# Dead Writing Theory

> Dead Writing Theory says the web's writing is collapsing into one machine-made voice. This site tests the claim with data, one case at a time. Published by Adapt Progress Evolve Limited (UK).

Last updated: 2026-10-18

Dead Writing Theory tests one claim, the theory that since ChatGPT, the writing on the web has been collapsing into one machine voice. Same words, same rhythm, same punctuation, everywhere. It is framed as a trial: the prosecution argues the theory (evidence for), the defence answers it (evidence against), every finding is an exhibit (P-n for the prosecution, D-n for the defence) and every study is a case on the docket (Case No. DWT-nnn) with a verdict. We don't argue for the theory or against it. We test it, case by case, and publish the evidence on both sides. Our own studies are measured with deterministic code, never a language model; other people's studies are credited, linked and weighed with their caveats. "The Court of Public Prose" is a figure of speech, not a real court.

Dead Writing Theory is a nod to the Dead Internet Theory: the idea that much of the web is now made by machines, for machines. We take the narrower, measurable corner of it, the writing: prose, copy, advertising and marketing. Not affiliated with the Dead Internet Theory or any site, forum or person associated with it.

## Key pages
- [Home](https://deadwritingtheory.com/): the theory, our stance, every exhibit (18 for the prosecution, 19 for the defence) and the docket. Markdown: https://deadwritingtheory.com/index.md
- [The theory](https://deadwritingtheory.com/theory/): the indictment: three counts (same words, same rhythm, same punctuation), both sides' opening statements, all 37 exhibits by side, how evidence is admitted, and FAQ. Markdown: https://deadwritingtheory.com/theory.md
- [The docket](https://deadwritingtheory.com/cases/): every case with its verdict and the next matter being measured. Markdown: https://deadwritingtheory.com/cases.md

