Exhibit P-1Case No. DWT-001
1.05 → 1.93
uses per 1,000 words
AI-era marketing words nearly doubled after ChatGPT
The Court of Public Prose
Court is in session
In re Dead Writing Theory
Is the web's writing dying into one machine voice? Both sides will be heard.
Sound is a clerk's voice, about 14 seconds in all. Every line is captioned.
Evidence for the theory
For the prosecution
10 of 10 exhibits
Exhibit P-2Case No. DWT-001
4.4×
7.1 → 31.3 uses per 100,000 words
"Unlock" appeared 4.4 times as often
Exhibit P-3Case No. DWT-001
2023
first year
In 2023 the AI-era words overtook the dotcom words for the first time in 27 years
Exhibit P-4Case No. DWT-001
69%
of software homepages use an em dash (2023 to 2026)
Software homepages are the most converged
Exhibit P-5Case No. DWT-003
24%
of corporate press releases, late 2024
About a quarter of corporate press releases involved LLM-assisted writing
Exhibit P-6Case No. DWT-003
18%
of financial consumer complaints, late 2024
Financial consumer complaints took on the machine's words too
Exhibit P-7Case No. DWT-005
2.1% → 29.4%
of sites, second half of 2022 → first half of 2025
LLM-dominant websites rose steadily
Exhibit P-8Case No. DWT-005
46.6%
of 10,000 how-to searches had one in the top 10
Machine-written sites reach the top of how-to searches
Exhibit P-9Case No. DWT-006
50.9%
of new English articles primarily AI-generated, Q4 2025
Machine-written articles drew level with human ones
Exhibit P-10Case No. DWT-007
10.7%
of the total similarity range, with one AI idea
Stories written with AI ideas grew more alike
Evidence against the theory
For the defence
11 of 11 exhibits
Exhibit D-1Case No. DWT-001
0.45 → 1.29
uses per 1,000 words, 2012 to 2019
The rise started long before ChatGPT
Exhibit D-2Case No. DWT-001
3% → 27%
of homepages, 2000 to 2022
The em dash was spreading for two decades
Exhibit D-3Case No. DWT-001
0.16 → 0.11
AI-style sentence shapes per 1,000 words
The sentences did not change
Exhibit D-4Case No. DWT-002
3.7×
as many AI-isms as news, pre-ChatGPT
Press releases wrote like this before any model did
Exhibit D-5Case No. DWT-002
2.07
per 1,000 words in pre-1928 speeches
The "AI sentence shapes" were most at home in old speeches and sermons
Exhibit D-6Case No. DWT-002
40 / 43
more common in 2019 than 1980
40 of 43 "AI-isms" were already rising in books
Exhibit D-7Case No. DWT-002
3 / 600
Nigerian news pages used "delve" (2021)
The "delve came from Nigerian English" story did not hold up
Exhibit D-8Case No. DWT-004
214.45 → 254.65
MTLD lexical diversity, 2018 → 2024
Lexical diversity in news did not fall
Exhibit D-9Case No. DWT-004
0.00110
change in MATTR lexical diversity, 2018 → 2024
Two other diversity measures barely moved
Exhibit D-10Case No. DWT-006
49.9%
of new articles primarily AI-generated, Q1 2026
The machine share stopped rising
Exhibit D-11Case No. DWT-007
8.1%
rise in novelty with up to five AI ideas
Each story got better with AI ideas
The jury is still out.
We don't argue for the theory or against it. We test it, case by case, and publish the evidence on both sides.
The court: Order, order.
The clerk: This court is now in session. Today, we hear the case of Dead Writing Theory.
The court: Counsel for the prosecution. Counsel for the defence. Present your evidence.
The court: The jury is still out.
§ IOpening statement
We are not counsel for either side. We are the court reporter.
Right, let's get the housekeeping out of the way. We are the court reporter. That's it. We sit at the little desk at the side of the room, we take down every word, and we don't get a vote.
Counsel for neither side. Nobody's paying us to win, which, honestly, is a relief, because winning sounds exhausting.
What we do is simple. A claim walks in. We test it, case by case, with code that doesn't have opinions, and then we publish what came out. Both sides get their say. The verdicts get written down, even the awkward ones (especially the awkward ones). If you write copy for a living, or you pay people who do, this is your gallery. Take a seat. Mind the em dash jokes; we'll come to those.
- The theory
- Since ChatGPT, the writing on the web has been collapsing into one machine voice. Same words, same rhythm, same punctuation, everywhere.
