MENU
Built in Japanese, not translated.
  1. Home
  2. SEO in Japan
  3. How to Get Recommended by ChatGPT in Japanese: Who Writes the List

How to Get Recommended by ChatGPT in Japanese: Who Writes the List

SEO in Japan
Five plain kraft-covered notebooks standing in a row on a wooden shelf in a Japanese stationery shop, one pulled slightly forward

When someone in Japan asks ChatGPT, Perplexity, Gemini or Google “おすすめの◯◯は？” (osusume no ◯◯ wa, “what is a good ◯◯?”), the answer is usually a short list of product names. Many guides on how to get recommended by ChatGPT are written for questions in English. For a question asked in Japanese, much of that list tends to come from the Japanese pages that already rank for it: listicles titled おすすめ◯選 (osusume ◯-sen, “◯ recommended picks”) and comparison media.

We have seen this with a service of our own. Between June 23 and September 26, 2026, our monitoring of AI answers checked Google’s AI Overview for “おすすめの選書サービス” (recommended book-selection services) 18 times. A book-selection service we operate was named 0 times. The AI Overview followed the page ranking first on Google, a competitor’s listicle, although our service was listed in the listicles ranking second and fourth. Over the same period Perplexity, ChatGPT and Claude did name it.

This article is about being named as a product or vendor in recommendation answers. Being cited as a source, and measuring that, is covered in AI search in Japan. Here we cover where the candidates come from, what to fix first and how to measure the result honestly.

Table of Contents

The short answer

There is no documented way to apply to be recommended. ChatGPT, Perplexity, Gemini, Copilot, Claude and Google’s AI features search the web for questions that need current information, and the products they name probably come largely from what those searches return. In our monitoring, the pages behind recommendation answers were mostly lists and comparisons on other companies’ sites.

In our view, that sets the order of work. First, publish accurate Japanese information about the product, so there is something correct to find and copy. Second, make sure the Japanese listicles and comparison pages that already rank for your category describe you correctly. Third, publish your own pages that answer the buyer’s question directly, from your company’s point of view.

Expect the assistants to differ. Google’s AI Overview was the hardest in our case, because it followed the top-ranking page. Measure mentions, citations and visits separately, every week, and avoid judging a change on one check.

Where ChatGPT, Perplexity, Gemini and Copilot find candidates

Each assistant documents how it searches. None explains how it decides which products to name. What the documentation does suggest is that candidates come from a web search, which turns the problem into a familiar one: which pages that search returns for your category.

AssistantWhat its own documentation saysWhat it means for a product
ChatGPTMay search automatically when a question benefits from current information; sometimes sends rewritten queries to third-party search providers (it links Microsoft’s and Shopify’s privacy policies); sites must allow OAI-SearchBot, and placement is not guaranteedBeing crawlable makes you eligible, nothing more
ChatGPT product resultsSelected independently and not ads; draw on structured data from first- and third-party providers, other third-party content and the model’s own answer before any new search; merchants can apply to supply a product feedWhat other sites say about you is an input
PerplexitySearches the web in real time with numbered citations; its crawler, PerplexityBot, surfaces and links sites in its resultsIts own index; allow PerplexityBot
Google AI Overviews and AI ModeMay run several related searches (“query fan-out”); pages must be indexed and eligible for a snippet; no special markup; AI Mode in Japanese since September 8, 2025Built on Google Search; in our case the top result shaped the answer
Gemini appUses external sources “such as Google Search” to generate responsesGoogle’s index again
Microsoft CopilotIn Microsoft 365 Copilot Chat, sends a short query to Bing; Bing Webmaster Tools counts citations in Copilot and Bing’s AI summariesBing’s index; check your Bing ranking too
ClaudeCalls a search tool to ground answers in the live web, with citationsSearch-backed like the others
What each provider’s own help or developer pages say, checked October 9, 2026. None describes how products are chosen for a recommendation.

Two practical points follow. Check that your robots.txt and your CDN let these crawlers in, as OpenAI and Perplexity both ask. And Copilot works from Bing, whose share of Japanese desktop search has grown since 2025, as we show in search engines in Japan. A product sold to Japanese companies should check where it ranks on Bing for its Japanese category terms, not only on Google.