## Cases
- [Case No. DWT-001: 27 years of company homepages. The words changed. The sentences did not.](https://deadwritingtheory.com/cases/did-chatgpt-change-how-companies-write/) (split verdict; ScriptGrain SGR-004; updated 2026-10-04): On company homepages the "AI voice" is mostly a vocabulary, and an older one than its name suggests. ChatGPT appears to have sped the drift up rather than started it. Markdown: https://deadwritingtheory.com/cases/did-chatgpt-change-how-companies-write.md
- [Case No. DWT-002: Where the AI-isms came from. Not an invention. An inheritance.](https://deadwritingtheory.com/cases/where-did-ai-isms-come-from/) (verdict for the defence; ScriptGrain SGR-007; updated 2026-10-04): The "AI voice" looks less like an invention than an inheritance: the vocabulary of corporate announcements laid over the rhetorical shapes of older speeches and sermons. Markdown: https://deadwritingtheory.com/cases/where-did-ai-isms-come-from.md
- [Case No. DWT-003: A quarter of the press release. The machine is in the room. In a lot of rooms.](https://deadwritingtheory.com/cases/a-quarter-of-the-press-release/) (verdict for the prosecution; third-party study: Liang et al., Patterns 6(12), 2025; updated 2026-10-04): By late 2024, roughly 24% of corporate press releases involved LLM-assisted writing. The study measures adoption, not sameness, but the machine is in a lot of rooms. Markdown: https://deadwritingtheory.com/cases/a-quarter-of-the-press-release.md
- [Case No. DWT-004: The news that didn't flatten. The fingerprints are there. The flattening is not.](https://deadwritingtheory.com/cases/the-news-that-didnt-flatten/) (verdict for the defence; third-party study: Fitterer et al., Proceedings of ACL, Student Research Workshop, 2025; updated 2026-10-04): Machine-style words rose in English news between 2018 and 2024, but lexical diversity did not fall. A real point for the defence on vocabulary, not a knockout. Markdown: https://deadwritingtheory.com/cases/the-news-that-didnt-flatten.md
- [Case No. DWT-005: The websites nobody wrote. More sites with little human input. More of them selling.](https://deadwritingtheory.com/cases/the-websites-nobody-wrote/) (verdict for the prosecution; third-party study: He et al., arXiv preprint 2605.00087, 2026; updated 2026-10-04): LLM-dominant websites rose from 2.1% to 29.4% of sites sampled between late 2022 and early 2025, and they turn up in how-to search results. A preprint, detector-based. Markdown: https://deadwritingtheory.com/cases/the-websites-nobody-wrote.md
- [Case No. DWT-006: Half the articles, and a plateau. Half the articles. Then the rise stopped.](https://deadwritingtheory.com/cases/half-the-articles-and-a-plateau/) (split verdict; third-party study, vendor: Paredes et al., Graphite research (Five Percent), 2026; updated 2026-10-09): A vendor study: primarily AI-generated articles reached about half of new English articles by Q1 2025, then the share stopped rising. Detector-based, not peer reviewed. Markdown: https://deadwritingtheory.com/cases/half-the-articles-and-a-plateau.md
- [Case No. DWT-007: Better stories, more alike. The stories grew alike. Each one got better.](https://deadwritingtheory.com/cases/better-stories-more-alike/) (split verdict; third-party study: Doshi and Hauser, Science Advances, 2024; updated 2026-10-04): In a preregistered experiment, AI story ideas made each story better and the stories as a group more alike: individually better off, collectively narrower. Markdown: https://deadwritingtheory.com/cases/better-stories-more-alike.md
- [Case No. DWT-008: The delve study, done properly. The machine's words spread through medicine, measured without a detector.](https://deadwritingtheory.com/cases/the-delve-study-done-properly/) (verdict for the prosecution; third-party study: Kobak et al., Science Advances 11(27), 2025; updated 2026-10-09): Excess words in 15.1 million PubMed abstracts: "delves" ran 28.0 times its expected rate in 2024, and at least 13.5% of 2024 abstracts were processed with LLMs. Markdown: https://deadwritingtheory.com/cases/the-delve-study-done-properly.md
- [Case No. DWT-009: The detector that convicts the innocent. The detectors flagged human writers as machines.](https://deadwritingtheory.com/cases/the-detector-that-convicts-the-innocent/) (verdict for the defence; third-party study: Liang et al., Patterns 4(7), 2023; updated 2026-10-09): Seven AI detectors labelled more than half of 91 human-written TOEFL essays as AI (average false-positive rate 61.3%). Any exhibit that rests on a detector needs that caveat. Markdown: https://deadwritingtheory.com/cases/the-detector-that-convicts-the-innocent.md
- [Case No. DWT-010: The shrinking range. Polished by a model, the styles drew closer together.](https://deadwritingtheory.com/cases/the-shrinking-range/) (verdict for the prosecution; third-party study: Sourati et al., Nature Human Behaviour, 2026; updated 2026-10-09): Across 880,000+ texts, writing-complexity variance fell after ChatGPT, and LLM polishing cut it by 21 to 50%. The strongest evidence for convergence so far, mostly from rewrite tests. Markdown: https://deadwritingtheory.com/cases/the-shrinking-range.md
- [Case No. DWT-011: Half the articles, but who reads them?. The machine writes half. Readers are mostly shown the other half.](https://deadwritingtheory.com/cases/half-the-articles-but-who-reads-them/) (verdict for the defence; third-party study, vendor: Paredes et al., Graphite research (Five Percent), 2025; updated 2026-10-09): A vendor study: 86% of articles ranking in Google and 82% cited by ChatGPT and Perplexity were human-written. Published is not the same as read. Detector-based. Markdown: https://deadwritingtheory.com/cases/half-the-articles-but-who-reads-them.md
- [Case No. DWT-012: The scientists got there first. Computer scientists took up the machine's words first. Mathematicians, much less.](https://deadwritingtheory.com/cases/the-scientists-got-there-first/) (verdict for the prosecution; third-party study: Liang et al., Nature Human Behaviour 9(12), 2025; updated 2026-10-09): By September 2024, 22.5% of computer science abstract sentences were LLM-modified, against 7.7% in mathematics. The machine is in science, but not evenly. Markdown: https://deadwritingtheory.com/cases/the-scientists-got-there-first.md
- [Case No. DWT-013: Delve, out loud. People started saying "delve". Then they stopped.](https://deadwritingtheory.com/cases/delve-out-loud/) (split verdict; third-party study: Yakura et al., arXiv preprint (version 4, July 2026), 2026; updated 2026-10-09): ChatGPT's favourite words rose in unscripted podcast speech, "delve" to about 44% above expected. Then use fell below expected once people noticed. A preprint. Markdown: https://deadwritingtheory.com/cases/delve-out-loud.md
- [Case No. DWT-014: Three quarters touched. Three quarters of new pages had some AI in them. 2.5% were all AI.](https://deadwritingtheory.com/cases/three-quarters-touched/) (split verdict; third-party study, vendor: Law et al., Ahrefs blog, 2025; updated 2026-10-09): A vendor study: 74.2% of 900,000 new web pages contained AI-generated content, but only 2.5% were pure AI. It counts any AI at all, with an unpublished error rate. Markdown: https://deadwritingtheory.com/cases/three-quarters-touched.md
- [Case No. DWT-015: The em dash goes to Congress. Congress's press releases doubled their em dashes. Two years after ChatGPT.](https://deadwritingtheory.com/cases/the-em-dash-goes-to-congress/) (split verdict; third-party study: Czuma, arXiv preprint (preregistered on OSF), 2026; updated 2026-10-18): Unspaced em dashes in 146,239 US congressional press releases more than doubled in 2025, within the same offices. But nothing moved when ChatGPT arrived. A preregistered preprint. Markdown: https://deadwritingtheory.com/cases/the-em-dash-goes-to-congress.md

