We checked 114 major Japanese ecommerce sites. 96% run GA4 — and 46% still carry a dead tag — What's really installed: tracking tags, consent banners, and load times

“Analytics is installed, but nobody looks at it.” After writing that in the first post of this series, I wanted to find out how common it really is. You can’t see inside people’s heads from the outside — but you can see what’s installed on a site. So I opened 114 major ecommerce sites, one by one, in a real browser and checked. All of them are Japanese stores, which matters for the consent-banner numbers in particular.

The short version:

• 96% have GA4 or Google Tag Manager installed (110/114). Analytics is “in nearly every store”
• 46% (52/114) still carry a tag or ID from the old Universal Analytics (UA), which was discontinued two years ago
• Marketing tags per site: median 7. 76% have five or more, and 71% run three or more kinds of ad pixel
• Only 25% show a cookie consent banner (28/114)
• Page load completes at a median of 3.9 seconds. 20% finish under 2 seconds; 36% take 5 seconds or more
• The number of external domains tracks load time. Sites with fewer than 20 external domains have a median of 1.9 seconds; sites with 40 or more, 5.4 seconds

How I did it (and the limits, up front)

• Sample: Brand-owned ecommerce sites (marketplaces excluded) across 14 categories: Food & beverage 21, Fashion 17, Cosmetics & beauty 12, Furniture & interior 11, Electronics 10, Sports & outdoor 8, Household & stationery 8, Eyewear & shoes 6, Baby & kids 5, Books & hobby 5, Pet 4, Health & supplements 3, Jewelry 2, Flowers & gifts 2. I picked the official stores of well-known brands, so this is not a random sample. It skews toward larger players
• Method: On September 5, 2026, using desktop Chrome on an ordinary consumer connection in Japan, I opened each site’s home page once, waited five seconds after load, and mechanically read tracking tags, platform, consent banner, load time, and transfer size from the page source and the resources it loaded. I attempted 124 sites, dropped 10 for load failures and similar problems, and tallied 114
• Limits: Home pages only (no product or checkout pages). One desktop measurement per site, so this is not comparable to a mobile Lighthouse score. Tag detection only counts what showed up in the source and the network — tags that fire conditionally were not counted (so the real number is probably higher). Consent-banner detection means “a fixed-position element containing the words cookie and consent” or a known consent-management tool, so some may have been missed
• No store names. The point of this survey is to see the pattern, not to grade individual stores. Only category-level tallies are published

1. Analytics is in nearly every store

I could directly detect a GA4 measurement ID or tag on 107 sites (94%), Google Tag Manager on 102 (89%), and one or the other on 110 (96%). The remaining four may well have it installed in a form that’s harder to detect.

In other words, the “should we install analytics?” stage is long over for major Japanese ecommerce. The question isn’t whether it’s installed — it’s whether anything gets decided because of it. That part isn’t visible from the outside. But the next number is a clue.

2. 46% still carry discontinued UA

The old Universal Analytics (UA) stopped collecting data in July 2023, and the data itself was deleted in July 2024. More than two years later, 52 home pages (46%) still contain a UA-XXXXXXX-X style ID or a load of the old analytics.js.

Looking at a few of these sources by hand, I found the old ga('create', 'UA-XXXXXXXX-1') code sitting right next to the GA4 tag, untouched. These tags do nothing now. There’s nowhere for the data to go. The reason they’re still there is probably simple: nobody is looking. Someone who opens the analytics screen every week notices the old tag during a GA4 migration and removes it. In a store where analytics is “installed and forgotten,” that never happens.

I think this is one of the few signals of “installed and forgotten” you can observe from outside. By category, it was left on 12 of 21 Food & beverage sites, 5 of 8 Household & stationery sites, and 5 of 10 Electronics sites.

3. Median 7 tags. 71% run three or more ad pixels

Here’s the breakdown of detected tags (out of 114).

Tag	Sites	Share
GA4	107	94%
Google Tag Manager	102	89%
Google Ads	89	78%
Meta Pixel	89	78%
Yahoo! JAPAN Ads	80	70%
Microsoft Clarity (session recording)	61	54%
LINE Tag	58	51%
Old UA (discontinued)	52	46%
Criteo	49	43%
TikTok	43	38%
KARTE	15	13%
Ptengine	10	9%

Tags per site: median 7. 87 sites (76%) have five or more, and 13 have ten or more. 81 sites (71%) run three or more kinds of ad pixel, and 72 (63%) run a session-recording or on-site engagement tool like Clarity or KARTE.

None of this is bad in itself. If you run ads, you need each platform’s pixel. But put it next to the following two numbers and the meaning changes.