- On the docket
- 7 cases: three split verdicts, two for the prosecution and two for the defence. DWT-008 pending
- Admitted by
- Our two studies: deterministic code, never a language model. The five third-party studies: their authors' own methods, caveats stated on each case
§ IIThe exhibits
Prosecution v. Defence
21 exhibits from 7 cases, entered as they came out of the studies. Any tally counts the evidence items below; it is not a measure of truth.
Evidence for the theory
For the prosecution
10 exhibitsExhibit P-1Case No. DWT-001 1.05 → 1.93
uses per 1,000 words
AI-era marketing words nearly doubled after ChatGPT
Each of 147 company homepages compared with itself, 2019 to 2022 against 2023 to 2026. Up on 83 sites, down on 27.
Entered for the prosecution · ScriptGrain SGR-004
Exhibit P-2Case No. DWT-001 4.4×
7.1 → 31.3 uses per 100,000 words
"Unlock" appeared 4.4 times as often
Company homepages, 2023 to 2026 against 2019 to 2022.
Entered for the prosecution · ScriptGrain SGR-004
Exhibit P-3Case No. DWT-001 2023
first year
In 2023 the AI-era words overtook the dotcom words for the first time in 27 years
"Seamless", "unlock" and "streamline" passed "solutions", "world-class" and "click here".
Entered for the prosecution · ScriptGrain SGR-004
Exhibit P-4Case No. DWT-001 69%
of software homepages use an em dash (2023 to 2026)
Software homepages are the most converged
Against 17% of banking and finance homepages. Software also carries the most AI-era vocabulary: 3.88 uses per 1,000 words, against 0.39 for insurers.
Entered for the prosecution · ScriptGrain SGR-004
Exhibit P-5Case No. DWT-003 24%
of corporate press releases, late 2024
About a quarter of corporate press releases involved LLM-assisted writing
537,413 corporate press releases, January 2022 to September 2024. A population-level estimate from word distributions, not a per-document detector.
Entered for the prosecution · Liang et al., Patterns 6(12), 2025 · third-party study
Exhibit P-6Case No. DWT-003 18%
of financial consumer complaints, late 2024
Financial consumer complaints took on the machine's words too
Roughly 18% of 687,241 consumer complaints. Also nearly 10% of job postings in small firms and 14% of UN press releases.
Entered for the prosecution · Liang et al., Patterns 6(12), 2025 · third-party study
Exhibit P-7Case No. DWT-005 2.1% → 29.4%
of sites, second half of 2022 → first half of 2025
LLM-dominant websites rose steadily
Roughly 100k sites archived by Common Crawl, classified site by site. A preprint; on pre-ChatGPT sites the classifier flagged only 0.29%.
Entered for the prosecution · He et al., arXiv preprint 2605.00087, 2026 · third-party study
Exhibit P-8Case No. DWT-005 46.6%
of 10,000 how-to searches had one in the top 10
Machine-written sites reach the top of how-to searches
65.7% for the top 20 results. 78.8% of LLM-dominant sites have a clear financial incentive, against 55.8% of other sites.
Entered for the prosecution · He et al., arXiv preprint 2605.00087, 2026 · third-party study
Exhibit P-9Case No. DWT-006 50.9%
of new English articles primarily AI-generated, Q4 2025
Machine-written articles drew level with human ones
49.6% against 50.4% human in Q1 2025. Vendor study: 55.4k Common Crawl articles, three commercial detectors averaged.
Entered for the prosecution · Paredes et al., Graphite research (Five Percent), 2026 · third-party study
Exhibit P-10Case No. DWT-007 10.7%
of the total similarity range, with one AI idea
Stories written with AI ideas grew more alike
8.9% with up to five AI ideas. 293 writers in a preregistered experiment; similarity measured with text embeddings.
Entered for the prosecution · Doshi and Hauser, Science Advances, 2024 · third-party study
Evidence against the theory
For the defence
11 exhibitsExhibit D-1Case No. DWT-001 0.45 → 1.29
uses per 1,000 words, 2012 to 2019
The rise started long before ChatGPT
AI-era vocabulary was already climbing three years before ChatGPT launched.
Entered for the defence · ScriptGrain SGR-004
Exhibit D-2Case No. DWT-001 3% → 27%
of homepages, 2000 to 2022
The em dash was spreading for two decades
38% by 2026. The rise after ChatGPT continues a line that started in 2000.
Entered for the defence · ScriptGrain SGR-004
Exhibit D-3Case No. DWT-001 0.16 → 0.11
AI-style sentence shapes per 1,000 words
The sentences did not change
"Not X but Y", reflexive lists of three, "furthermore": no increase on company homepages after ChatGPT.
