Building an Asia AI Can See: Japan National AI Profile

Japan has given artificial intelligence a statutory home, a prime-minister-led command structure and a national plan. The five English-language homepages that carry that story to the world score 52.4/100 on average (Grade C) in an AI Visibility Inspector audit, and the average hides a 21-point gap between the best page and the weakest.

The audit covers five institutions across five government domains: the Digital Agency, the Prime Minister’s Office, the Ministry of Economy, Trade and Industry (METI), the Ministry of Internal Affairs and Communications (MIC) and the Cabinet Office. Together they sit at the centre of the AI Promotion Act, the AI Basic Plan, the AI Guidelines for Business and Japan’s government-wide generative AI rollout. Their pages range from 44 to 65. Two reach Grade C. Three do not.

In This Article

• National average: 52.4, Grade C. That is 3.8 points above Estonia’s 48.6 in the previous series, but the shape is very different. Estonia’s five pages sat in a one-point band. Japan’s span 21 points.
• Two Grade C pages, three Grade D. The Digital Agency leads at 65. The Prime Minister’s Office follows at 57. METI (49), MIC (47) and the Cabinet Office (44) fall below the line.
• Structure is near-perfect. Identity is not. Structural integrity averages 99 across the five pages. Entity clarity averages 44 and schema and metadata average 15.
• No machine-readable identity on any page. The Inspector detects no JSON-LD on all five homepages, and entity connectivity and knowledge-graph anchoring both score 0/100 on every one.
• Freshness separates the leaders from the rest. The Digital Agency (57) and the Prime Minister’s Office (49) carry recognisable date signals. The Cabinet Office (4), METI (8) and MIC (0) do not, and all three are flagged for structural decay.
• AI is hard to find on the pages that govern AI. The Inspector lists artificial intelligence as a missing topic on the Cabinet Office, Prime Minister’s Office and Digital Agency homepages.

Why Japan, and why now

Japan’s AI framework is no longer a set of advisory papers. The AI Promotion Act came fully into force in September 2025 and established an AI Strategy Headquarters under the Prime Minister. In December 2025 the government adopted its first AI Basic Plan, and in the months since, the Headquarters has continued to meet, with a fifth session held on July 10, 2026 under Prime Minister Sanae Takaichi. The Digital Agency’s generative AI platform for civil servants, Gennai, was slated to reach more than 100,000 government officials from May 2026, and METI and MIC jointly publish the AI Guidelines for Business.

The political calendar has also moved. Prime Minister Takaichi carried out her first cabinet reshuffle on September 17, 2026, and Toshiharu Furukawa took over as Minister for Digital Transformation. Ministers, mandates and policy documents are changing in real time, and these are exactly the facts that AI systems are asked about when someone searches for Japan’s AI strategy.

The English-language homepages matter more than their traffic suggests. They are the first pages international investors, researchers, journalists and, increasingly, retrieval engines consult to understand who does what in Japanese AI policy. This profile looks at how those pages read to the machines.

The profile in one view

The table below shows the five audited institutions (rows) and the five dimension scores that determine the overall index. Scores are from the AI Visibility Inspector (v1.9.3), audited on 5 October 2026.

Shading convention (for your CMS): 75 and above strong, 50 to 74 moderate, 20 to 49 weak, below 20 absent.

The pattern is consistent. Where Estonia’s pages were almost identical, Japan’s are differentiated, and the differentiation comes almost entirely from three columns: extractability, schema and metadata, and freshness. Structure is solved everywhere. Identity is solved nowhere.

How the five engines read the Japanese pages

The Inspector reports compatibility scores for five engines: Perplexity, OpenAI/ChatGPT, Claude, Google Gemini and Microsoft Copilot. The table below shows the five institutions (rows) and the five engine scores (columns).

Engines that reward readable prose and clean hierarchy (Claude at 74.2, ChatGPT at 69.0) read Japan’s pages comfortably. Engines that lean on structured data, dates and authorship (Gemini at 36.6, Perplexity at 45.2) are far less generous. The Digital Agency is the only page where Gemini clears 50, and it is also where Perplexity reaches 64, the highest in the set. Both results track the same factor: it is the one homepage whose recency is legible to a machine.

For comparison, the Estonian pages averaged 17.0 on Gemini and a flat 27 on Perplexity. Japan’s averages are stronger on both, but the lowest individual scores, METI at 27 on Gemini and the Cabinet Office at 33 on Perplexity, sit in the same territory.

What the five audits reveal

1. A 21-point spread inside one government

The Digital Agency scores 65 and the Cabinet Office scores 44. Both are central institutions in the same national AI framework, publishing on the same kind of site, yet an engine meets two very different pages. For anyone asking a retrieval system “who runs Japan’s AI policy?”, the answer depends heavily on which homepage the engine happens to read first.

