Last week, i saw my friend’s phone before she left for Tokyo. She hadn’t opened google. Instead of that, she had opened chat gpt and typed “where should i stay in Asakusa for a traditional vibe under 25000 yen a night”. It gave her three properties, with a paragraph about each one. It also gave her a convincing pick. Not one was the small, actually beautiful guesthouse two blocks away from me with a 4.7 star rating on google. It looks great on Google Maps to the algorithm, but didn’t exist to the ai.
Hotels in Tokyo are experiencing a Discovery gap by ai visibility. It is not because demand is low, it is because how we discover hotels has shifted. Roughly 40 percent of global travelers now use ai tools to plan their trip. Just last year, 1 study found that only about 16 percent of hotels worldwide appear in results when using these tool. Record tourism does not help if your property is never mentioned by the ai.
The Tokyo paradox: Record Demand, Missing Hotels
Japan has never been more wanted than now. Japan reached an all-time high of 42.7 million international visitors in 2025; this marks a change of about 15.8 percent from the previous high of 36.9 million visitors in 2024, while those visitors spent a new high of 9.5 trillion yen (about $60 billion usd). The government is publicly pursuing an aim of 60 million annual visitors by 2030. You don’t have a demand issue.
You have a Discovery issue. How travelers find where they will stay has quietly shifterd. Industry data suggests that roughly 40% of global travelers now use ai-tool during planning their trips. Among millennials & gen z that figure climbs as high as 60%. Those are exactly the type of travelers currently flooding Tokyo right now; many of them decide where to sleep before they ever even open a booking site.
The 16 percent problem
Here is the number every independent operator in Tokyo needs to pay attention to. A 2025 analysis by HotelWorld AI ranked over 130,000 properties across 30 countries and found that only about 16 percent of hotel supply world-wide is visible inside of chat gpt, google’s ai, or Perplexity. The same report stated that chain-affiliated properties are much more likely to be surface than independents. This creates what they called a “two-tier system of hotels” where the ai knows some and most do not.
This means that if you run an independent property in Tokyo, there is a good chance you fall into that unvisible 84 percent. And unless you can get found, then the traveler who is ready to book after asking an ai for recommendations will be directed to other properties that are found by the ai. This is expensive in today’s hot market, since these types of travelers are high intent and ready to book.
Why isn’t a great listing on Google Maps enough?
For the past fifteen years, “being found” meant ranking well in google, winning local pack positions and collecting reviews. However, this instinct is now actively misleading because the machine that provides the answer for ai is reading different signals and the channel used to find you in google vs the channel used by the ai to provide answers are separate and do not feed off each other.
Google Maps evaluates hotels based on their proximity to a search, and the completeness and activity level of their google business profiles. Also, the number of positive reviews a hotel has received, as well as its categorization accuracy. This allows a hotel to potentially win on maps through having a high volume of positive reviews, an accurate categorization, and keeping their Google Business Profile active and complete. Additionally, the travel industry sees 79% of the links clicked from google’s AI Mode, redirected to a traveler’s Google Business Profile. Thusly, the Google Business Profile is becoming a primary source of traffic, and not simply a directory listing.
AI tools evaluate recommendations based on authority signals
Ai tools aggregate data. For example, when you ask the tool “what is the best quiet ryokan style hotel to stay in Tokyo”, it will typically name several properties. Unlike traditional search methods, such as Google Maps, there is no second page of results. One hospitality professional stated that large language models generally provide users with limited options, therefore the content that they provide needs to be sufficient enough to allow them to be included as an option.
The primary sources used to validate the responses provided by ai tools are generally not from the owner/operator of the property being recommended. Research was completed across multiple platforms and revealed that ai tools favor third party signals, including social media and travel blogs over official travel guides. Furthermore, each of the ai engines utilized today utilizes a different database. The chatgpt search engine uses Bing to determine if the user has created content that would be visible to the chatgpt search engine. Therefore, the level of visibility that a hotel or travel business has on Bing determines if that business will be referenced in chatgpt. The Perplexity search engine references approximately 21.87 sources per response, making it the reference source with the highest frequency. Additionally, unlike google, which creates a single set of criteria for referencing businesses, Gemini operates as a much more restrictive version of google. An analysis conducted by Yext examined 6.8 million citations across six platforms and found that 52.15% of Gemini citations were attributed to brand owned websites. Therefore, due to the varying levels of visibility that exist among the various platforms cited above, an audit completed solely on One platform provides very little insight into overall citation performance.
