Japan's AI citation hierarchy: which sources AI engines actually trust (ITmedia, BOXIL, note...)

Here’s the bottom line first. In Japan, counting “how many times the AI cited you” is nearly meaningless on its own — because being cited on ITmedia versus a personal blog differs in business value by an order of magnitude. Yet most global GEO tools record a citation as just a URL; they don’t understand that itmedia.co.jp carries a different authority than note.com. This article maps the real landscape of Japanese AI citation sources using public data, explains why weighting citations is the core of localization, and shows how VeReach GEO’s Japanese 4-tier weighted source dictionary and its “tier1-only” view answer that problem.

---

1. The facts first: which sources AI cites most in Japan

Before the abstractions, the measured data. Per Ahrefs’ 2026 cross-platform Japanese citation ranking (reported by Web担当者Forum in April 2026), the overall picture of which domains AI engines cited most on Japanese queries looks like this:

Rank	Domain	Est. AI citations	Type
1	YouTube	~3.32M	Video UGC
2	Wikipedia (Japanese)	~0.98M	Encyclopedia
3	Google	~0.91M	Search / properties
4	Ameblo (Ameba Blog)	—	Blog UGC
5	note.com	—	UGC / publishing
6	Amazon.co.jp	—	E-commerce

What matters here isn’t the ranking itself — it’s that the “overall ranking” and “the citations that are valuable to your business” barely overlap. YouTube and Wikipedia rank high because they cut across nearly every topic, which is a different thing from where a Japanese B2B SaaS competes within its own category.

Ahrefs also showed that engines differ sharply in their habits:

Engine	Tendency in most-cited sources
ChatGPT	Ameblo #1, PR TIMES #2 (it treats press releases as authoritative), Reddit prominent
Perplexity	Favors Wikipedia JA, note.com
Copilot	Wikipedia JA, note.com, Ameblo

The fact that ChatGPT elevates PR TIMES as an authoritative source carries a strategic implication of its own: for Japanese B2B, press-release distribution isn’t only an SEO play — it’s a route into a specific engine’s citation pool. If engines “believe” different sources, your playbook has to differ by engine too.

And one finding you can’t ignore: note.com’s anomalous efficiency. Per Ahrefs, note.com captures roughly 4× the AI traffic its search-traffic share would predict. This suggests note’s article format (clean headings, answer-first, well-chunked) sits close to an “AI-optimal format” the models find easy to extract from. The first-principle of GEO — that structure itself drives citation probability — shows up here too.

---

2. The blind spot in global tools: “we see the URL, but not the authority”

This is where a structural weakness in most global GEO tools surfaces.

Every tool extracts citation URLs from AI answers. The problem is what comes next. A citation from itmedia.co.jp and one from an obscure personal blog get tallied as the same “1 count.” Tools built with an English-market mindset never encode the authority ITmedia holds in Japan’s IT industry, the role BOXIL plays in SaaS evaluation, or the context that note is UGC.

Ahrefs’ study left two other important observations: AI crawlers read HTML and do not directly read schema (structured data), and SEO rank and LLM mentions correlate at roughly 0.65. That means traditional SEO equity carries partway into AI visibility but not fully — which makes per-domain resolution (“on which domain did I earn the mention”) decisive.

Distilled, measuring AI visibility in Japan requires at minimum three things:

1. Classify the citation source correctly (authority IT media? comparison site? UGC?).
2. Weight by that classification (don’t count every citation as 1).
3. Be able to isolate “authority sources only” (the raw landscape with noisy UGC set aside).

Having all three natively for Japanese sources is precisely the territory a global tool struggles to cover by construction — and where a localized tool earns its edge.

---

3. VeReach GEO’s Japanese 4-tier weighted source dictionary

To bake in that “difference in authority” from the start, VeReach GEO implements a dictionary that classifies Japanese sources into four tiers by authority and assigns each a weight. Rather than counting every citation as equal, the core is localizing how the citation landscape is read and weighted.

Tier	Category	Weight	Representative domains
tier1	Authority IT media	5.0	ITmedia, @IT / atmarkit, EnterpriseZine, CNET Japan, ASCII.jp, ZDNet Japan, Nikkei / Nikkei xTECH, GIGAZINE
tier2	B2B SaaS comparison & review	3.0	BOXIL, IT Trend, ITreview, 起業ログ (Kigyo Log), Nikkei Business
tier3	Marketing / trade press	1.5	MarkeZine, Web担当者Forum, PR TIMES, ExchangeWire, BRIDGE
tier4	UGC / blogs	1.0	note, Zenn, Qiita, Wantedly

Unknown domains default to 1.0 — a conservative default that says “if we don’t know the weight, don’t overvalue it.”