4. Only 25% show a cookie consent banner

I detected a fixed-position consent banner or a known consent-management tool (OneTrust and the like) on 28 sites (25%). Three quarters of the sites run those seven tags with no banner at all.

This isn’t about breaking the law. As I laid out in another post, Japanese ecommerce stores selling physical goods are treated as outside the scope of the external-transmission rules, and whether consent is needed depends on how the ad integrations are used. The point here is the fact that the mainstream in Japanese ecommerce is “no banner” — and the structural consequence: if a banner ever becomes necessary, the denominator behind all seven tags starts shrinking at once. My position is that cookieless measurement protects that denominator, but that’s a product matter, so I’ll leave it for the end.

By category, banners were shown on 4 of 10 Electronics sites, 3 of 6 Eyewear & shoes sites, and 3 of 8 Sports & outdoor sites; Household & stationery was 0 of 8. Categories with international brands or overseas operations seem more likely to show one, but the counts are small, so treat that as a tendency, not a finding.

5. Median load time 3.9 seconds. More external domains, slower page

On an ordinary desktop connection, time to page load completion (the load event) was a median of 3.9 seconds.

Time to load complete	Sites	Share
Under 2 s	23	20%
2–3 s	21	19%
3–5 s	28	25%
5–10 s	34	30%
10 s or more	7	6%

Transfer size was a median of 2.8MB, with 18 sites (16%) exceeding 10MB. The number of external domains loaded (hosts other than the site’s own) was a median of 32, with a maximum of 82.

And the number of external domains clearly tracks load time.

External domains	Sites	Median load complete
Under 20	33	1.9 s
20–39	37	3.5 s
40 or more	43	5.4 s

The correlation coefficient between tag count and load time was 0.54. Of course, sites with many tags tend to be large players with more images and more features, so tags aren’t the only cause. Still, the direction — “every tag you add opens more connections to external domains and stretches the load” — is consistent with these numbers.

A 2020 study by Google and Deloitte found that a 0.1-second improvement in mobile page speed raised retail conversion rates by 8.4%. This survey measured desktop, so the numbers aren’t directly comparable, but given that a third of sites take 5 seconds or more, it’s reasonable to assume mobile is even worse.

Median load time by category: Health & supplements 0.9 s (3 sites), Food & beverage 2.4 s, Electronics 2.9 s, Cosmetics & beauty 3.5 s, Furniture & interior 3.7 s, Fashion 4.3 s, Sports & outdoor 4.8 s, Books & hobby 6.7 s, Eyewear & shoes 7.6 s. Fashion and Sports & outdoor have the highest median external-domain counts, at 47–49, and sit on the slow side of load times.

6. Takeaway: “installed and forgotten” is visible from the outside

Three things stand out from looking at 114 major ecommerce sites.

1. Analytics is in every store. Almost no store has “not installed” as its problem anymore
2. Nearly half the stores have a tag that died two years ago sitting there untouched. “Installed and forgotten” really is common
3. Tags keep piling up, and once external domains pass 40, load times land in the 5-second range. Nobody is doing an inventory

To check your own store, there are two things to do. Open the Network tab in your browser’s developer tools, load your home page, and count how many domains other than your own appear. Then search the page source for UA-. If it’s still there, it tells you roughly when someone last reviewed the analytics setup.

What this survey was built on

This survey ran the same detection logic (identifying tracking tags and platforms) as the free diagnosis at NextRise Analytics, in a browser. Enter your store’s URL into the free diagnosis and, alongside the page-speed score, you get the same tracking-tag detection results — no sign-up required.

And the NextRise tracking tag is about 5KB, one external domain, and no cookies — built so that adding it to the table above doesn’t add to your load time or your consent denominator. It sits next to GA4, so there’s no need to remove anything you already have.

Here is the data the tallies came from. 114 rows, one row per site, with the category and each metric — and no store names or URLs.

▶ Download the CSV (114 rows, 6.8KB)

The columns are genre, platform_detected, ga4, gtm, ua_remnant, ad_pixels, behavior_tools, tags_total, consent_banner, load_ms, transfer_kb, js_kb, third_party_hosts, scripts. Every number in this post comes out of that file, so you are welcome to recount them yourself.

---

Survey details: September 5, 2026 / 124 Japanese ecommerce sites across 14 categories attempted, 114 tallied / desktop Chrome, ordinary consumer connection in Japan, one visit to the home page / detection by string matching against page source and loaded resources (GA4, GTM, UA, ad pixels, session-recording tools, consent-management tools) / load time is Navigation Timing loadEventEnd / transfer size is the sum of Resource Timing transferSize. To avoid cache effects from repeat visits in the same browser, each site was in principle visited only once (a few sites had been opened earlier during preliminary checks).