The question foreign teams ask about AI search in Japan is usually “has it arrived yet”. It is the wrong question, and it produces a wait-and-see answer that costs a year.
The better question is narrower and answerable: when somebody asks about your category in Japanese, what is the model actually reading? Because whatever it reads is what it repeats, and for most foreign companies the honest answer is that it is reading somebody else.
What is actually running in Japan
Dates matter here, because a lot of English-language guidance about Japan is written from a snapshot that has moved.
- Google AI Overviews reached Japan on 15 August 2024, in the second wave of countries after the US. Japanese-language AI Overviews followed on 28 October 2024, when Google took the feature to more than 100 countries and added Japanese alongside Hindi, Indonesian, Portuguese and Spanish. Japanese has been a supported language for well over a year.
- ChatGPT is widely used in Japan, including inside companies, and is frequently the first place a research question goes rather than a search box.
- Perplexity had an unusual distribution advantage here. From 19 June 2024, SoftBank gave subscribers on its SoftBank, Y!mobile and LINEMO brands a free year of Perplexity Pro — three consumer mobile brands at once. That consumer promotion has since ended, and SoftBank now resells Perplexity Enterprise Pro to Japanese businesses instead. But for a year, a large cohort of Japanese users was seeded onto an answer engine at no cost, and habits formed in that year did not evaporate when the promotion did.
The practical reading: AI answers are not a future consideration in this market. They are a present one, and the Perplexity episode is a reminder that distribution in Japan can move through carriers rather than through the open web.
Japanese questions get answered from Japanese sources
This is the rule that decides almost everything else, and it is the same rule that governs ordinary Japanese search — only enforced harder.
When the question is asked in Japanese, the candidate pool is overwhelmingly Japanese-language material. Your English documentation may be the most authoritative writing that exists on your product anywhere in the world. If the question is in Japanese, it is not in the room.
Two consequences follow, and they are worth stating plainly because they are usually discovered late.
- Having no Japanese content is not neutral. The model still answers the question. It answers it from your competitors, from Japanese comparison media, and from whatever a Japanese blogger wrote about you three years ago.
- What is said about you gets repeated. Where a search result at least showed the reader a list of sources to judge, an answer flattens that into a single confident paragraph. An outdated price or a discontinued feature, sitting in a Japanese article you have never seen, becomes the answer.

What gets cited, and why you are not in it
We have started recording which domains appear as sources in Japanese AI answers in the categories we work in. It is a small, hand-collected sample rather than a study, so treat what follows as an observation rather than a measurement. It has been consistent so far, and it is not flattering to newcomers.
The cited set is narrow, and it leans towards Japanese aggregators and category media rather than the companies that make the products. Comparison sites, established Japanese trade media and a handful of large operators take most of the citations. A vendor’s own site appears, but usually only when the question names the vendor.
That has an uncomfortable implication for anyone entering the market. For the questions worth winning — the ones where the buyer has not yet decided who to consider — being cited often means being covered by Japanese media you are not in yet, not merely publishing a better page. That is a longer game than a content calendar, and pretending otherwise wastes budget.
It also means the useful early goal is narrower: be the source that gets cited when the question does name you, and make sure what is said is current and correct.
Being citable is not the same as ranking
Ranking rewards a page that covers a topic. Citation rewards a passage that answers a question and can be lifted without ambiguity. Those pull in slightly different directions, and the difference is mostly in the writing.
- Answer in the first sentence under the heading. A section that spends three paragraphs building to the point gives a model nothing clean to take.
- Attach numbers to dates and sources. “About 80%” is unusable. “80.1% as of March 2026, per Statcounter” survives extraction, and it survives a reader checking it.
- One idea per heading. A heading covering three things produces a citation that misrepresents all three.
- Be a resolvable entity. A Japanese company profile page with your registered name, address and representative does more for machine confidence than another blog post. In Japan this page is a genuine convention rather than a formality.
- Name a qualified reviewer in regulated categories. A claim attached to a named professional gives a model something to attribute. In Japanese health, finance and legal content this is close to an entry condition.
None of that is exotic. It is the writing standard that good Japanese editorial already uses, which is why content written natively in Japanese tends to do better here than content translated into it — the translated version usually preserves an English structure that was never built to be quoted.

You cannot measure this properly yet
We would rather say this than sell around it. There is no reliable tool for measuring Japanese-language AI visibility. The Western platforms that offer AI-visibility tracking handle Japanese poorly or not at all, and the Japanese SEO tools have not built it.
What we do instead is unglamorous: a fixed list of Japanese questions, put to the assistants by hand, with the answers and their cited sources written down. It is manual and the sample is small, and we would rather say so than dress it up. It is still better than a dashboard that quietly measures English.
If a vendor offers you Japanese AI-visibility reporting, ask how the queries were generated and in which language the answers were collected. The answer is informative.
Where we would start
- Ask about yourself in Japanese. Take ten questions a Japanese buyer would actually ask, put them to the assistants your buyers use, and read what comes back. Most companies have never done this and are surprised by it.
- Fix what is wrong about you first. Correcting an outdated fact that is being repeated is cheaper and faster than trying to rank for a new term.
- Publish the boring reference pages in Japanese. Specifications, pricing basis, company profile, what the product does not do. These get cited disproportionately because they are unambiguous.
- Then write for the questions. Not keywords — questions, in the form a Japanese buyer would type them.
Common questions
Is there a separate discipline for this, or is it just SEO?
Mostly it is the same work done more carefully. The inputs overlap almost entirely. What changes is the writing standard — clean, attributable passages instead of comprehensive pages — and the fact that you are now competing to be quoted rather than clicked.
Should we block AI crawlers?
If your goal is to be found in Japan, no. Blocking removes you from the candidate pool without removing the question, so the answer gets built from whoever wrote about you instead. Blocking is a defensible choice for a publisher whose product is the text itself. It is rarely the right one for a company trying to enter a market.
Do AI answers reduce our traffic?
For questions with a short factual answer, yes — that click was never going to convert anyway. For questions where somebody is choosing a supplier, the click still happens, because nobody signs a contract on the strength of a summary. Look at which of your pages serve which kind of question before drawing a conclusion.
Can we just translate our English content and be cited?
You will be in the pool, which is better than being absent. But you will be competing against material written natively for the question, and translated text usually keeps a structure that was built for a different reader. The gap shows up most in the first sentence under each heading, which is exactly the sentence that gets quoted.
Working on Japan?
Nihongo Search is published by Amanaut, Inc., a small Japanese SEO and content company in Tokyo working with companies entering the Japanese market, and we keep the hand-collected citation notes described above because nothing better exists yet. The tooling gap in more detail is in SEO tools for the Japanese market, the writing standard is in writing Japanese SEO content, and the wider sequence is in our our guide to SEO in Japan. More about how we work.
Sources
- Google, “AI Overviews in Google Search expanding to more than 100 countries”, 28 October 2024 — Japanese added as a supported language.
- Google, AI Overviews country expansion, 15 August 2024 — Japan included.
- SoftBank Corp. press release, 17 June 2024 — one-year free Perplexity Pro for SoftBank, Y!mobile and LINEMO subscribers from 19 June 2024. The consumer promotion has since ended; SoftBank remains an authorised reseller of Perplexity Enterprise Pro in Japan.
Last reviewed: August 2026.