Research results – what they really mean
At this point it becomes clear that we have moved beyond speculation. The foundational research is the princeton and Georgia Tech GEO study (Aggarwal et al., ACM SIGKDD 2024). This study was the first peer reviewed research focusing on generative engine optimization. Through testing numerous tactics across approximately 10,000 queries, researchers determined that the greatest influence on improving visibility among ai engines were credibility signals rather than keywords. Researchers found that adding statistics and providing credible sources with quotations resulted in increases ranging from 30-41%. Conversely, utilizing keyword stuffing did not produce any significant improvement. According to a BrightEdge analysis, updating pages every sixty days increased the likelihood of appearing in ai answers by nearly 2 times. Additionally, utilization of faqs combined with structured data resulted in an increase in ai citations by approximately 44%.
A somewhat transparent disclaimer exists. The actual percentage of occurrences of ai overview results is actually debated widely. Some studies indicate that ai overviews occur approximately 15.69 percent of searches (Semrush – late 2025). While others estimate that ai overviews occur in nearly 48 percent (BrightEdge) of searches or even higher estimates of 60 percent (advanced Web ranking – April 2026). Irrespective of methodology utilized to calculate percentages, One thing is certain: the trend indicates movement in One direction.
Furthermore, it is estimated that approximately eighty percent of URLs listed as references in ai engine results are not ranked in google’s top 100 for the corresponding original query. Therefore, a hotel could be the most prominent hotel in a yearbook and never be mentioned by an ai engine searching for the same information. It is akin to being the most prominent restaurant on a street route avoided by GPS systems. The food may be excellent but travelers using GPS will never know to stop.
How to Get Your Tokyo Hotel Cited in AI Search: The 2026 Fix
Google Maps ranks hotels on proximity, Google Business Profile completeness, and reviews. AI tools like ChatGPT and Perplexity build recommendations from third-party citations, structured data, freshness, and consistent information across the web. The Princeton GEO study found that adding statistics, citations, and quotations lifts AI visibility by 30 to 41 percent, while keyword stuffing does nothing.
Google Maps Rewards Proximity and Profile
Maps is a local ranking system. It cares where the searcher is standing, how complete and active your Google Business Profile is, your review count and rating, and your category accuracy. You can win Maps by being close, well-reviewed, and diligent about your profile. That channel is real and not going away. In travel specifically, 79 percent of hotel link clicks from Google’s AI Mode still flow back into a Google Business Profile, which means your profile is now a top-of-funnel landing page, not a directory afterthought.
AI Answers Reward Citations and Consensus
AI tools synthesize. Ask for “the best quiet ryokan-style stay in Tokyo” and the model names two or three properties. There is no page two. As one hospitality analysis put it, large language models surface only a handful of results, so your content must be authoritative enough to make the cut.
The sources those models trust are mostly not you. Research compiled across platforms found AI tools lean on third-party signals like Reddit, travel blogs, and review aggregators, and that niche travel blogs often outperform major travel guides. Each engine also draws from a different index. ChatGPT Search runs primarily on Bing, so your Bing visibility decides whether ChatGPT can cite you at all. Perplexity cites about 21.87 sources per response, the most of any major platform. Gemini behaves like a stricter Google, with one Yext analysis of 6.8 million citations finding 52.15 percent of Gemini citations come from brand-owned websites. The same brand can see citation volumes differ by up to 615x across platforms, which is why a one-platform audit tells you almost nothing.
