Skip to the content

The Court of Public Prose

The docket · In re Dead Writing Theory

One matter at a time Every case, with its verdict.

Every case is a real study. Ours fix the question before measuring and publish the method and data on ScriptGrain; third-party studies are credited to their authors and linked to the original, caveats stated. 15 cases so far (two ours, 13 third-party), 37 exhibits in the record, the next one being measured.

§ IOn the docket

The cases

  1. Case No. DWT-001Filed In re
    Dead Writing
    Theory
    27 years of company homepagesScriptGrain SGR-004
    The words changed. The sentences did not.2,950 homepages from 158 long-lived UK and US companies, one capture a year, 2000 to 2026.SplitverdictExhibits P-1 to P-4 · D-1 to D-3Read the case
  2. Case No. DWT-002Filed In re
    Dead Writing
    Theory
    Where the AI-isms came fromScriptGrain SGR-007
    Not an invention. An inheritance.17,000 pieces of writing from before ChatGPT: 6,600 web pages in 16 kinds of writing, 103 homepages, 10,000 arXiv abstracts, 600 passages from pre-1928 books, Google Books 1800 to 2019.Verdict for the defenceExhibits D-4 to D-7Read the case
  3. Case No. DWT-003Filed In re
    Dead Writing
    Theory
    A quarter of the press releaseLiang et al., Patterns 6(12), 2025
    Third-party studyThe machine is in the room. In a lot of rooms.687,241 consumer complaints, 537,413 corporate press releases, 304.3 million job postings and 15,919 UN press releases, January 2022 to September 2024.Verdict for the prosecutionExhibits P-5 to P-6Read the case
  4. Case No. DWT-004Filed In re
    Dead Writing
    Theory
    The news that didn't flattenFitterer et al., Proceedings of ACL, Student Research Workshop, 2025
    Third-party studyThe fingerprints are there. The flattening is not.Two random samples of roughly 30,000 news articles each from the News on the Web (NOW) corpus, one from 2018 and one from 2024.Verdict for the defenceExhibits D-8 to D-9Read the case
  5. Case No. DWT-005Filed In re
    Dead Writing
    Theory
    The websites nobody wroteHe et al., arXiv preprint 2605.00087, 2026
    Third-party studyPreprintMore sites with little human input. More of them selling.Roughly 100k sites archived by Common Crawl and roughly 20k sites found in Bing results for 10k how-to search queries, each site classified from several of its pages.Verdict for the prosecutionExhibits P-7 to P-8Read the case
  6. Case No. DWT-006Filed In re
    Dead Writing
    Theory
    Half the articles, and a plateauParedes et al., Graphite research (Five Percent), 2026
    Third-party studyVendor studyHalf the articles. Then the rise stopped.55.4k random English article URLs from Common Crawl, published between January 2020 and March 2026, each scored by three commercial AI detectors (Pangram, Copyleaks, GPTZero) and averaged.SplitverdictExhibits P-9 · D-10Read the case
  7. Case No. DWT-007Filed In re
    Dead Writing
    Theory
    Better stories, more alikeDoshi and Hauser, Science Advances, 2024
    Third-party studyThe stories grew alike. Each one got better.293 writers wrote eight-sentence stories with no AI help, one GPT-4 story idea or up to five; 600 evaluators made 3,519 evaluations; similarity measured with text embeddings.SplitverdictExhibits P-10 · D-11Read the case
  8. Case No. DWT-008Filed In re
    Dead Writing
    Theory
    The delve study, done properlyKobak et al., Science Advances 11(27), 2025
    Third-party studyThe machine's words spread through medicine, measured without a detector.15.1 million English-language PubMed abstracts from 2010 to 2024; each word's 2024 frequency compared with a projection from its 2021 to 2022 trend.Verdict for the prosecutionExhibits P-11 to P-12Read the case
  9. Case No. DWT-009Filed In re
    Dead Writing
    Theory
    The detector that convicts the innocentLiang et al., Patterns 4(7), 2023
    Third-party studyPreprintThe detectors flagged human writers as machines.Seven widely used GPT detectors, run in 2023 on 91 human-written TOEFL essays from a Chinese forum and 88 US eighth-grade essays from the Hewlett Foundation's ASAP dataset.Verdict for the defenceExhibits D-12 to D-13Read the case
  10. Case No. DWT-010Filed In re
    Dead Writing
    Theory
    The shrinking rangeSourati et al., Nature Human Behaviour, 2026
    Third-party studyPolished by a model, the styles drew closer together.Over 880,000 texts in seven datasets. Monthly series from January 2018: 80,238 arXiv computer science abstracts, 318,490 Reddit creative stories (both to November 2024) and 379,583 Patch community news articles (to November 2023); plus 1,000 pre-ChatGPT Reddit and arXiv texts each, polished by GPT-3.5, Llama 3 70B and Gemini Pro.Verdict for the prosecutionExhibits P-13 to P-14Read the case
  11. Case No. DWT-011Filed In re
    Dead Writing
    Theory
    Half the articles, but who reads them?Paredes et al., Graphite research (Five Percent), 2025
    Third-party studyVendor studyThe machine writes half. Readers are mostly shown the other half.31,493 keywords across 10 categories; the first two Google results pages for each in June 2025, plus ChatGPT and Perplexity citations for 100 keywords per category; English articles and listicles classified in 500-word chunks with Surfer's AI detector.Verdict for the defenceExhibits D-14 to D-15Read the case
  12. Case No. DWT-012Filed In re
    Dead Writing
    Theory
    The scientists got there firstLiang et al., Nature Human Behaviour 9(12), 2025
    Third-party studyComputer scientists took up the machine's words first. Mathematicians, much less.1,121,912 papers from January 2020 to September 2024 (861,253 from arXiv, 205,094 from bioRxiv, 55,565 from Nature portfolio journals); abstracts and introductions, estimated with a population-level word-frequency model.Verdict for the prosecutionExhibits P-15 · D-16Read the case
  13. Case No. DWT-013Filed In re
    Dead Writing
    Theory
    Delve, out loudYakura et al., arXiv preprint (version 4, July 2026), 2026
    Third-party studyPreprintPeople started saying "delve". Then they stopped.737,083 hours of conversation from 824,634 podcast episodes, screened for unscripted speech, compared with a synthetic control built from words ChatGPT does not favour; plus a preregistered experiment with 496 people.SplitverdictExhibits P-16 · D-17Read the case
  14. Case No. DWT-014Filed In re
    Dead Writing
    Theory
    Three quarters touchedLaw et al., Ahrefs blog, 2025
    Third-party studyVendor studyThree quarters of new pages had some AI in them. 2.5% were all AI.900,000 English-language pages newly found by Ahrefs' crawler in April 2025, one page per domain, scored by Ahrefs' own AI detector (bot_or_not) for the share of each page detected as AI.SplitverdictExhibits P-17 · D-18Read the case
  15. Case No. DWT-015Filed In re
    Dead Writing
    Theory
    The em dash goes to CongressCzuma, arXiv preprint (preregistered on OSF), 2026
    Third-party studyPreprintCongress's press releases doubled their em dashes. Two years after ChatGPT.146,239 press releases from 480 House and Senate offices, 2021 to 2025, from the open congress-press dataset; unspaced em dashes in prose counted per 1,000 characters after stripping the dateline, against the preregistered ChatGPT cut-off of 30 November 2022.SplitverdictExhibits P-18 · D-19Read the case

