“Sure, But Those Numbers Are All From English”
You have probably heard some version of that in a meeting.
Someone shares a deck about AI search, the citation leaderboard says Reddit and LinkedIn, and the person selling into a non-English market shrugs. Fair enough. A leaderboard built on English questions tells you very little about who gets quoted when the question is asked in Japanese, or German, or Spanish.
Which is why a Japanese-market count is worth a look even if you never sell a thing in Japan.
Someone measured which domains five AI assistants actually cited for Japanese-language questions, twice, two months apart. The interesting part is not the winner. It is how much of the list moved.
Which Sites Do AI Assistants Cite In Japan?
YouTube ranked first and note.com second in the June 2026 count, with note climbing from fifth in the April measurement R. Four of the ten slots went to domains that were not on the list in April.
The five assistants covered were Google AI Mode, Google’s AI Overviews, ChatGPT, Perplexity and Microsoft Copilot.
A cited domain is simply a site the model leaned on while assembling its answer. It does not have to be yours to matter to you: if the models keep quoting a review site or a press-release wire when your category comes up, that is a surface you can be present on.
That is the practical reason to read a list like this at all. Watching your own domain’s rank tells you whether you are winning. Watching the whole list tells you where you could go.
Four Slots Opened, Four Slots Filled
The newcomers were Yahoo! Chiebukuro, Japan’s big Q&A board (fifth), Reddit (seventh), the comparison site my-best.com (eighth) and the Rakuten marketplace (ninth).
Falling out entirely: Google itself, previously third, along with Instagram, Amazon and the Yahoo! news portal.
Inside the list, the blogging platform Ameblo moved from seventh to fourth and the press-release wire PR TIMES from ninth to sixth, while Japanese Wikipedia slipped from second to third.
Per assistant, the sharpest move belonged to Reddit inside ChatGPT: eighth place to first.
English Wikipedia also lands fourth in ChatGPT’s Japanese-language results, above the Japanese edition — which the publisher reads as a sign that ChatGPT is expanding the question into English behind the scenes and searching there.
Now, there is a tempting conclusion here, which is that a list this volatile is not worth optimizing for.
I would read it the other way. Four slots turning over in two months means those slots are not locked, and a leaderboard where the same names have sat for years is a much harder thing to break into than one that turns over at this rate.
The Industry Cut Looks Nothing Like The Headline List
The same analysis published a second ranking, this one restricted to car-related questions: new cars, EVs, used cars.
That top ten reads: Goo-net second, carview third, Nextage fifth, Car Sensor sixth, MOTA eighth. More than half of it is automotive trade media and used-car listings.
Neither Ameblo nor PR TIMES, both strong on the general list, appears in it at all.
So copying the general leaderboard into your own content plan will point you at the wrong surfaces. The general list nudges you toward blogging platforms and video. In the automotive context, what the models were actually reading was used-car inventory sites.
The publisher’s read is that AI assistants preferentially cite specialist media inside a specialist domain. They add two more observations:
- Official first-party sources carry weight. PR TIMES does well because it accumulates dated, factual, company-issued announcements
- The arrival of Reddit, Yahoo! Chiebukuro and my-best.com suggests the models are leaning harder on actual user opinion
Surfaces you publish on yourself, and surfaces where other people talk about you. The lesson is not that one of them wins.
What To Flag When This Goes Into A Deck
Five things, and they matter if anyone is going to set a target off these numbers.
The analysis comes from Ahrefs, which sells AI visibility tooling. “Make AI citations visible and act on them” is a conclusion that happens to suit the product. The analysis tool used, Agent A by Letaido, is also theirs.
Only ranks were published. The publisher says citation counts and shares move too much with the query set and time window to release, which means you cannot tell how far second place sits behind first.
That gap matters the moment someone in the room proposes a percentage target. This data cannot support one.
The query set is not disclosed either — described only as spanning Japanese queries broadly, with no question count and no selection criteria. And the whole thing is a vendor measuring with its own tool, with no independent replication.
Finally, the industry cut covers exactly one industry. “Specialist media probably dominate in our sector too” is a hypothesis, not a finding.
Strip those five away and “note.com is number two in Japan” starts walking around the company on its own. Keep them, and the useful part survives: this list moves, and it moves fast.
Count Your Own Category Before You Copy Anyone’s Leaderboard
Two things are worth doing here, and the first one is cheap.
Write down ten questions a prospect would plausibly type into an assistant about your category, run them, and count which domains show up in the citations. That is your leaderboard. The general one belongs to someone else’s question set.
The automotive numbers exist to make exactly that point.
Then pick one surface from your own list where you can put first-party information — a company announcement, a bylined piece in a trade outlet, an entry in a comparison roundup — and go be present there. You do not need all of them at once.
The part that catches people out is what happens next. Your own analytics will show you traffic, but they will never show you whether a model quoted you, because that answer never touched your site. You have to count it from the model’s side.
In a list where four slots turn over in two months, being able to notice the turn is most of the value.
Source
- Takanori Kawaharada, “note Jumps To #2, Reddit Takes #1 In ChatGPT: Latest Ranking Of Domains AI Cites In Japan, June 2026”, Ahrefs Blog, 2026-06-23, https://ahrefs.com/ja/blog/brand-radar-top-10-cited-domains-2026-june/
