Direct answer: across the seven sites with at least two cached pages — 33 article pages — visible text segments of 8+ characters total 68,839 characters. Of these, 24,573 characters (35.7%) also appear on another page of the same site; per page, the median repeat share is 33.6% (3.6%–72.6%), leaving 64.3% unique to the page.
In other words: study 89 found that visible text is only 4.6% of HTML; about one third of that text is repeated long segments within the same site. The repetition may come from navigation, footer, contact details or template copy; this study does not separate DOM positions.
Method: exact cross-page matching within a site
We reused the 36 cached article pages from study 83 (10 sites, fetched 2026-11-03). Only sites with at least two cached pages are included: 7 sites, 33 pages; E19, E26 and E28 have one page each and are excluded. After removing script/style/noscript/comments, tags and extra whitespace, text is split at 。！？!?；;; only segments of 8+ characters are kept; a segment appearing more than once on the same page counts once.
| Item | Rule | Boundary |
|---|---|---|
| Text segment | Visible text between sentence delimiters, 8+ characters | Short words/buttons excluded |
| Within-site repeat | Same string on at least two pages of the site | Exact match, no fuzzy or synonym matching |
| Repeat characters | Characters in repeated segments on that page | Repeats on one page count once |
| Unique characters | Visible characters minus repeat characters | Not a measure of original copy |
Result: 35.7% repeated, median page 33.6%
| Measure | Value | Note |
|---|---|---|
| Pages included | 33 | 7 multi-page sites |
| Visible segment characters | 68,839 | 8+ character segments |
| Repeated within site | 24,573 (35.7%) | On at least two pages |
| Unique characters | 44,266 (64.3%) | Only on that page |
| Median page repeat | 33.6% | Range 3.6%–72.6% |
| Page repeat share | Pages | Of 33 |
|---|---|---|
| Below 20% | 8 | 24.2% |
| 20%–40% | 10 | 30.3% |
| 40%–60% | 11 | 33.3% |
| Above 60% | 4 | 12.1% |
By site: from 4.8% to 51.5%
| Site | Pages | Visible chars | Repeated chars | Repeat share |
|---|---|---|---|---|
| E03 | 5 | 16,576 | 8,539 | 51.5% |
| E06 | 5 | 9,727 | 2,965 | 30.5% |
| E09 | 5 | 9,324 | 4,641 | 49.8% |
| E14 | 5 | 10,340 | 1,626 | 15.7% |
| E16 | 3 | 6,205 | 3,123 | 50.3% |
| E22 | 5 | 8,903 | 430 | 4.8% |
| E23 | 5 | 7,764 | 3,249 | 41.8% |
E22 has only 4.8% repeated long text and E14 15.7%; E03, E16 and E09 are around 50%. The site differences are more informative than one industry-style average.
Text repetition and SVG repetition do not move together
Pairing this study with the SVG byte repetition from study 91: E22 has 4.8% text repetition but 83.6% SVG byte repetition; E23 has 41.8% text repetition but 19.7% SVG byte repetition; E16 has 50.3% text repetition and no inline SVG. The two measures describe different layers and cannot replace each other.
| Site | Text repeat | SVG byte repeat (study 91) |
|---|---|---|
| E03 | 51.5% | 82.3% |
| E06 | 30.5% | 94.1% |
| E09 | 49.8% | 67.5% |
| E14 | 15.7% | 74.5% |
| E16 | 50.3% | 0.0% |
| E22 | 4.8% | 83.6% |
| E23 | 41.8% | 19.7% |
What this means for acceptance
- Split visible text into repeated and unique. Reporting only total characters overstates the information on each page; list the repeat share separately.
- Separate DOM regions first. This study does not distinguish navigation/footer/body; acceptance should split template text from article copy before judging repetition.
- Check text and SVG separately. The two layers do not track; fixing one does not fix the other.
- Exact matching is a conservative floor. Paraphrase, translation or small edits are not counted; real repetition may be higher, but this study does not extrapolate.
- Repetition is not plagiarism or a quality score. Navigation, footer and contact details repeat by design; this study only quantifies identical long text across pages of one site.
Limits
- Template text included: no article-body extraction; navigation, footer and contact details raise the repeat share.
- Exact matching: no fuzzy, synonym or translation matching; small edits are not counted.
- 8+ characters: short words and button labels are excluded from numerator and denominator.
- Site scope: only the 7 sites with at least two cached pages (33 pages); single-page sites are excluded.
- Cached snapshot: fetched 2026-11-03; no redesign recheck.
- Pilot: not an industry estimate; sites anonymised; no segment text is published.
Data and reproduction
Anonymised data: equipment-article-visible-text-duplication-2026.csv (33 rows × 7 columns: site, page, visible chars, segments, repeated chars, repeat share, unique chars). In-repo copy: docs/research/2026-11-article-visible-text-duplication.csv. No segment text is published.
Probe: scripts/article_text_duplication_probe.py; tests: scripts/test_article_text_duplication_probe.py (no network, 8 tests: splitting, short-segment filtering, script/style removal, within-site matching, same-page dedup, single-page exclusion, character conservation). Run:
python -m unittest scripts.test_article_text_duplication_probe
Related: the article HTML weight composition pilot, the article inline-SVG duplication pilot, and the article body-structure pilot.
Sources and definitions
- Page cache: study 83, 10 sites and 36 pages (2026-11-03 08:40–08:41); this study uses 7 sites and 33 pages.
- Segments: remove script/style/comments/tags, split at sentence delimiters, keep 8+ characters; exact matching.
- Measured 2026-11-06 with no network requests.
Related: the article author-signal pilot.