cd /news/ai-tools/ai-writing-patterns-across-10126-pag… · home topics ai-tools article
[ARTICLE · art-131404] src=getsitetell.com ↗ pub= topic=ai-tools verified=true sentiment=· neutral

AI writing patterns across 10,126 pages of real marketing copy

A scan of 10,126 pages of real marketing copy across 103 sites, run between 2026-08-14 and 2026-09-14, found a median site score of 90/100 but at least one page scoring under 60 on 17% of sites, according to SiteTell. The clearest pattern was that larger sites scored worse, with single-page sites averaging 95 versus 70.6 for sites over 200 pages, and 58% of flags were structural rather than vocabulary, led by redundant closing paragraphs (66% of sites) and em-dash frequency (63%). SiteTell reported that 51% of sites carried at least one word from its own blacklist, with "unlock" (35%), "landscape" (29%) and "leverage" (29%) the most common.

read5 min views2 publishedSep 16, 2026
AI writing patterns across 10,126 pages of real marketing copy
Image: source

Every figure here comes from scans people actually ran, between 2026-08-14 and 2026-09-14. One scan per site, the most recent. No site is named: these are other people's sites, crawled from a URL someone pasted in, and none of their owners signed up to appear in a write-up.

Measured . Figures are a fixed snapshot, not a live counter.

the median page is far cleaner than the mean

Most sites are fine. The tail is not. #

The median site scores 90/100, and 10 of 103 carry no flags at all. The number worth attention is at the other end: on 17% of sites, at least one page scores under 60. A visitor reads a page, not an average, which is why a report leads with the worst one.

Distribution of overall site scores out of 100, across 103 sites
Score band Sites Count
--- --- ---
0-19 1
20-39 2
40-59 7
60-79 12
80-89 25
90-100 56

Bigger sites score worse #

This was the clearest pattern in the set, and not one we went looking for. A one-page site is usually hand-written and stays that way. Past a couple of hundred pages, most of that copy is a blog nobody re-reads, and it shows: 95 average for a single page against 70.6 for sites over 200.

Mean overall site score by number of pages crawled
Site size Mean score Sites
--- --- ---
1 page 95 25
2-9 pages 84.4 18
10-49 pages 87.6 17
50-199 pages 85 22
200+ pages 70.6 21

Structural habits beat word choice #

58% of flags are structural rather than vocabulary. Swapping out buzzwords is the easy part; the habits below are what make a page read like it was generated. Ordered by how many sites carry each one, which is steadier than raw instance counts.

Structural patterns by share of the 103 sites scanned
Pattern Share of sites Instances
--- --- ---
redundant_closing_paragraph 66% 2,209
em_dash_frequency 63% 2,757
rule_of_three_pattern 63% 2,436
vocabulary_diversity 60% 2,105
punctuation_density 53% 856
listicle_heavy 51% 1,524
sentence_length_variance 41% 494
generic_opener_pattern 22% 165
dangling_ing_puffery 18% 213
nominalization_pattern 17% 61
latinate_word_ratio 12% 28
paragraph_symmetry 10% 306
glyph_led_bullets 6% 191

The words #

By share of sites. 51% of sites carry at least one word from SiteTell's own blacklist. Entries marked * are single common English words, where a share of the hits are ordinary usage rather than a tell: “auth boundaries” and “timezone boundaries” count the same as “beyond boundaries”. Marked rather than quietly dropped, because the multi-word phrases are the ones worth trusting.

Flagged words and phrases by share of the 103 sites scanned
Pattern Share of sites Instances
--- --- ---
unlock* 35% 390
landscape* 29% 1,769
leverage* 29% 705
robust* 27% 228
enhance* 25% 653
boundaries* 25% 168
seamlessly 22% 366
facilitate* 22% 99
seamless 21% 307
showcase* 19% 321
streamline* 18% 478
crucial* 17% 872
align with* 17% 208
streamlined* 17% 102
game-changer 14% 467
when it comes to 14% 232
utilize* 14% 114
dive into 14% 83
elevate* 13% 105
furthermore 10% 229
delve 10% 87
underscores* 8% 220
revolutionize 8% 219
moreover 8% 103
holistic* 6% 71

How this was measured #

  • One scan per site.
  • A site scanned eight times would otherwise count eight times and drag every frequency with it.
  • Nav and boilerplate stripped before scoring.
  • A phrase sitting in a footer on every page is not counted on every page.
  • Structural rules need length.
  • Most need 100 to 150 words before they run, so a short one-page site can only trip vocabulary rules. That is part of why small sites score higher.
  • Em-dash instance counts are not comparable over time.
  • That rule changed on 2026-09-12 from one flag per page to one per offending sentence, so read its share of sites rather than its instance count.
  • Score is 100 minus penalties.
  • A short page has less opportunity to lose points, so compare pages of similar length.

Your own site is not in here unless you scanned it. If you want the same read on your copy, it is free and takes about twenty seconds: scan your site.

── more in #ai-tools 4 stories · sorted by recency
── more on @sitetell 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/ai-writing-patterns-…] indexed:0 read:5min 2026-09-16 ·