2. Perfect structure, absent identity

Four of the five pages score a flat 100 on structural integrity and the fifth scores 95. Headings are clean, nesting is logical, and there is one H1 per page. Alongside that, schema and metadata scores range from 0 to 35, and entity connectivity and knowledge-graph anchoring are 0/100 on every page. E-E-A-T density ranges from 0 to 8.

The pages are readable, but they are not recognisable. A machine can follow the layout and still be unable to say, with confidence, which organisation it is reading, who leads it and how it relates to the others.

3. Freshness is the real dividing line

The two Grade C pages are also the two with the strongest freshness scores, 57 for the Digital Agency and 49 for the Prime Minister’s Office. The three Grade D pages score 8, 4 and 0. On those three, the Inspector reports that no date signal could be found and that content age is unverifiable, and it flags structural decay.

This matters because the pages are not stale. The Digital Agency homepage carries items such as the September 18 inaugural press conference of the new minister and the September 24 ministerial handover. The Prime Minister’s Office page references the reshuffled Takaichi cabinet. Recency exists in the content. Whether it exists for the engines is a different question, and on three of five pages the answer is no.

4. The AI story is not on the front page

Japan’s AI architecture runs through the Cabinet, the Digital Agency, METI and MIC. Yet the Inspector’s gap analysis lists artificial intelligence as missing on the Cabinet Office, Prime Minister’s Office and Digital Agency homepages. Only METI and MIC register AI as a detected concept, and even there it is one concept among many on a broad ministry page.

Homepages are not policy pages, and the detail may well live deeper in each site. But engines frequently form their first impression from the front door, and the front doors of three key institutions say little about AI.

5. What engines think the pages are about

The Inspector generates candidate queries from what each page exposes. The results are instructive. The Cabinet Office page produces queries about its corporate number and copyright line. The Prime Minister’s Office page produces “skip to main content” prompts. METI’s page surfaces six bare dates as separate question candidates. MIC’s English page produces queries built from Japanese-language navigation labels and a street address.

These are heuristic estimates, not predictions of real user behaviour. They do show where the machine’s attention falls on a page, and it often falls on interface elements rather than on policy.

6. Entity stability is highest where scores are lowest

The entity graph stability score runs from 49 (Digital Agency, fragmented) to 73 (METI and MIC, moderate). The best-scoring page overall has the least stable entity graph, while the Cabinet Office, METI and MIC rate higher on stability despite scoring lower overall. A strong page can have a weak identity, and a weak page can have a coherent one. Overall scores alone do not tell the full story.

Institution profiles

digital.go.jp/en (Digital Agency) – Grade C – 65/100

The highest-scoring page in the set, and the only one with a strong freshness signal

Strengths: Structural integrity of 95 and data extractability of 85, the highest extractability in the audit. Freshness reaches 57, the only page above 50. Schema and metadata scores 35, ahead of every other page despite no JSON-LD. Engine scores are the best in the set, with Perplexity at 64, ChatGPT at 76, Claude at 78, Gemini at 51 and Copilot at 58.

What AI sees: A page about the Digital Agency with recent, concrete content: the new minister’s inaugural press conference, the ministerial handover, My Number Card services, the 2025 activity report. Of its 25 candidate queries, the strongest reach 54% eligibility. Dates are detected through page text and meta tags only.

What stays unseen: Entity graph stability is 49/100, the lowest of the five and rated fragmented. Only the Agency, a press room and a generic product entity are detected. Artificial intelligence is flagged as a missing topic, despite the Agency’s role in Japan’s government-wide generative AI platform. E-E-A-T density is 4/100, with two trust signals and no expertise or authority signals.

japan.kantei.go.jp (Prime Minister’s Office) – Grade C – 57/100

A clean, recent page that leaves the Prime Minister’s role unexpressed

Strengths: Perfect structural integrity (100) and a freshness score of 49, the second highest. Claude scores 73 and ChatGPT 67. The page is correctly identified as the Prime Minister’s Office of Japan, and its leading candidate query reaches 55% eligibility.

What AI sees: The institution itself, a generic organisation entity, and interface concepts such as “Important Notice” and “Main Content”. Entity graph stability is 51/100, rated fragmented. The page references the reshuffled Takaichi cabinet, but the Inspector reads it through text and meta signals rather than structured data.

What stays unseen: The Prime Minister chairs the AI Strategy Headquarters, yet the homepage registers no AI concept. Thirty images lack alt text, leaving coverage at 17%, the lowest in the set, and Gemini sits at 44. The meta description is missing or too short. E-E-A-T density is 4/100 and semantic richness 39/100.

meti.go.jp/english (Ministry of Economy, Trade and Industry) – Grade D – 49/100

Strong extractability and the broadest concept coverage, with the lowest schema score in the set

Strengths: Perfect structural integrity (100) and extractability of 80, the second highest. Entity graph stability is 73/100, rated moderate. METI is one of only two pages where artificial intelligence registers as a detected concept, alongside policy index, white papers, news releases and data reports. ChatGPT scores 74 and Claude 77.