One honest scope note: this weight is not claimed to be fully fused into the final headline score. What the dictionary does is govern how VeReach GEO reads and weights the citation landscape when presenting it, and on top of that enable a “tier1-only” view (isolating only citations from authority IT media). Staring at a raw pile of citation counts versus being able to flip instantly to “what does this look like if I only count top-authority sources” yields completely different insight.

It also supports domain filtering. Select “tier1 only” in project settings, and at run time VeReach passes only tier1 domains to the target engines, reconstructing a landscape narrowed to the top authority sources. Because VeReach GEO tracks across ChatGPT, Gemini, Claude, and Perplexity, you can align “weighting × tier1-only × cross-engine” in a single view.

---

4. How it works in practice: rebalancing a B2B SaaS PR budget

Let’s bring the abstract dictionary down to a concrete decision. What a Japanese B2B SaaS really wants from GEO is: where should we go earn the next mention?

For B2B/IT, the tier1 targets GEO guides name again and again are ITmedia, BOXIL, and ITreview. ITmedia is authority IT media (tier1, weight 5.0); BOXIL and ITreview are SaaS comparison/review (tier2, weight 3.0). When an evaluation-stage user asks an AI for “best [category] / [category] comparison,” whether your brand appears on these domains heavily influences whether you get cited.

This is where the weighted dictionary earns its keep. Suppose a monitoring run returned this:

Source	Tier	Weight	Your citations	Weighted contribution
ITmedia	tier1	5.0	12	60.0
BOXIL	tier2	3.0	0	0.0
ITreview	tier2	3.0	2	6.0
note (own / employee posts)	tier4	1.0	18	18.0

Look only at raw counts and you’d read “18 citations on note — we’re doing great.” But re-read the landscape with weights and a different story emerges: you’re strong on ITmedia (12 citations at authority tier1), but completely absent on BOXIL (0), the main battleground for evaluation. Evaluation-stage buyers narrow their shortlist through comparison sites like BOXIL, so absence there is a hole that maps straight to conversion.

The prescription is clear: “Cited 12× on ITmedia (weight 5.0), 0× on BOXIL (weight 3.0) → prioritize BOXIL placement in the next PR / content investment.” A decision that was invisible behind “note: 18” in raw counts becomes actionable only once you weight and tier-classify the landscape.

That is what it means to elevate citations from “counting” to “reading.” The sum of citation counts is a vanity metric all too easily; weight by tier and look at which tier has the hole, and your PR and content budget allocation changes.

---

5. The hierarchy isn’t fixed: operating caveats

A few honest reservations to close.

First, exact B2B citation-share numbers are not public. The overall ranking and per-engine tendencies here come from Ahrefs’ public data, but figures at the granularity of “what % of the Japanese B2B SaaS category does BOXIL hold” don’t exist as a public benchmark. That’s exactly why continuously measuring on your own category and your own query set beats guessing.

Second, the hierarchy shifts by engine. ChatGPT weights PR TIMES heavily while Perplexity prefers Wikipedia and note — even with the same tier classification, which sources “work” changes with which engine you look at. That’s why VeReach GEO tracks across all four engines and breaks results down per engine.

Third, weights are a lens for reading the landscape, not the single truth. tier1 isn’t always the right answer; depending on the product and stage, a tier2 comparison site can be the very key to conversion. The dictionary is a tool for showing where the holes are — the final call stays with your business context.

---

Japan’s AI citations have a different topology from the English-speaking world’s. A tool that counts ITmedia and note as the same “1 citation” simply can’t see that topology. VeReach GEO’s Japanese 4-tier weighted source dictionary and tier1-only view are a localized lens for moving citations from “counting” to “reading, weighting, and turning into action.” If you want to see who AI actually trusts in your category — read against the authority of Japanese sources — talk to VeReach GEO.

Want to diagnose which tier of sources cite your brand, and where the holes are? From contact you can receive a weighted × tier1-only landscape report for your own category.

---

A note on data: the overall citation ranking, per-engine tendencies, note’s AI-traffic efficiency, and the SEO↔LLM correlation (~0.65) cited here come from Ahrefs’ 2026 cross-platform Japanese citation study (reported by Web担当者Forum, April 2026). Exact per-category B2B citation shares are not public, so treat them as estimates. Source hierarchies and engine behavior change fast — as of June 2026, verify against your own fresh measurements before any key decision.