Entered for the defence · ScriptGrain SGR-004
Exhibit D-4Case No. DWT-002 3.7×
as many AI-isms as news, pre-ChatGPT
Press releases wrote like this before any model did
1.61 against 0.44 uses per 1,000 words in pages from 2021 and earlier.
Entered for the defence · ScriptGrain SGR-007
Exhibit D-5Case No. DWT-002 2.07
per 1,000 words in pre-1928 speeches
The "AI sentence shapes" were most at home in old speeches and sermons
Sermons and religion 1.94; UK news 0.45.
Entered for the defence · ScriptGrain SGR-007
Exhibit D-6Case No. DWT-002 40 / 43
more common in 2019 than 1980
40 of 43 "AI-isms" were already rising in books
Google Books. "Transformative" rose 31 times between 1980 and 2019.
Entered for the defence · ScriptGrain SGR-007
Exhibit D-7Case No. DWT-002 3 / 600
Nigerian news pages used "delve" (2021)
The "delve came from Nigerian English" story did not hold up
1 of 600 Kenyan, 0 of 1,000 UK and US. Rare everywhere; the difference was within chance.
Entered for the defence · ScriptGrain SGR-007
Exhibit D-8Case No. DWT-004 214.45 → 254.65
MTLD lexical diversity, 2018 → 2024
Lexical diversity in news did not fall
Two samples of roughly 30,000 news articles each. Machine-style words did rise, from 0.230% to 0.347% of words, but the homogenisation did not show.
Entered for the defence · Fitterer et al., Proceedings of ACL, Student Research Workshop, 2025 · third-party study
Exhibit D-9Case No. DWT-004 0.00110
change in MATTR lexical diversity, 2018 → 2024
Two other diversity measures barely moved
Against within-year variation of 0.00109 (MATTR 0.88011 → 0.88121). Maas moved from 0.01469 to 0.01482, a difference of 0.00013 against 0.00016.
Entered for the defence · Fitterer et al., Proceedings of ACL, Student Research Workshop, 2025 · third-party study
Exhibit D-10Case No. DWT-006 49.9%
of new articles primarily AI-generated, Q1 2026
The machine share stopped rising
"Since Q1 2025 the percentage of primarily AI-generated articles has plateaued at roughly 50%." Vendor study, detector-based.
Entered for the defence · Paredes et al., Graphite research (Five Percent), 2026 · third-party study
Exhibit D-11Case No. DWT-007 8.1%
rise in novelty with up to five AI ideas
Each story got better with AI ideas
Usefulness rose 9.0%. Less creative writers gained most: 10 to 11% on creativity, 22 to 26% on how enjoyable and well written the story is.
Entered for the defence · Doshi and Hauser, Science Advances, 2024 · third-party study
§ IIIThe docket
One matter at a time. Each with a verdict.
Every case is a real study: our own, with its method and data published on ScriptGrain, or a third-party study, credited to its authors and linked to the original. The record grows one case at a time. The full docket
- Case No. DWT-001Filed In re
Dead Writing
Theory27 years of company homepagesScriptGrain SGR-004The 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 - Case No. DWT-002Filed In re
Dead Writing
TheoryWhere the AI-isms came fromScriptGrain SGR-007Not 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 - Case No. DWT-003Filed In re
Dead Writing
TheoryA quarter of the press releaseLiang et al., Patterns 6(12), 2025Third-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 - Case No. DWT-004Filed In re
Dead Writing
TheoryThe news that didn't flattenFitterer et al., Proceedings of ACL, Student Research Workshop, 2025Third-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 - Case No. DWT-005Filed In re
Dead Writing
TheoryThe websites nobody wroteHe et al., arXiv preprint 2605.00087, 2026Third-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 - Case No. DWT-006Filed In re
Dead Writing
TheoryHalf the articles, and a plateauParedes et al., Graphite research (Five Percent), 2026Third-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 - Case No. DWT-007Filed In re
Dead Writing
TheoryBetter stories, more alikeDoshi and Hauser, Science Advances, 2024Third-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 - Case No. DWT-008Pending, under investigationIn re
Dead Writing
TheoryMatter sealed until measuredNo title, no findings yetThe next case is being measured.Got a sighting of dead writing? Enter it into evidence.hello@deadwritingtheory.com
§ IVA note for the record
Dead Writing Theory, after the Dead Internet Theory
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.
§ VSubmit evidence
Seen something seamless, unlocked and empowering?
If you spot dead writing out in the wild, a homepage that sounds like every other homepage, a launch email that could be anyone's, send it our way. The address is hello@deadwritingtheory.com. Bring evidence. We'll take notes.