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
The Court of Public Prose
In re Dead Writing Theory
- Case No.
- DWT-001
- Matter
- 27 years of company homepages
- Filed
- Verdict
- Splitverdict
The words changed. The sentences did not.
- The question
- How has the writing on company homepages changed over 27 years, and did anything shift after ChatGPT arrived in November 2022?
- The sample
- 2,950 homepages from 158 long-lived UK and US companies, one capture a year, 2000 to 2026
- Filed
- The study
- ScriptGrain SGR-004 Ours: made by the same studio that runs this site.
§ IFindings of fact
What the study found.
- AI-era marketing words nearly doubled after ChatGPT: 1.05 to 1.93 uses per 1,000 words, each of 147 sites compared with itself.
- "Unlock" appeared 4.4 times as often in 2023 to 2026 as in 2019 to 2022.
- AI-era vocabulary was already rising: 0.45 per 1,000 words in 2012, 1.29 in 2019.
- In 2023, AI-era words overtook the dotcom vocabulary for the first time in 27 years.
- 3% of homepages used an em dash in 2000, 27% in 2022, 38% in 2026.
- The sentence shapes associated with AI writing did not increase: 0.16 per 1,000 words before, 0.11 after.
- The median homepage carried 169 words of prose in 2000 and 606 in 2026.
§ IIThe court reporter's account
27 years of company homepages.
By Jack Stovell
Our first case is a big, patient one. Study SGR-004, from ScriptGrain, took 2,950 homepages from 158 long-lived UK and US companies, one archived capture a year, 2000 to 2026. The clever bit is that each site is compared with itself: a company against its own younger self.
The words moved. That much is clear. AI-era marketing language nearly doubled after ChatGPT, rose on 83 sites and fell on 27, and in 2023 it overtook the dotcom vocabulary for the first time in 27 years. The prosecution had a good day.
But the clock tells a different story. That same word family was already climbing: 0.45 per 1,000 words in 2012, 1.29 in 2019. The em dash told a similar tale: 3% in 2000, then 27% in 2022, and 38% by 2026. Something was already happening.
And the sentences? They sat still. The shapes people blame on AI (“not X but Y”, reflexive lists of three, “furthermore”) ran at 0.16 per 1,000 words before and 0.11 after. No surge. No flood.
So the verdict is split. The words changed; the sentences did not. ChatGPT appears to have sped up a drift, not started it. Which is, I mean, a very unsatisfying thing to hand a jury, but it’s what the data said, and we don’t get to edit the data to be more exciting.
§ IIIThe crossover
Exhibit P-3Exhibit D-1
In 2023 the AI-era words overtook the dotcom words.
Seamless, unlock and streamline passed solutions, world-class and click here for the first time in 27 years. But the rise began long before: 0.45 per 1,000 words in 2012, 1.29 in 2019. Mean uses per 1,000 words of homepage prose, every company measured each year.
AI-era vs dotcom vocabulary, per 1,000 words
01234
200020052010201520202026
ChatGPT, Nov 20220.45 in 20122023: 1.50 vs 1.422.17- AI-era words
- Dotcom words
| Year | AI-era words | Dotcom words |
|---|---|---|
| 2000 | 0.42 | 3.65 |
| 2001 | 0.49 | 3.18 |
| 2002 | 0.29 | 2.41 |
| 2003 | 0.24 | 3.28 |
| 2004 | 0.11 | 3.45 |
| 2005 | 0.14 | 2.62 |
| 2006 | 0.24 | 2.7 |
| 2007 | 0.58 | 2.07 |
| 2008 | 0.34 | 1.89 |
| 2009 | 0.24 | 1.61 |
| 2010 | 0.44 | 2.06 |
| 2011 | 0.32 | 2.34 |
| 2012 | 0.45 | 2.07 |
| 2013 | 0.42 | 2.17 |
| 2014 | 0.57 | 2.46 |
| 2015 | 0.83 | 1.69 |
| 2016 | 0.64 | 1.94 |
| 2017 | 0.82 | 2 |
| 2018 | 0.96 | 2 |
| 2019 | 1.29 | 1.93 |
| 2020 | 0.86 | 1.55 |
| 2021 | 1.12 | 1.68 |
| 2022 | 1.07 | 1.42 |
| 2023 | 1.5 | 1.42 |
| 2024 | 1.86 | 1.38 |
| 2025 | 1.79 | 1.35 |
| 2026 | 2.17 | 1.77 |
§ IVThe em dash
Exhibit D-2 · for the defence
The em dash was spreading long before any model.
Share of homepages with at least one em dash. 3% in 2000, 27% in 2022, 38% in 2026: the rise after ChatGPT continues a line that started in 2000.