## Blog
https://deadwritingtheory.com/blog/
- [What does "AI slop" mean, and is it the same thing as machine voice?](https://deadwritingtheory.com/blog/what-does-ai-slop-mean/) (2026-10-27): AI slop is low-quality content made in bulk with AI and pushed at people who never asked for it. It is a verdict on quality. It does not prove who wrote a page, or that the web is converging. Markdown: https://deadwritingtheory.com/blog/what-does-ai-slop-mean.md
- [Kurzgesagt on AI Slop: Which Claims Can Evidence Check, and Which Are Opinion?](https://deadwritingtheory.com/blog/kurzgesagt-ai-slop-what-the-evidence-can-check/) (2026-10-22): Kurzgesagt's video says AI slop is destroying the internet. Some of its claims can be measured and some are judgement. YouTube's reported mislabelling of the channel shows how fragile a one-item 'slop' verdict is. Markdown: https://deadwritingtheory.com/blog/kurzgesagt-ai-slop-what-the-evidence-can-check.md
- [What Is Model Collapse, and Is It the Same Thing as the Web Writing Alike?](https://deadwritingtheory.com/blog/what-is-model-collapse/) (2026-10-20): Model collapse is the degradation of models trained, generation after generation, on earlier models' output, rare content first. It is a training-loop result, not a reading of today's web. Markdown: https://deadwritingtheory.com/blog/what-is-model-collapse.md
- [The Microsoft CEO Wants to Move Past "AI Slop". What Can the Evidence Actually Say?](https://deadwritingtheory.com/blog/microsoft-ceo-ai-slop-what-the-court-can-say/) (2026-10-15): Satya Nadella wrote on 29 December 2025 that we need to get beyond "slop vs sophistication". Slop is a quality judgement and machine voice is a measurable pattern, and the court record is built to test the second. Markdown: https://deadwritingtheory.com/blog/microsoft-ceo-ai-slop-what-the-court-can-say.md
- ["Your AI Slop Bores Me": what the joke site can and cannot tell us about machine voice](https://deadwritingtheory.com/blog/your-ai-slop-bores-me-what-it-shows/) (2026-10-13): Your AI Slop Bores Me is a browser game in which people answer each other's prompts while pretending to be an AI. It shows that readers can recognise and imitate a machine voice. It does not measure whether the web's writing has converged on one. Markdown: https://deadwritingtheory.com/blog/your-ai-slop-bores-me-what-it-shows.md

