Direct answer: the 13 product entry pages from study 104 hold 1,179,680 bytes (1.18 MB) of HTML, with a median of 96,035 bytes. The median visible-text share is only 4.4% (2.0%–10.3%), and inline SVG is 6.2% median (present on 7 of 13 pages). Script is 5.7% median, style 0% (max 25.9%), markup 48.6%, whitespace 17.2% and comments 1.2%.
This is a third HTML profile after article pages and homepages: text is as scarce as on article pages (4.4% vs 4.0%), but inline SVG is far below article pages (6.2% vs 30.8%); script/style sit between the two, while markup is the highest (48.6%). Product pages cannot inherit either page type's optimisation playbook.
Method: the same byte splitter
The sample is the 13 product entry-page caches from study 104 (captured 2026-11-01; P04/P14 have no product link). The study-89 method applies: remove comments, script, style and inline SVG, then measure the remaining tags and text nodes; UTF-8 bytes, percentages over that page's HTML bytes. No network and no pixel decoding.
| Bucket | Contains | Boundary |
|---|---|---|
| Visible text | Text after removing script/style/SVG/comments/tags | Includes navigation and footer, not body copy |
| Inline SVG | The whole <svg>…</svg> block | Not <img src=…svg> image files |
| Script/style | Whole <script> / <style> blocks | Includes JSON-LD and external tags |
| Markup | Other tags, including class/style/data attributes | Attribute overhead in class_attr_bytes |
Result: text 4.4%, SVG 6.2%, markup 48.6%
| Bucket | Median share | Min–max |
|---|---|---|
| Visible text | 4.4% | 2.0%–10.3% |
| Inline SVG | 6.2% | 0%–41.3% |
| Markup | 48.6% | 25.6%–71.9% |
| Whitespace/other | 17.2% | 0.2%–53.9% |
| Script | 5.7% | 2.5%–22.1% |
| Style | 0.0% | 0%–25.9% |
| Comment | 1.2% | 0.2%–6.5% |
Ten of 13 pages are below 6% text, with P11 at 2.0% and P02 at 2.5%; P15 is highest at 10.3%. Product entry pages are mostly list/template pages, so thin text is expected; but thin text does not mean small HTML: the 13 pages total 1.18 MB and the largest is 236.7 KB.
Three page profiles
| Metric (median) | 13 product pages | 14 homepages | 36 article pages |
|---|---|---|---|
| HTML bytes | 96,035 | 54,666 | 106,260 |
| Visible text | 4.4% | 6.8% | 4.0% |
| Inline SVG | 6.2% | 0.0% | 30.8% |
| Script | 5.7% | 10.7% | 4.1% |
| Style | 0.0% | 0.3% | 0.0% |
| Markup | 48.6% | 47.3% | 39.2% |
| Whitespace/other | 17.2% | 15.0% | 16.5% |
Each page type has a different centre of gravity: article pages are SVG-heavy; homepages lean on script/style with almost no SVG; product pages have little text, moderate SVG and the highest markup. Applying one page type's conclusion to another will mislead.
Page-by-page (13 pages)
| Page | HTML KB | Text | Inline SVG | Script | Style | Markup | Whitespace |
|---|---|---|---|---|---|---|---|
| P01 | 7.8 | 9.7% | 0.0% | 5.5% | 0.0% | 71.9% | 10.4% |
| P02 | 231.2 | 2.5% | 10.3% | 5.2% | 0.0% | 25.6% | 53.9% |
| P03 | 159.2 | 2.6% | 38.4% | 3.2% | 0.6% | 34.7% | 19.5% |
| P05 | 158.0 | 3.8% | 41.3% | 2.5% | 0.0% | 33.2% | 18.4% |
| P06 | 99.5 | 6.2% | 6.2% | 18.4% | 0.0% | 32.5% | 30.1% |
| P07 | 93.8 | 2.6% | 26.5% | 4.0% | 0.0% | 48.6% | 17.2% |
| P08 | 29.4 | 9.0% | 0.0% | 22.1% | 1.3% | 46.2% | 15.5% |
| P09 | 41.1 | 4.4% | 0.0% | 4.2% | 11.5% | 54.0% | 21.2% |
| P10 | 14.6 | 5.3% | 0.0% | 14.8% | 0.0% | 69.8% | 9.9% |
| P11 | 94.9 | 2.0% | 26.3% | 5.7% | 0.0% | 48.5% | 16.4% |
| P12 | 13.0 | 9.6% | 0.0% | 14.9% | 4.6% | 62.7% | 5.2% |
| P13 | 170.2 | 4.2% | 13.1% | 6.5% | 25.9% | 49.7% | 0.2% |
| P15 | 39.4 | 10.3% | 0.0% | 13.0% | 1.6% | 51.5% | 22.4% |
Three tails: P02 is 231 KB with 53.9% whitespace/other and 10.3% inline SVG; P03 and P05 have 38.4%/41.3% inline SVG; P13 is 170 KB with 25.9% style. "Large" comes from completely different buckets on different pages.
Inline SVG is not an SVG image file
Study 105 counted <img src=…svg> image files (234); this study counts inline <svg> blocks in the HTML. Both are SVG but differ in delivery: an image file can be cached separately, while an inline block is repeated in every page. Product pages contain both, so acceptance should count them separately.
What this means for acceptance
- Measure product pages separately. Text 4.4%, SVG 6.2%, markup 48.6% differ from both article pages and homepages.
- Look at markup and whitespace first. Markup is 48.6% median and whitespace 17.2%; template structure may outweigh content.
- Report inline SVG separately. P03/P05 are 38–41% SVG; if repeated icons, consider a sprite or external file.
- Check the style-block tail. P13 has 25.9% style; inlining template CSS makes every page larger.
- Thin text is not a quality score. Product list pages are naturally thin; read them with product details, specs and images.
Limits
- 13-page pilot: the same product entry pages as study 104; not all product pages on the sites and not an industry estimate.
- Raw HTML only: no compressed transfer, images, external CSS/JS, fonts or HTTP headers.
- Text includes template: navigation, footer and product names are included, not body copy.
- Inline SVG purpose unknown: icons, illustrations or decoration; no deduplication.
- Single snapshot: cached 2026-11-01; no redesign recheck.
Data and reproduction
Anonymised data: equipment-product-entry-html-weight-2026.csv (13 rows × 23 columns: HTML bytes, text, script/style/SVG/comment/markup/whitespace bytes and shares, tag counts, class-attribute bytes). In-repo copy: docs/research/2026-11-product-entry-html-weight.csv.
Parser: scripts/article_html_weight_probe.py (the same splitter as study 89); tests: scripts/test_article_html_weight_probe.py (9 tests). Run:
python -m unittest discover -s scripts -p "test_article_html_weight_probe.py"
Related: the product entry-page AI-readability pilot, the product entry image-delivery pilot, and the article HTML weight composition pilot.
Sources and definitions
- Sample: 13 product entry-page caches from study 104 (captured 2026-11-01; P04/P14 have no product link).
- Split: remove comments/script/style/inline SVG, then measure tags and text; UTF-8 bytes.
- Measured 2026-11-07 with no network requests; site and product URLs are not published.
Related: the product entry HTML-compression pilot.