Share of homepages using an em dash
0%10%20%30%40%
200020052010201520202026
ChatGPT, Nov 20223%27%38%| Year | Share with an em dash |
|---|---|
| 2000 | 0.03 |
| 2001 | 0.07 |
| 2002 | 0.08 |
| 2003 | 0.05 |
| 2004 | 0.08 |
| 2005 | 0.1 |
| 2006 | 0.1 |
| 2007 | 0.09 |
| 2008 | 0.12 |
| 2009 | 0.1 |
| 2010 | 0.16 |
| 2011 | 0.18 |
| 2012 | 0.19 |
| 2013 | 0.16 |
| 2014 | 0.18 |
| 2015 | 0.21 |
| 2016 | 0.2 |
| 2017 | 0.23 |
| 2018 | 0.27 |
| 2019 | 0.29 |
| 2020 | 0.3 |
| 2021 | 0.28 |
| 2022 | 0.27 |
| 2023 | 0.32 |
| 2024 | 0.29 |
| 2025 | 0.35 |
| 2026 | 0.38 |
§ VThe sentences
Exhibit D-3 · for the defence
The sentences did not change.
AI-style sentence shapes (not X but Y, reflexive lists of three, furthermore) went from 0.16 to 0.11 per 1,000 words after ChatGPT. Median sentence length: 7.9 words in 2000, 9.02 in 2026.
Median sentence length, words
04812
200020052010201520202026
ChatGPT, Nov 20227.99.02| Year | Median sentence length |
|---|---|
| 2000 | 7.9 |
| 2001 | 8.22 |
| 2002 | 7.69 |
| 2003 | 7.84 |
| 2004 | 8.11 |
| 2005 | 8.09 |
| 2006 | 8.19 |
| 2007 | 8.03 |
| 2008 | 8 |
| 2009 | 7.79 |
| 2010 | 7.31 |
| 2011 | 7.58 |
| 2012 | 7.36 |
| 2013 | 7.73 |
| 2014 | 7.88 |
| 2015 | 8.09 |
| 2016 | 8.06 |
| 2017 | 8.55 |
| 2018 | 8.48 |
| 2019 | 8.83 |
| 2020 | 9.23 |
| 2021 | 9.2 |
| 2022 | 9.14 |
| 2023 | 8.8 |
| 2024 | 9.11 |
| 2025 | 9.53 |
| 2026 | 9.02 |
§ VIThe homepage
For the record
Homepages got longer: 169 words to 606.
The median homepage carried 169 words of prose in 2000 and 606 in 2026. Which is why every rate in this record is per 1,000 words, never a raw count.
§ VIIThe word-by-word record
Exhibit P-1 · for the prosecutionExhibit P-2
Word by word, before and after.
The same 147 homepages, each compared with itself: 351,878 words of prose in 2019 to 2022, 421,966 in 2023 to 2026. Outline: uses per 1,000 words before. Solid: after. Type area is drawn to the rate.
AI-era words
- unlock 0.071 → 0.31325 → 132 uses
- ai-powered 0.031 → 0.20411 → 86 uses
- seamless 0.080 → 0.16628 → 70 uses
- journey 0.125 → 0.14544 → 61 uses
- empower 0.148 → 0.14252 → 60 uses
- at scale 0.040 → 0.10214 → 43 uses
- streamline 0.031 → 0.08511 → 36 uses
- elevate 0.054 → 0.07819 → 33 uses
- unleash 0.026 → 0.0559 → 23 uses
- reimagin 0.026 → 0.0509 → 21 uses
- effortless 0.028 → 0.04510 → 19 uses
- in seconds 0.017 → 0.0406 → 17 uses
- revolutioniz 0.014 → 0.0385 → 16 uses
- landscape 0.031 → 0.03611 → 15 uses
- cutting-edge 0.020 → 0.0367 → 15 uses
- harness 0.023 → 0.0368 → 15 uses
Dotcom words
- solutions 0.634 → 0.550223 → 232 uses
- click here 0.298 → 0.149105 → 63 uses
- welcome to 0.119 → 0.10442 → 44 uses
- premier 0.063 → 0.06422 → 27 uses
- industry-leading 0.054 → 0.05019 → 21 uses
- world-class 0.028 → 0.06410 → 27 uses
- mission-critical 0.011 → 0.0124 → 5 uses
- value-added 0.009 → 0.0023 → 1 uses
- synergy 0.006 → 0.0002 → 0 uses
- one-stop 0.006 → 0.0122 → 5 uses
| Word | Set | Uses 2019–22 | Uses 2023–26 | Per 1,000 before | Per 1,000 after |
|---|---|---|---|---|---|
| unlock | AI-era | 25 | 132 | 0.071 | 0.313 |
| ai-powered | AI-era | 11 | 86 | 0.031 | 0.204 |
| seamless | AI-era | 28 | 70 | 0.080 | 0.166 |
| journey | AI-era | 44 | 61 | 0.125 | 0.145 |
| empower | AI-era | 52 | 60 | 0.148 | 0.142 |
| at scale | AI-era | 14 | 43 | 0.040 | 0.102 |
| streamline | AI-era | 11 | 36 | 0.031 | 0.085 |
| elevate | AI-era | 19 | 33 | 0.054 | 0.078 |
| unleash | AI-era | 9 | 23 | 0.026 | 0.055 |
| reimagin | AI-era | 9 | 21 | 0.026 | 0.050 |
