Of 363,583 live Japanese websites, only 3,628 (1.0%) block an AI crawler — the lowest of all 43 countries. Yet Japanese news sites block AI at 23%, one of the sharpest press-vs-web gaps anywhere.
Analysis of 41,818,678 live, content-validated websites worldwide · snapshot 2026-07-13 · by Piperic
Japan runs the most open web in the entire world. Of 363,583 live Japanese sites, just 1.0% block an AI crawler — the lowest figure of all 43 countries analysed, edging out China (1.08%) and Germany (1.58%). The Japanese web is, by this measure, almost entirely open to AI.
The contradiction is the press. Japanese news sites block AI at 23.13% — more than twenty times the national average and one of the sharpest press-vs-web gaps in this report. Against a web that welcomes AI almost universally, Japanese publishers have taken a decisively defensive stance. The most-blocked non-press sectors — shopping (1.87%), tech (1.76%) — barely register.
Where Japan does block, it's slightly imprecise: 10.0% of GPTBot-blockers accidentally lost ChatGPT Search, above the global 6.1%. But with 94% of the web silent and open, Japan is the clearest case anywhere of a near-fully-exposed internet paired with a fortress press.
Share of sites blocking at least one AI crawler — lower = more open. Japan ranks #1 of 44 countries analysed.
Neighbours shown for context. Full ranking in the data download.
Every website falls into one of four groups, based on what it blocks and what it declares:
2,897 Japan websites block GPTBot to keep their content out of AI training. But 290 of them — 10.0% — ALSO block OAI-SearchBot, most likely by accident. Those sites have removed themselves from ChatGPT Search results entirely: protected from training, but invisible where their customers now search.
GPTBot (training) and OAI-SearchBot (ChatGPT Search) are independent robots.txt tokens — blocking one does not block the other.
Share of Japan sites blocking each crawler in robots.txt.
Bytespider is ByteDance's (TikTok) crawler — it ignores robots.txt more often than the majors, yet it is among the most-blocked.
AI-blocking and llms.txt adoption across Japan's biggest website categories.
| Sector | Sites | Blocks AI | llms.txt |
|---|---|---|---|
| 📰 News & media | 134 | 23.13% | 2.24% |
| business and finance | 88,838 | 0.56% | 4.2% |
| technology and computing | 25,796 | 1.76% | 4.99% |
| home and garden | 23,261 | 0.53% | 5.2% |
| style and fashion | 18,201 | 1.44% | 10.75% |
| real estate | 17,958 | 1.14% | 3.62% |
| medical health | 15,828 | 0.47% | 4.59% |
| family and relationships | 14,264 | 0.46% | 4.23% |
| food and drink | 13,032 | 1.0% | 6.79% |
| fine art | 10,953 | 0.95% | 4.6% |
| shopping | 10,649 | 1.87% | 8.53% |
| careers | 10,359 | 1.14% | 5.06% |
Japan news sites block AI 23.1× more often than the national average (23.13% vs 1.0%) — publishers are the most defensive segment of the web.
18,708 Japan online shops are open to AI crawlers (99.1% of 18,880). Their product content can already be read, compared and recommended by AI assistants — whether they know it or not.
Share of sites blocking at least one AI bot, by platform:
This is the baseline edition (2026-07-13). Piperic re-measures continuously; the next edition will show how fast Japan's web is closing — or opening.
Free, no signup — see your site the way AI crawlers do:
How to cite: “According to Piperic's AI Access Report (2026-07-13), based on 363,583 live Japan websites…” — link to this page.
Download the country dataset (JSON) Global rankings dataset
Sample: 363,583 live, content-validated websites (Japan, ccTLD-based assignment), out of 41,818,678 live domains analysed worldwide. “Live” = HTTP 200 and not parked. “Blocks AI” = the site's robots.txt disallows at least one of 14 known AI crawlers. llms.txt/ai.txt = the file exists at the site root. Snapshot: 2026-07-13 (frozen — numbers do not move with the live crawl). Full-web note: these figures cover the ENTIRE live web including the long tail; industry headlines like “25% of top sites block GPTBot” measure only large publishers — both are true, they measure different things. Country assignment via ccTLD under-counts .com/.org sites; treat US figures as indicative.
Press contact: press@piperic.com · Data: Piperic Business Intelligence
Every number on this page is reproducible from the public dataset above.