Why Japanese recommendation answers often lean on listicles

Japanese recommendation content usually follows a settled format. A page titled with a number of picks, such as おすすめ9選 (“nine recommended picks”), lists products, often groups them by type or use, and describes each in a few lines. For the query we tracked, the pages ranking first, second and fourth on Google were all listicles. In Japanese B2B software the same role is played by comparison media (比較サイト, hikaku saito, “comparison sites”), which often hold several of the top results for a category, as we describe in Japanese B2B SaaS and comparison sites.

An assistant asked for a list often has a ready-made one in front of it. When its background searches return a page that already names, groups and summarizes the products in a category, it gets candidates, categories and descriptions from one source. A vendor’s own product page describes one product, and a recommendation answer needs several, compared. The pages that compare are mostly written by someone else, for Japanese readers, with prices in yen and the category words Japanese users type.

What our monitoring showed for two kinds of question

Our monitoring is a fixed list of Japanese questions put to the same assistants every week, with the answers and cited sources recorded. Two results matter here.

A consumer question: the AI Overview followed the top page

選書サービス (sensho sābisu) are services that choose books for a reader. In the 18 checks of Google’s AI Overview for “おすすめの選書サービス,” our service was never named. The AI Overview’s structure, its grouping of services and the services it listed followed the page ranking first, a competitor’s listicle of nine picks. Being listed in the second- and fourth-ranked listicles did not get our service in.

Perplexity, ChatGPT and Claude named the service over the same period, and Gemini’s background searches included its name, so Gemini had at least found it. One query in one category cannot show how AI Overviews always work, and we do not claim it does. It does show that being in a listicle that ranks and being in the listicle that ranks first can lead to different outcomes in Google’s answer.

A vendor question: other companies’ lists did the work

The second case is our own company. In September 2026 our monitoring included Japanese questions asking which firms offer LLMO (optimization for large language models, meaning for AI answers). When the assistants named Amanaut, the sources they cited were mostly other companies’ comparison articles, lists of the “best LLMO companies” type. Our own 20-plus explainer columns on LLMO were almost never used.

The pages of ours that did get cited were different in kind. By September 28, 2026, three had been cited by Perplexity and ChatGPT, and each answered the exact question a buyer asks, from a company’s point of view. An explainer answers “what is LLMO?” while a buyer asks “which firm should we use?” It was the second kind of page that got cited.

What a product without Japanese pages can and cannot do

Without Japanese pages of your own, you depend on what others write. Assistants often name products that appear in the pages they retrieve, whoever wrote them, so Japanese listicles alone can get a product named. What you lose is control. A Japanese author working from an English site has to translate your category, convert or guess your price, and work out whether the product can be bought in Japan. A literal translation can leave you described in words few people search for: in our test of 24 machine-translated category terms, 19 were not the term Japanese searchers use, as we report in translation vs. localization for Japan.

Some things cannot be done at all. You have no Japanese page to point an editor to when you ask for a correction, and no answer in Japanese to the buyer’s follow-up questions about price, support and payment. Nor can you buy your way into ChatGPT’s answer: OpenAI says ChatGPT’s product results are selected independently and are not ads.

How to get recommended by ChatGPT: what to fix first

Each step gives the next one something to work with, and none of it is specific to ChatGPT: the same Japanese pages can feed every assistant in the table above.

Accurate Japanese product information

Start with one Japanese page per product that states what a listicle author or an assistant needs: the product name as Japanese users write it, the category in the words Japanese searchers use, the price in yen, and whether and how it can be bought and supported in Japan. Add a Japanese company profile page (会社概要, kaisha gaiyō) so the company behind the product can be identified; its contents are set out in the Japanese company profile page.

Correct descriptions in the listicles that already rank

Search your category in Japanese with おすすめ (osusume, “recommended”) and 比較 (hikaku, “comparison”), and note the top five pages for each. Check that you are included and that what they say is current: name, price, plans, availability in Japan. Where something is wrong, write to the editor with the correction and your Japanese page as the source. On comparison media, correct the free listing before paying for anything.

