Brands over time

innocent
over time

Every year keeps its colour across every chart.│ ChatGPT, 30 Nov 2022

54 of innocent's own blog posts, press releases and announcements, 2015 to 2018, measured sentence by sentence, 54 of them recovered from the Internet Archive. How the writing has moved over the years, how it compares with 9 other food and drink brands, and what it sounds like in its own words.

Pages measured
54
Years covered
2015–18
Archived before ChatGPT
54
Blog · press
54 · 0

01 · Homepage

One capture a year

innocent's homepage from 2000 to 2026, closest to 1 July, from the Internet Archive. Pick a year or press play. Each headline is quoted from its capture and links to the archived copy.

2026innocentdrinks.co.uk25 Jun 2026
innocent homepage, 25 Jun 2026

Screenshot of innocent’s homepage (innocentdrinks.co.uk) as archived by the Internet Archive on 25 Jun 2026. © innocent or its licensors. Shown for comparison and review; not endorsed by innocent.

Homepage

2026

“start the day right”

Page title

home

Open on the Internet Archive
  1. 2012Stories from our blog…
  2. 2021you will become unexpectedly interested in crushed fruit and companies that give 10% of profits to charity
  3. 2024our little bottles
  4. 2026start the day right

21 captures. Screenshots are rendered from the archive and can pull in images archived later than the page itself; dates come from the capture. Years without a capture are greyed out in the strip.

02 · Measures by year

How the writing moved

Every year’s pages, averaged. Pick a measure. Hatched bars rest on fewer than 10 pages: context, not a finding. The page count sits under each year.

Sentence lengthAverage words per sentence.
5
10
15
Food avg 12.7
10.7
11.5
12.7
11.6
ChatGPT · Nov 2022
201516 pp
201614 pp
201713 pp
201811 pp
2019no pages
2020no pages
2021no pages
2022no pages
2023no pages
2024no pages
2025no pages
2026no pages
10 or more pagesFewer than 10 pages: context, not a findingSector averageChatGPT released
Show as a table
YearSentence lengthPages
201510.716
201611.514
201712.713
201811.611

Years with pages: 2015 (16), 2016 (14), 2017 (13), 2018 (11). Dates come from page metadata, structured data, the URL, or the month a page was first archived (a latest possible date).

03 · Against its sector

innocent among food and drink brands

Each dot is one brand’s mean across all its measured pages; the tick is the sector average. Only brands with 10 or more measured pages are shown or counted. Select a dot to open that brand.

  1. Em dashes

    80% less than its sector

    0.39 against a sector average of 1.97; lowest of 7.

  2. Questions

    69% more than its sector

    6.0 against a sector average of 3.6; 2nd highest of 7.

  3. Long words

    25% less than its sector

    16% against a sector average of 21%; lowest of 7.

Sentence lengthMid-table, 4th of 7
innocent 11.6
8.420.8
Contractions2nd highest of 7
innocent 28.1
8.746.8
"You"3rd lowest of 7
innocent 49%
10%63%
"We"3rd highest of 7
innocent 45%
29%87%
Questions2nd highest of 7
innocent 6.0
0.89.0
Long wordsLowest of 7
innocent 16%
16%29%
Em dashesLowest of 7
innocent 0.39
0.394.05
AI-associated words2nd lowest of 7
innocent 0.32
0.181.61
ExclamationsMid-table, 4th of 7
innocent 1.89
0.389.76
Page length3rd lowest of 7
innocent 464
3311,158

Pages measured: innocent 54 · Pret A Manger 5 · Greggs 63 · Nando's 184 · LEON 0 · wagamama 0 · itsu 129 · COOK 168 · GAIL's 15 · Deliciously Ella 133. Measures and method: SGR-010; the brand list and sectors come from SGR-009.

04 · Vocabulary

Words innocent uses more than its peers

The words this brand reaches for more often than the rest of its sector. This is subject matter more than voice.

you're60.0 vs 4.2
ask37.7 vs 0.7
information48.5 vs 3.0
date35.4 vs 1.6
bit57.7 vs 5.5
fruit42.3 vs 3.3
pineapple21.5 vs 0.5
message20.8 vs 0.4
person26.9 vs 1.4
seems20.0 vs 0.3
sorry19.2 vs 0.4
words19.2 vs 0.4
innocentSector peers

Uses per 10,000 words, ranked by weighted log-odds (Monroe et al. 2008) across 54 pages. Product names and page furniture filtered automatically; list not yet reviewed.

05 · In its own words

What it sounded like

Openings of real pages, quoted word for word, oldest first. Each links to the page, or to the Internet Archive’s copy when it has gone.

2016 · BlogArchived copy
“I've always loved Innocent (ever since you sent my flu-bound Sister an email telling her to get better in 2001, just because I asked you to!) - and now I think you're even greater! Love what you're doing with FareShare and helping them to do amazing stuff. Maybe I'll get my Sister to email you a thank you as well!”
less waste in the world - innocent
2017May (archived by) · BlogArchived copy
“When we went up to the roof the other day, we found a secret message next to a ladder on the fifth floor.”
secret messages - innocent

06 · Every page

All 58 pages, by year

The full list behind every number above, with links to the live page or the archived copy. 54 count towards the measures; the rest were not measured (2 pdf report (annual report, info pack), 1 under 150 words of text, 1 mostly short lines (lists, tables, link pages) rather than prose). Pages with no reliable date count in the overall numbers but in no year.

201811 pages
201713 pages
201614 pages
201516 pages
Not measured4 pages

07 · Method

How this was measured

The pages

innocent's own blog posts, press releases and announcements, fetched from the live site in 2026 and from the Internet Archive for 2015 to November 2022. Homepages are shown as captures; product and help pages are not measured.

The measures

Code counts from ScriptGrain’s writing instruments: sentence length and rhythm, contractions, pronouns, punctuation, long words, and three published lists of AI-associated words. No AI model reads or judges any page.

What it cannot say

These measures describe style. They cannot tell who or what wrote a page, and nothing here says or implies that any brand used AI to write.

Sample sizes

Every number carries its page count. Years under 10 pages, and before-and-after comparisons with under 10 pages on a side, are shown as context only.

Sentence length
Average words per sentence.
Sentence rhythm
How much sentence length varies (standard deviation over mean). Higher reads less even.
Contractions
Contractions (we're, you'll, it's) per 1,000 words. A rough gauge of how conversational the writing is.
"You"
Share of I/we/you pronouns that are "you": how much the writing addresses the reader.
"We"
Share of I/we/you pronouns that are "we": how much the writing talks about the organisation.
Questions
Question marks per 1,000 words.
Exclamations
Exclamation marks per 1,000 words.
Long words
Share of words with seven or more letters.
Em dashes
Em dashes per 1,000 words.
AI-associated words
Words from SGR-007's three AI-associated lists, per 1,000 words. These are style markers that people used long before ChatGPT; a higher rate says nothing about who or what wrote the page.
AI-era marketing words
Marketing words that became more common in AI-era copy (seamless, elevate, unlock...), per 1,000 words. A style measure only.
Quieter LLM-favoured words
Everyday words language models are known to over-use, per 1,000 words. A style measure only.
AI-associated sentence shapes
Sentence patterns such as "not just X, but Y", found by ScriptGrain's cliche checker, per 1,000 words. A style measure only.
Page length
Average words per page.

Studies: SGR-009: Do brands in the same sector sound the same? · SGR-010: How UK brands write · SGR-011: Did brand writing change after ChatGPT?.