Google AI Overviews Visibility Audit, Japanese and English

In the last query set we ran, 600 Japanese commercial queries across electricity retail, packaged food and skincare, 41 percent of AI Overview citations pointed at pages that did not rank in the top ten organic results for the same query. Ranking and being cited are separate outcomes with separate causes. This audit tells you which of the two you are losing on, query by query, and what the cause is. How we build the query set: From your Search Console export and from the questions your sales and support teams answer in Japanese every week. Then we classify every query by whether an Overview fires at all. In our energy work, roughly a third of 電力比較 queries return no Overview, because Google suppresses them where the intent is transactional and the comparison sites already own the page. Optimising against those is spend with no ceiling, and we will tell you to stop before you brief anyone to write for them. For queries that do fire, we capture the Overview and the AI Mode response from a Tokyo exit, DOM and screenshot, ja-JP and en-US kept separate, three times across ten days. One capture is an anecdote; the response is not stable enough to read from a single pull. Then we take the citations apart. Which domains hold the answer in your category. Which of your URLs appear and on what. Which passage of a cited page was actually used, and what that passage does that yours does not. In skincare this is rarely a content-quality gap. It is that the cited page states a plain claim and yours hedges. 薬機法 forces the hedge, and the hedge deletes the extractable sentence. We show you where that trade sits page by page, and what a legally clean direct answer reads like when the review is done by someone who has read the 表示 guidance rather than guessed at it. What you get: A query-level table: trigger rate, who holds the citations, where you sit. A written read of the four or five structural reasons your pages are passed over, each with its evidence. A ranked list of pages to change and, as importantly, the pages we would leave alone. The raw capture corpus as JSONL with timestamps and the surface each response came from, so you can rerun it in six months against the same baseline instead of a new one. What is not in this: We do not rewrite the pages here. We do not build links, place PR or approach publishers. We do not report an AI visibility score, because there is no denominator that makes one honest, and a number you cannot decompose is a number you cannot act on. We will not commit to a citation count. Nobody controls Google's selection, and a firm that quotes you one is selling you something other than this. Who this is not for: Brands whose Japanese site is translated from English by machine. Those pages are not retrieved, and no audit changes that; fix the source language first and come back. Companies with no third-party presence in Japanese, no trade press, no retailer listings, no product database records, because the audit returns one finding you already suspect and you will have paid us to confirm it. And anyone who wants this run somewhere we do not work. We cover Japan and English-language global search. We decline the rest rather than learn on your budget. How it runs: Kickoff, then we work alone through the capture window. One working session mid-way to test our reading of your category against yours, in Japanese or English. A written deliverable and a two-hour readout with the people who will action it. We ask that your search lead and whoever owns claim approval both attend, because most of the decisions land in the space between them.