Homepage Image Delivery: 538 Images, 43% PNG, 31% JPG, 23% SVG, 38% Alt Coverage

Direct answer: the 14 homepages from study 95 contain 563 <img> tags, of which 538 have a usable src. By URL extension: PNG 232 (43.1%), JPG 169 (31.4%), SVG 122 (22.7%), GIF 9 (1.7%); WebP and AVIF are 0, and 6 have no extension. No image is external and none carries a size query parameter. Alt text is non-empty on 214 (38.0%), empty on 185 (32.9%) and missing on 164 (29.1%).

We also byte-sampled the first three images per page (37 images): 36 were fetchable and one URL was invalid; one .png was actually a JPEG. Details are in the byte-sample section below.

This differs from the article pages: among 499 images there, PNG was 403 (80.8%), external 188 (37.7%), alt non-empty 76.8%, and SVG image files 0. Homepages use more JPG and SVG image files, host everything themselves, and have lower alt coverage. Image delivery cannot be inferred from one page type to the other.

Method: same homepages, static img only

The sample is the same as studies 95/96: 14 identifiable homepage URLs from the 2024 Su–Xi–Chang first-set equipment list (16 companies, excluding E21 HTTP 502 and E20 1,329 bytes). With no network, the cached HTML is parsed for <img>; the extension is taken from the URL path, external means a different registrable domain, and alt is split into non-empty/empty/missing; width+height, loading=lazy and srcset are recorded too.

Result: PNG 43%, JPG 31%, SVG 23%

Homepage PNG is 43.1%, much lower than the article-page 80.8%; JPG (31.4% vs 10.8%) and SVG image files (22.7% vs 0) are higher. The 6 extension-less images could be served as any format and cannot be judged from the URL.

No external images, no size parameters, no lazy/srcset

All 538 images come from the homepage's own registrable domain; none carries a size query such as ?w=, and none uses loading=lazy or srcset. The 13 pages with images each have one image host (13 in total), so "external 0" does not mean "no CDN"; a same-domain CDN subdomain still counts as internal.

Width and height appear on only 17 of 563 img tags (3.0%). As with the article pages, these homepages mostly do not declare image dimensions and have no lazy loading or responsive candidates.

alt: 38.0% non-empty, 29.1% missing

Article pages had 76.8% non-empty and 18.5% missing; homepages have only 38.0% non-empty and 29.1% missing. An empty alt may mark a decorative image, but it may also simply be unwritten; this study does not judge each case.

Byte sample: 36 of 37 fetched, one format mismatch

Between 04:40 and 04:42 on 2026-11-07 we requested the first three distinct images per homepage (14 pages; E18 has no images and E24's first URL is invalid, giving 37 images). 36 returned 200 and 1 URL was invalid. On all 36, the Content-Type matched the file header (PNG 22, JPEG 9, SVG 5); but E29's first URL ends in .png, its Content-Type says image/png, and the file header is JPEG — neither the extension nor the Content-Type alone is proof of format.

Bytes: minimum 233, median 17,638, maximum 1,147,926 (about 1.15 MB); the 36 images total 2.63 MB. By actual format, the median is 19,257 bytes for PNG, 27,926 for JPEG and 1,346 for SVG; two are above 200 KB and one above 500 KB.

Compared with article pages: different page types

The two page types differ widely: article pages lean on PNG, external CDNs and higher alt coverage; homepages lean on JPG/SVG image files, self-hosting and lower alt coverage. The article-page pattern cannot be applied to homepages.

SVG image files are not inline SVG

Homepages contain 122 <img src="…svg"> image files, while the SVG in studies 89/91 was mainly inline <svg> blocks (1,698 tags, 40.7% of bytes). Both are SVG, but the delivery differs: an image file versus markup written into the HTML. Acceptance should count them separately.

What this means for acceptance

1. Sample homepages and article pages separately. PNG/JPG/SVG shares, external share and alt coverage all differ.
2. Fix alt on homepages first. 29.1% missing and 38.0% non-empty is the clearest gap in this sample.
3. Separate SVG image files from inline SVG. Same format, different delivery and caching.
4. External 0 does not mean no CDN. A subdomain or same-domain self-host is internal; inspect the actual host.
5. An extension is not the encoding. The 6 extension-less images need a real request to confirm their format.

Limits

• 14-page pilot: two of 16 identifiable homepages excluded (E21 502, E20 1,329 bytes); E18 has no images.
• Extension plus sample: the extension count covers 538 images; the byte sample covers only the first three per page (37), not the full 538, and does not judge copyright.
• alt states only: copy quality is not judged and an empty alt may be decorative.
• Single snapshot: fetched 2026-11-07; no redesign recheck.
• Page level: includes navigation, footer and decorative images; body images are not separated.
• Pilot: not an industry estimate; sites anonymised.

Data and reproduction

Anonymised data: equipment-homepage-image-delivery-2026.csv (14 rows × 19 columns: img tags, with src, external, hosts, query params, format counts, three alt states, width/height, lazy, srcset). In-repo copy: docs/research/2026-11-homepage-image-delivery.csv.

Byte sample: equipment-homepage-image-bytes-sample-2026.csv (37 rows × 11 columns: site, index, URL extension, status, Content-Type, length, bytes read, magic, format match, error, time). In-repo copy: docs/research/2026-11-homepage-image-bytes-sample.csv.

Parser: scripts/homepage_image_probe.py (merges article_image_source_probe.py and article_img_attrs_probe.py); tests: scripts/test_homepage_image_probe.py (no network, 6 tests: formats/alt, data-src, external, query params, attributes, empty page). Run:

python -m unittest scripts.test_homepage_image_probe

Related: the homepage HTML weight composition pilot, the article image-delivery pilot, and the homepage HTML compression pilot.

Sources and definitions

1. Sample: 14 homepage caches from study 95 (fetched 2026-11-07; 16 companies, excluding E21 502 and E20 1,329 bytes).
2. Images: static img src; extension from the URL path; external by registrable domain; alt in three states.
3. Measured 2026-11-07 with no network requests.