§ IIThe docket so far

A ledger, never one score.

The exhibits each case entered, side by side. A count of exhibits is not a measure of truth. How evidence is admitted

Cases by verdict, with the exhibits each one entered
Case No.MatterProsecutionDefenceVerdict
DWT-00127 years of company homepages43Splitverdict
DWT-002Where the AI-isms came from04Verdict for the defence
DWT-003A quarter of the press release · Third-party study20Verdict for the prosecution
DWT-004The news that didn't flatten · Third-party study02Verdict for the defence
DWT-005The websites nobody wrote · Third-party study20Verdict for the prosecution
DWT-006Half the articles, and a plateau · Third-party study (vendor)11Splitverdict
DWT-007Better stories, more alike · Third-party study11Splitverdict
DWT-008The delve study, done properly · Third-party study20Verdict for the prosecution
DWT-009The detector that convicts the innocent · Third-party study02Verdict for the defence
DWT-010The shrinking range · Third-party study20Verdict for the prosecution
DWT-011Half the articles, but who reads them? · Third-party study (vendor)02Verdict for the defence
DWT-012The scientists got there first · Third-party study11Verdict for the prosecution
DWT-013Delve, out loud · Third-party study11Splitverdict
DWT-014Three quarters touched · Third-party study (vendor)11Splitverdict
DWT-015The em dash goes to Congress · Third-party study11Splitverdict
DWT-016Being measured––Pending, under investigation