What the Research Actually Shows
This is where it stops being mysterious. The foundational work is the Princeton and Georgia Tech GEO study (Aggarwal et al., ACM SIGKDD 2024), the first peer-reviewed research on generative engine optimization. Testing tactics across roughly 10,000 queries, it found the changes that lift AI visibility most are credibility signals, not keywords. Adding statistics, citing reliable sources, and including quotations each produced gains in the range of 30 to 41 percent, while keyword stuffing produced essentially zero benefit. BrightEdge analysis points the same way: pages updated within 60 days are roughly 1.9 times more likely to appear in AI answers, and FAQ plus structured data lifted AI citations about 44 percent.
One honest caveat. How often AI Overviews appear is genuinely contested, ranging from about 15.69 percent of queries (Semrush, late 2025) to roughly 48 percent (BrightEdge) to 60 percent (Advanced Web Ranking, April 2026), depending on methodology. Anyone quoting one tidy figure is selling certainty the data does not support. What is not contested is the direction.
And the kicker: around 80 percent of URLs that AI engines cite do not rank in Google’s top 100 for the original query. So you can be the most popular hotel in a yearbook the AI never opens. It is a little like being the best restaurant on a street the GPS routes everyone around. The food is great. Nobody is told to turn.
Frequently Asked Questions (FAQ)
Question: is ai search really effecting bookings at hotels in Japan already?
Answer: yes. Approximately 40 percent of global travelers now use ai tool during trip planning, up to 60% for younger travelers who make up a large share of Tokyo’s record inbound numbers. Many decide where to stay within an ai conversation before opening a booking site.
Question: what does “only 16 percent of hotels are known to ai” actually mean?
Answer: a 2025 HotelWorld AI study found that only about 16 percent of hotels appeared in results for chat gpt, google’s ai and Perplexity. The rest are basically unknown to these tool with chains far more likely to be surfaced than independents.
Question: if demand for Tokyo is at an all-time high why should i care about being visible?
Answer: because demand and Discovery are not the same things. Record tourism only helps you if travelers can find you and the way people find where they will stay has quietly shifted. A growing share of them now first ask an ai about where they will stay. If the ai never named your property then the demand will flush through to the properties that it did name
Question: why does my hotel show on Google Maps but not in chatgpt?
Answer: maps assesses hotels on proximity, Google Business Profile and reviews. However, chatgpt generates answers from another database – Bing – and third party citations and structured data. Winning maps does not guarantee inclusion in ai-based recommendation sources.
Question: are all ai engines utilizing hotels the same way?
Answer: no. While both chatgpt and Gemini utilize Bing databases as a starting point for generating answers – each tool processes data differently. Chatgtp relies heavily on third-party opinions – Gemini focuses primarily on brand-owned sites with correct schema – and Perplexity places greater emphasis on recency of citations and references many sources per answer. Additionally, citation patterns can vary significantly – up to 615x – between platforms. Therefore, each platform must be measured separately.
Question: does keyword stuffing improve my chances of being included in ai answer sources?
Answer: no. The Princeton GEO study showed that keyword stuffing generated virtually no value and slightly decreased performance on some platforms. Statistically valid sources (such as statistics), citations from reputable organizations and quotes generate a thirty-two to forty-One percent increase in ai exposure.
Question: Why does my hotel show on Google Maps but not in ChatGPT?
Answer: Maps ranks you on proximity, your Google Business Profile, and reviews. ChatGPT builds answers from a different index, primarily Bing, plus third-party citations and structured data. A strong Maps presence does not automatically feed those AI signals, so you can win one and be absent from the other.
Question: Do all AI engines pick hotels the same way?
Answer: No. ChatGPT runs on Bing and leans on third-party consensus, Gemini favors brand-owned sites with clean schema, and Perplexity weights fresh citations and names many sources per answer. Citation patterns can vary up to 615x across platforms, so each one needs separate measurement.
Question: Does keyword stuffing help me get into AI answers?
Answer: No. The Princeton GEO study found keyword stuffing produced essentially no benefit and slightly hurt on some platforms. Statistics, citations, and quotations are what lift AI visibility, by 30 to 41 percent in that research.