We ran the same check over 162 Japanese and 84 global SaaS domains, then compared accounting against accounting, HR against HR so that a difference in category mix could not masquerade as a difference between markets. In this sample the gap concentrates in llms.txt adoption; everything else is close.
The comparison below uses the 8 categories with at least 5 domains on both sides (98 Japanese, 55 global). Categories too thin on either side were left out, because a single company moves the rate too far. The full 162 and 84 are not the base of the conclusion.
Where the gap sits
llms.txt — the plain text file telling an AI which pages explain what. The Japanese side was higher (i.e. worse) in every one of the 8 compared categories.
| Category | Japanese sample (98) | Global sample (55) |
|---|---|---|
| Accounting & invoicing | 82.4% (n=17) | 50.0% (n=6) |
| HR & payroll | 78.9% (n=19) | 20.0% (n=5) |
| Communication | 84.6% (n=13) | 77.8% (n=9) |
| Payments | 58.3% (n=12) | 33.3% (n=6) |
| Marketing | 83.3% (n=12) | 33.3% (n=12) |
| E-commerce | 88.9% (n=9) | 60.0% (n=5) |
| Project management | 37.5% (n=8) | 0.0% (n=6) |
| Customer support | 62.5% (n=8) | 16.7% (n=6) |
Mean of category rates: Japanese 72.1% · global 36.4% (-35.7pt)
Everything else came out close
| Check | Japanese sample | Global sample | Difference |
|---|---|---|---|
| Blocks AI crawlers | 5.4% | 1.4% | -4.0pt |
| No structured data | 37.3% | 29.7% | -7.6pt |
| Contact details not machine-readable | 45.2% | 36.5% | -8.8pt |
| No sitemap.xml | 22.2% | 32.8% | +10.7pt |
| No llms.txt | 72.1% | 36.4% | -35.7pt |
Mean of category rates. A positive difference means the global sample failed that check more often.
- Almost nobody blocks AI crawlers in either sample. Among major SaaS, "the site is configured to keep AI out" is the exception, not the rule.
- sitemap.xml is missing less often on the Japanese side (22.2% against 32.8%). Whatever else this sample shows, it is not that Japan trails on every check.
- Structured data and contact details differ by under ten points, and the Japanese side is ahead in 3 and 3 of the 8 categories respectively — the difference is not one-directional.
Median score: Japanese 81, global 85 out of 100.
What this can and cannot say
This is a descriptive comparison of the sites we selected. It carries the real composition of each market and does not match companies on size — the global side holds more large vendors.
So "there is a gap in this sample" is supportable; "because it is Japan" is not. Nothing in this design separates market practice from company size. Doing that would need stratification or matching.
The 8 categories agreeing in direction is a clear observation, but categories are not independent trials — they share market conditions — so we do not treat that agreement as evidence of statistical significance.
And we are not claiming llms.txt works
What we measured is adoption, not effect. Whether publishing llms.txt makes an AI more likely to surface you is a separate question, and we have no evidence for it either way.
What we can say is that the difference concentrates here, in the same direction across every compared category, and that the file costs one text file to publish. A small effect and a small cost is still a decision you can make.
If you read an article claiming llms.txt gets you found by AI, ask for the evidence. We do not have it.
KanseiLink measured 162 leading Japanese SaaS domains and 84 leading global SaaS domains with one tool in September 2026 and compared the 8 categories holding at least 5 domains on each side (98 Japanese, 55 global). llms.txt was absent from 72.1% of the Japanese sample against 36.4% of the global sample, with the Japanese side higher in all 8 compared categories, while sitemap.xml was absent from 22.2% of the Japanese sample against 32.8% globally. The study is a descriptive comparison of selected sites and does not match companies on size, so it does not describe the markets as a whole (KanseiLink, measured 2026-09-03). Source: https://kansei-link.com/en/insights/jp-vs-global-ai-readability-2026-09.html
The same check that produced these figures. Enter a URL — free, no signup, 5-15 seconds.
FAQ
Does publishing llms.txt get you found by AI?
This study does not answer that. We measured how many sites publish it, not what happens when they do. We have no evidence of an effect, and neither, as far as we can tell, does anyone else. What we can say is that this is where the difference sits and that the file is cheap to publish.
Does this mean Japanese SaaS is behind?
Not from this. In this sample sitemap.xml was missing less often on the Japanese side, and structured data and contact details differed by under ten points. It is also a comparison of selected sites rather than of markets.
Why not just compare the overall rates?
The two samples have different category mixes — the Japanese side is heavier on accounting and HR, the global side on marketing — so an overall rate would report that mix as a market difference. We compare within categories and average the category rates instead.
Do you publish per-company scores?
No. This is a lightweight readability check, not the ARI rating, and attaching company names would blur the two. Only the distribution is published.
Can we cite this?
Please do. Attribute it to KanseiLink with the URL of this page. A ready-made sentence is provided above.
Method and limits
Measured 2026-09-03 using the same public API as the free check on this site, so the figures are reproducible with the same tool. The Japanese sample is the 162 domains resolved from the 200 services verified for ARI Award 2026 Summer; the global sample is the 84 business SaaS domains in KanseiLink's database with a resolvable product domain. Comparison uses the 8 categories with at least 5 domains on both sides, via category rates and their unweighted mean.
Limits: the samples are unequal (162 against 84) and per-category the global side runs n=5-12, so one company moves a rate noticeably. The two were not selected by the same rule. The global side skews toward large vendors, so company size cannot be separated from market. This is one point in time and sites change. The check measures machine readability; it is not a quality judgement and not the ARI rating.
What we do not promise: we cannot guarantee visibility or ranking inside any AI product, and we do not treat a single AI answer as market share. Principles of Independence