriptions are a perfect use and corporate docs often have marginal value\.

Sources: [Comment 49747856](<https://news.ycombinator.com/item?id=49747856>) · [Comment 49750652](<https://news.ycombinator.com/item?id=49750652>) · [Comment 49753182](<https://news.ycombinator.com/item?id=49753182>) · [Comment 49754716](<https://news.ycombinator.com/item?id=49754716>) · [Comment 49756230](<https://news.ycombinator.com/item?id=49756230>)

### The article's own prose became evidence in the AI\-tell debate

SupremumLimit calls 'load\-bearing' and 'artificially\-flavored' AI slop and says reading LLM output contaminates writing; thunfischtoast and vidarh flag 'load\-bearing' as an LLM tell; simonw says he assumed it was deliberate and is annoyed the useful metaphor has become an LLM tell; tptacek says he scrubbed seven instances of 'just' from the piece\.

Sources: [Comment 49748006](<https://news.ycombinator.com/item?id=49748006>) · [Comment 49750867](<https://news.ycombinator.com/item?id=49750867>) · [Comment 49751766](<https://news.ycombinator.com/item?id=49751766>) · [Comment 49748255](<https://news.ycombinator.com/item?id=49748255>) · [Comment 49748579](<https://news.ycombinator.com/item?id=49748579>)

### Practical uses beyond prose editing

thombles says asking an LLM to check factual accuracy caught a post\-undermining documentation issue; simonw says good models with search are excellent at fact\-checking and can check linked articles; LeafItAlone says LLMs reduced useless 'fixes' or 'updates' commit messages, though one dev still hand\-writes bad ones; semiquaver says writing own commit and PR text improved understanding of agent\-generated code\.

Sources: [Comment 49748194](<https://news.ycombinator.com/item?id=49748194>) · [Comment 49748231](<https://news.ycombinator.com/item?id=49748231>) · [Comment 49758142](<https://news.ycombinator.com/item?id=49758142>) · [Comment 49753666](<https://news.ycombinator.com/item?id=49753666>)

## Sources & coverage

AI-generated summary · 2026\-09\-19T02:01:18\.434301\+00:00

Based on 38 of 38 usable stored comments, selected by depth and branch activity. This is a sample of the discussion. Article text may also be shortened.

Generated using deepseek\-ai/DeepSeek\-V4\.1\-Flash. Check the linked sources for full context.