Why AI Assistants Don’t Recommend Your Brand in Japanese

Source: UDX, “Japan Listed Companies AI-Search Readiness Survey,” n=3,709 domain-resolved companies on the Prime, Standard, and Growth markets of the Japan Exchange Group, as of July 20, 2026. Full methodology. The competitive implication: in most Japanese categories, the incumbents are not defending this ground.

The five reasons, in the order they usually apply

1. Nothing in Japanese refers to you except you

This is the dominant cause and the least technical one. Assistants answering a recommendation question lean on third-party sources — reviews, listings, directories, press, forum discussion, comparison content. A brand whose entire Japanese-language footprint is its own website gives the model nothing corroborating to work from, and self-description alone rarely earns a recommendation slot. No amount of on-site optimisation substitutes for this.

2. You are legible in English and invisible in Japanese

A brand can be well represented in English answers and entirely absent from Japanese ones. The two are effectively separate questions with separate evidence bases. Testing your visibility by prompting in English will tell you almost nothing about what a Japanese buyer is shown — the prompts have to be in Japanese, phrased the way a Japanese buyer would actually phrase them.

3. Your site is unreadable to the crawlers, often unintentionally

Three recurring technical faults, in rough order of frequency:

• AI crawlers blocked in robots.txt — frequently inherited from a template or a security default that nobody revisited. Worth checking before anything else, because it invalidates all other work.
• No structured data — leaving the model to infer what your organisation, product, and pricing are from prose.
• Key content rendered client-side — the specifications and pricing a buyer needs exist in the browser but not in the fetched document.

4. Your language signalling is ambiguous

If Japanese and English versions of a page are not explicitly related to each other, and the page language is not declared, systems must guess which audience each page serves. With hreflang implemented by 12.5% of listed Japanese companies, this is not an exotic failure — it is the norm. Correct signalling is cheap and it is one of the few items entirely within your control.

5. You have no machine-readable account of what you do

The llms.txt convention (see llmstxt.org) provides a plain summary of what a site is and where the substantive pages are. Adoption is not required and it is not a ranking mechanism — but at 4.5% adoption among listed Japanese companies, it costs an afternoon and puts you in a small minority that has stated its own case unambiguously.

How to measure it rather than guess

AI visibility feels unmeasurable, which is why most teams treat it as a vibe. It is measurable, with a method that is unglamorous but sound:

1. Fix a Japanese prompt set reflecting how buyers in your category actually ask — recommendation, comparison, and problem-first phrasings. Fix it once, then leave it alone; changing prompts between rounds destroys comparability.
2. Query multiple assistants, since they disagree, and disagreement is itself informative.
3. Score presence, position, and characterisation. Being mentioned dismissively is a different problem from not being mentioned, and needs a different fix.
4. Score your competitors on the same prompts, so you have a benchmark rather than an absolute number that means nothing on its own.
5. Re-run on a fixed schedule. Individual responses vary between runs; the trend across rounds is the signal, a single run is noise.

One caveat worth stating plainly: these systems change, and any specific tactic here may age. What does not age is the underlying requirement — be technically readable, and be discussed by sources other than yourself.

The realistic order of work

Most teams do these in reverse, starting with content and never touching the technical gate that was suppressing everything. Check robots.txt first. It takes minutes and it is the most common single point of failure we find.

UDX Inc. is a digital marketing firm based in Japan. We publish our own survey of AI-search readiness across all listed Japanese companies, which is the same method used in the services above. Related reading: How to Enter the Japanese Market · Do You Need a Japanese Landing Page? Last reviewed August 2, 2026.