What AI sees: A broad policy ministry with a well-organised content inventory. Several detected items are incidental, including a “conversion rate” metric, and six bare dates are read as separate question candidates.

What stays unseen: Schema and metadata score 0/100, the only zero in that dimension. Freshness is 8/100 and the page is flagged for structural decay because no machine-readable date was found. The meta description and canonical URL are missing, and 19 images lack alt text (49% coverage). Gemini scores 27, the lowest in the audit.

soumu.go.jp/english (Ministry of Internal Affairs and Communications) – Grade D – 47/100

The highest single entity clarity score in the audit, and no date signal at all

Strengths: Perfect structural integrity (100) and entity graph stability of 73/100. “White Papers” is detected as a named entity at 80% clarity, the highest individual clarity figure in the set. AI is detected as a concept, alongside the Radio Use Portal and the Basic Resident Registration System. Claude scores 74.

What AI sees: A large ministry with many internal organisations, external agencies and services. Candidate queries include Japanese-language navigation labels on the English page, along with a corporate number and street address string at 54% eligibility.

What stays unseen: Freshness is 0/100 and the page is flagged for structural decay. Schema and metadata score 10/100, the meta description and canonical URL are missing, and alt-text coverage is 74%. Content depth is moderate at 316 words. MIC co-publishes the AI Guidelines for Business with METI, but nothing on the homepage connects it to that role in a way the Inspector can read.

cao.go.jp/index-e.html (Cabinet Office) – Grade D – 44/100

The lowest score in the set, on the page of the institution that publishes the AI Basic Plan

Strengths: Perfect structural integrity (100) and a moderate entity graph stability of 70/100. Primary and secondary entity signals each score 20/20. The page is correctly identified as the Cabinet Office, with state ministers, councils and white papers detected as related concepts.

What AI sees: A formal institutional page with 310 words of content, a corporate number and a copyright line. All six candidate queries sit at 43% eligibility and none concerns policy. Extractability is the lowest in the set at 41, and semantic richness is rated thin at 39/100.

What stays unseen: Zero expertise, experience, authority and trust signals were detected, giving E-E-A-T density of 0/100. Freshness is 4/100 and the page is flagged for structural decay. Ten images lack alt text (33% coverage), the meta description and canonical URL are missing, and artificial intelligence is listed as a missing topic. Perplexity scores 33 and Gemini 31.

What this means

Japan’s AI policy is clearly articulated on paper. It has a statute, a headquarters, a basic plan, joint ministry guidelines and a government-wide platform. The audits show that the front pages of the institutions responsible read to AI systems as well-built documents about organisations that the machines cannot yet fully identify or date.

That is a different problem from the one Estonia faces, and it has a different cost. A single high-scoring page (the Digital Agency) shows that Japanese government sites can read well to machines. The spread between that page and the weakest one is the measure of what is still uneven.

The timing is notable. A new cabinet lineup, a revised basic plan under discussion and an extraordinary Diet session starting in October all mean that the facts engines will be asked about are being rewritten right now. Which version of those facts reaches the answer depends on what the machines can see.

Scope and method

Each page was audited once on 5 October 2026 with the AI Visibility Inspector (v1.9.3) and the Srna SEO Framework. This is a snapshot of five page audits across five domains, not a full-site audit. All five are English-language homepages: cao.go.jp/index-e.html, digital.go.jp/en, japan.kantei.go.jp, meti.go.jp/english and soumu.go.jp/english. They were chosen because they belong to institutions directly involved in Japan’s AI policy and implementation.

Entity clarity scores of 40 to 45 are largely a feature of the scoring model, since the Inspector reports no JSON-LD on any page. This profile treats schema as absent and reads that credit as coming from page-level metadata and entity tokenisation. Query categories and engine citation percentages are the tool’s estimates and are heuristic. Facts about Japan’s AI framework and political calendar come from public sources checked on 7 October 2026. The Inspector measures what a page exposes, not what an institution does.

The national average of 52.4 is the arithmetic mean of the five overall index scores: 65, 57, 49, 47 and 44.

How does your institution read to AI?

The scores above describe what a page exposes. They do not explain why a given page lands where it does, how its signals interact, or what the right order of work would be. That analysis sits behind the numbers, and it is specific to each institution.

If you work for a Japanese ministry, agency, enterprise or any organisation whose public pages are meant to be understood, cited and trusted by AI systems, I would be glad to talk. I can clarify any of the findings in this profile, walk you through how your own pages are read, and run an audit tailored to your institution.

This research series continues from Building an Asia AI Can See, which includes National AI Profiles of Japan.

You can also explore the AI Search Readiness Audit, the Knowledge Exposure Audit and the AI Visibility Inspector.

Calculate your potential exposure below