## How pages work
- Our own cases (2) are ScriptGrain studies with the question fixed before measuring, measured with deterministic code (never a language model), and linked to their full method and data. Third-party cases (13) are other people's studies entered into evidence: each names its authors, institution, venue and year, says whether it is peer reviewed (and whether it is a vendor study), lists its caveats and links to the original. Their numbers are quoted from the original, not re-measured. A count of exhibits is a ledger, not a verdict.
- Every page is written by Jack Stovell, lists its sources, and shows when it was published and last updated (the updated date changes only when the content does).
- Every page has a markdown mirror: append .md to its path (e.g. https://deadwritingtheory.com/index.md, https://deadwritingtheory.com/cases.md).
- Research by ScriptGrain: Ours: made by the same studio that runs this site.

## Site
- [About](https://deadwritingtheory.com/about/): Who writes and publishes Dead Writing Theory, and how cases are made, sourced and kept up to date. Markdown: https://deadwritingtheory.com/about.md
- [Privacy](https://deadwritingtheory.com/privacy/): What Dead Writing Theory records when you read it: anonymous first-party page analytics, no cookies, no accounts. Markdown: https://deadwritingtheory.com/privacy.md
- Full text of every page in one file: https://deadwritingtheory.com/llms-full.txt
- Submit evidence (a page that sounds like every other page): hello@deadwritingtheory.com

## Publisher and author
- Publisher: Adapt Progress Evolve Limited, company 10047328 (England and Wales), registered office 5 Brayford Square, London E1 0SG, VAT GB362654684. https://adaptprogressevolve.com/ · Contact: hello@deadwritingtheory.com
- Author: Jack Stovell. Profile: https://i.hilyt.it/jack

## From the same studio
- [ScriptGrain](https://scriptgrain.com/): Measures your writing style from samples you upload, builds a voice profile and drafts new content in it, scoring every draft against the profile.
- [ReviewNudge](https://reviewnudge.app/): Free, no-account tools for small businesses: a review reply writer, a review request builder, a Google review QR card maker and a testimonial card maker.
- [Launch Video Maker](https://launchvideomaker.com/): Turns a business website into a short motion video with music, using only facts you approve, paid once per video.
- [How to get AI to](https://howtogetaito.com/): A free library of step-by-step guides to getting AI to do a specific task, with prompts, tools, cost and privacy notes.
- [What is GEO](https://whatisgeo.app/): A free, source-linked guide to Generative Engine Optimisation: how websites show up in answers from ChatGPT, Perplexity and Google's AI Overviews.
- [visibilitytoai](https://visibilitytoai.com/): Measures whether ChatGPT, Claude, Gemini and Perplexity name and cite a company when buyers ask them questions, quoting every answer.
- [Hilyt](https://hilyt.it/): Open profiles for people and businesses that AI assistants can read and cite: stable URLs, schema.org data and a plain-text card per profile.
- [Assets for three.js](https://assetsforthreejs.com/): Original 3D model packs (GLB) for three.js and WebGL, each model with three levels of detail and its triangle count published.
- [Forematter](https://forematter.com/): Employee-onboarding software that turns a company handbook into a paced first month for new starters. In development, not yet open to customers.
- [Can You Use AI](https://canyouuseai.com/): Straight answers to "Can you use AI to ...?" questions: a verdict, how to do it well, and the UK rules and risks, each with sources.