| effortless | AI-era | 10 | 19 | 0.028 | 0.045 |
| in seconds | AI-era | 6 | 17 | 0.017 | 0.040 |
| revolutioniz | AI-era | 5 | 16 | 0.014 | 0.038 |
| landscape | AI-era | 11 | 15 | 0.031 | 0.036 |
| cutting-edge | AI-era | 7 | 15 | 0.020 | 0.036 |
| harness | AI-era | 8 | 15 | 0.023 | 0.036 |
| redefin | AI-era | 17 | 14 | 0.048 | 0.033 |
| next-level | AI-era | 22 | 14 | 0.063 | 0.033 |
| supercharge | AI-era | 4 | 12 | 0.011 | 0.028 |
| powered by ai | AI-era | 1 | 4 | 0.003 | 0.009 |
| embark | AI-era | 2 | 4 | 0.006 | 0.009 |
| transformative | AI-era | 2 | 4 | 0.006 | 0.009 |
| revolutionis | AI-era | 0 | 3 | 0.000 | 0.007 |
| tapestry | AI-era | 2 | 2 | 0.006 | 0.005 |
| realm | AI-era | 0 | 2 | 0.000 | 0.005 |
| navigate the | AI-era | 6 | 2 | 0.017 | 0.005 |
| game changer | AI-era | 3 | 2 | 0.009 | 0.005 |
| game-changer | AI-era | 4 | 2 | 0.011 | 0.005 |
| solutions | Dotcom | 223 | 232 | 0.634 | 0.550 |
| click here | Dotcom | 105 | 63 | 0.298 | 0.149 |
| welcome to | Dotcom | 42 | 44 | 0.119 | 0.104 |
| premier | Dotcom | 22 | 27 | 0.063 | 0.064 |
| industry-leading | Dotcom | 19 | 21 | 0.054 | 0.050 |
| world-class | Dotcom | 10 | 27 | 0.028 | 0.064 |
| mission-critical | Dotcom | 4 | 5 | 0.011 | 0.012 |
| value-added | Dotcom | 3 | 1 | 0.009 | 0.002 |
| synergy | Dotcom | 2 | 0 | 0.006 | 0.000 |
| one-stop | Dotcom | 2 | 5 | 0.006 | 0.012 |
§ VIIIThe sectors
Exhibit P-4 · for the prosecution
Software is the most converged.
69% of software homepages use an em dash since 2023, against 17% of banking and finance. Software also carries the most AI-era vocabulary: 3.88 uses per 1,000 words, against 0.39 for insurers.
- Software28 sites2.69 → 3.8869%
- Hardware13 sites1.16 → 3.4024%
- Agencies and consultancies10 sites1.59 → 2.6139%
- Travel13 sites0.41 → 1.6516%
- Web and marketing tools9 sites0.75 → 1.2959%
- Consumer brands15 sites0.87 → 1.2717%
- Logistics and cars10 sites1.23 → 1.2126%
- Retail24 sites0.28 → 1.1718%
- Banking and finance18 sites0.31 → 0.9217%
- Telecoms and media10 sites0.48 → 0.8933%
- Insurance8 sites0.18 → 0.3921%
§ IXThe exhibits entered in this case
Prosecution v. Defence
Any tally counts the evidence items below; it is not a measure of truth.
Evidence for the theory
For the prosecution
4 exhibitsExhibit P-1Case No. DWT-001 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
Evidence against the theory
For the defence
3 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
§ XThe finding
The finding
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.
§ XIQuestions for the court
Asked and answered.
Did ChatGPT change how companies write their homepages?
Partly. AI-era marketing words nearly doubled after ChatGPT, from 1.05 to 1.93 uses per 1,000 words, but they were already rising before it (0.45 in 2012, 1.29 in 2019), and the sentence shapes associated with AI writing did not increase.
Did the em dash take off after ChatGPT?
It was already spreading. 3% of homepages used an em dash in 2000, 27% in 2022 and 38% in 2026: the rise after ChatGPT continues a line that started in 2000.
Why is the verdict split?
The words changed and the sentences did not. The vocabulary moved towards the theory; the sentence shapes, and how early the vocabulary and the em dash started rising, count against it.
How was it measured?
2,950 homepages from 158 long-lived UK and US companies, one capture a year from 2000 to 2026, each site compared with itself. Counted with deterministic code, never a language model. The full method and data are published on ScriptGrain.