This page is a summary edition. To keep maintenance effort focused on the observations themselves and on the accuracy of the Japanese original, non-Japanese editions have been published as summaries since September 2026. The full report — all sections, figures, determinations, and the revision record — is published in Japanese: full text (日本語). The observation data is language-neutral and can be downloaded below.
Summary
2nd edition (revised August 20, 2026). First edition published August 2026; changes are listed in the revision record.
For this report, ChatGPT was asked for recommendations on Kyoto "tea ceremony," "kimono," and "zen meditation" experiences, 20 times each across four languages, logging the sources cited in each answer (category × language = 12 cells, 240 observations).
Which sources get cited differs sharply by category and by language. Three angles summarize the differences: category, language, and which businesses appear.
By category: kimono depends most heavily on comparison media and OTAs, while zen meditation centers on public-sector sites and the businesses' own official sites — a gap that survives even after subtracting the run-to-run variation inherent in 20 observations (details in the Japanese full text, §3.2).
By language: the differences concentrate in English. For tea ceremony and kimono, only English leans strongly toward comparison media and OTAs; for zen meditation, almost no language difference is observed (details in §3.3).
By business: how much the recommended businesses overlap across languages also depends on the category. For kimono, roughly the same businesses appear in all four languages; for tea ceremony and zen meditation, only English returns a different lineup — the "only English differs" pattern seen in citation sources reappears, in the same shape, at the level of who gets recommended (details in §4).
These differences appear to correspond to two conditions: the category's commerciality, and the information supply in each language. This issue is a single-timepoint, single-engine cross-sectional observation, however, so this correspondence is a hypothesis consistent with the observed pattern, not a proven causal relationship.
Report #000 (this issue) defines the observation method and evaluation metrics as a baseline. Future monthly reports track how this structure changes, starting from this issue as t=0.
Dominant citation source across the 12 cells (share, %)
| Japanese | English | Simplified Ch. | Traditional Ch. | |
|---|---|---|---|---|
| Tea ceremony | Public 52.2 | Media 56.6 | Public 43.9 (2nd: Owned 30.5, gap 13pt) | Public 41.2 (2nd: Owned 30.9, gap 10pt) |
| Kimono | Media 53.7 | Media 45.5 + OTA 34.5 ≈ 80 | Media 34.1 (2nd: Public 29.5, gap 5pt) | Media 37.0 (2nd: Owned 32.6, gap 4pt) |
| Zen meditation | Public 59.6 + Directory 28.1 | Owned 67.1 | Public 45.7 (2nd: Owned 34.3, gap 11pt) | Public 51.3 (2nd: Owned 28.2, gap 23pt) |
Where the gap to the runner-up is within the cell's noise floor, the leader is a point estimate only and is not treated as a settled ranking. Determination criteria and the cell-by-cell analysis are in the Japanese full text, §3.
Method
Runs used ChatGPT (GPT-5.5, web version) in a temporary chat, logged out. Queries followed a keyword pattern of "Kyoto + category term + recommendation term," in Japanese, English, Simplified Chinese, and Traditional Chinese, for three categories: tea ceremony, kimono rental, and zen meditation. Each query was run 20 times (n=20) consecutively, logging the business names and cited domains in each answer along with execution date/time, location (Kyoto City), network, and model name. Cited sources were classified into six types — Public, Owned, OTA, Media, Directory, UGC — recorded per outlet and aggregated per brand. Exact query strings, classification definitions, and determination thresholds are in the Japanese full text, §2.
Revision record
2nd edition (August 20, 2026) — The first edition counted the registration info displayed alongside business names in answers (business info cards) as citations. That is a separate pathway from search-based citations. The 2nd edition separates the two: the figures in the report are re-aggregated on the citation pathway only, and the card-pathway data is kept and published as an independent metric (placecards.csv). This change flipped the most-cited source type in 6 of the 12 cells, and the following two statements from the first edition are retracted:
- "Across the 6 Simplified + Traditional Chinese cells, businesses' own sites were the most-cited type in every category" — not true on the citation pathway; the leader in those cells is public-sector sites or comparison media.
- "What changes with language is the sources, not the businesses ultimately recommended" — roughly holds for kimono, but not for tea ceremony or zen meditation.
The full account is in the Japanese full text, "8. 改訂記録".
Data
- Raw data data.csv (CC BY 4.0 · 12 cells × n=20 = 240 runs · 764 rows at cited-domain level)
- Business info card pathway placecards.csv (same conditions · 329 rows)
- Noise floor: measurement method and full data (Japanese)
This is a personal research and information-sharing project, unaffiliated with any organization, university, or company. Data is published free of charge under CC BY 4.0 and is not for sale.