A place in a listicle further down the results is still worth having, but for Google’s answer, in our case, the page in first place was what counted. If you pay to be included anywhere, the page has to say so. Since October 1, 2023, Japan’s Consumer Affairs Agency has treated advertising that consumers cannot recognize as advertising as a violation of the 景品表示法 (Keihin Hyōji Hō, the Act against Unjustifiable Premiums and Misleading Representations), and the party it regulates is the advertiser, not the publisher.

We offer this kind of placement ourselves: our editorial roundup format puts a product in a Japanese comparison article that our editors plan and write, labeled PR. In our monitoring a listicle that did not rank near the top did not move Google’s AI Overview, so treat it as one more accurate Japanese page for assistants that read more widely, not as a route into Google’s answer.

Your own comparison-style pages

Only then write your own pages, for the buyer’s question rather than the topic. Our vendor case points that way: explainers were almost never used, while pages answering a buyer’s question from a company’s point of view were cited. In Japanese that means pages built around 比較, 選び方 (erabikata, “how to choose”) and ◯◯と△△の違い (“the difference between ◯◯ and △△”), written in Japanese from the start. Make the comparison fair enough to be believed. A table in which you win every row is likely to read as an advertisement.

How to measure recommendations honestly

Recommendation answers change from run to run, so a single screenshot proves little. Use a fixed list of Japanese questions a buyer in your category would type, put them to the same assistants on the same day each week, and record whether you are named, where in the list, and which pages are cited. Keep three measures apart.

MeasureWhat it tells youWhere it comes from
MentionThe assistant named your productYour own weekly checks
CitationThe assistant linked one of your pages as a sourceYour checks; Bing Webmaster Tools for Copilot and Bing; Search Console, which counts Google’s AI features within its Web search type
VisitSomeone clicked through to your siteReferrals in your analytics

One query in our monitoring went from three mentions to none in a single week. Judge a change on several weeks of checks, not on one.

Citations are not visits either. For Book Crunch, the Japanese book media site we run, Bing Webmaster Tools counted roughly 1,165 citations a day on average in September 2026, in a report that covers Microsoft Copilot and Bing’s AI summaries. AI assistants sent Book Crunch about 80 visits in 30 days. Getting mentioned in ChatGPT or cited in Copilot is where the work starts; whether it sells is a question for your inquiry and sales figures.

Common questions

Three questions often come up when teams plan for Japanese AI answers.

Can we pay to be recommended by ChatGPT?

Not directly. OpenAI says ChatGPT’s product results are selected independently and are not ads, that ads are shown separately, and that placement in its search results is not guaranteed. What you can pay for is space in Japanese media that assistants read, and in Japan that space has to be recognizable as advertising. It puts you on a page. It does not guarantee you a place in an answer.

Which assistant should we work on first?

Work on the pages first, because the same Japanese pages feed all of them. Google’s AI Overview is probably the hardest to change if you are not already near the top of Google’s results; in our case it followed the first page. ChatGPT, Perplexity and Claude appear to read more widely. Copilot depends on Bing, which matters more for buyers who search on desktop.

An agency offers to get us ranked in ChatGPT in Japan. What should we ask?

Ask which Japanese questions they will track, in which assistants, how often, and whether they report mentions, citations and visits separately. Ask which pages they will change and why. Be wary of any promise of a place in an AI answer: OpenAI itself says placement in ChatGPT’s search results is not guaranteed, and our own results moved from week to week.

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. The citation side of AI answers is covered in AI search in Japan, and the comparison sites that hold Japanese software categories in Japanese B2B SaaS and comparison sites. More about how we work.

Sources

The monitoring results and the Book Crunch figures are our own. Everything else comes from the sources below.

Last reviewed: October 2026.

SEO in Japan
Found this useful? Send it to whoever is running your Japan launch.

Author

Portrait of Hiroki Tanaka Hiroki Tanaka

Founder and CEO of Amanaut, Inc., a Japanese content and SEO company, and a content director at Willgate since 2023. Holds SEO Certification Level 1 (All Japan SEO Association). Writes and edits Nihongo Search.

Related Articles

Table of Contents