Kääselä

Kamogawa AI Search Lab — Kyoto

Is your business talked about — in your customers' language?

These days, when people want to know something, they ask AI instead of Google.
So which businesses does AI choose to mention, and how does it describe them?

Ask the same question in a different language, and both the sources AI draws on and what it says can change.
Kääselä tracks this every month, under the same conditions, in Japanese, English, Simplified Chinese, and Traditional Chinese — and records and analyzes the differences.

How does AI search cross the line between languages?

Latest Monthly Report
Monthly Report #000 — How Generative AI Answers About Kyoto Experiences Published 2026.08.02

12 months of continuous observation
Starting 2026.08

2026.082027.07

Published the first Sunday of every month.

01

Findings

For recommendation-style questions (e.g. "best kimono rental in Kyoto?"), the sources AI cites show three patterns.

  1. Insight 1

    When comparison media exists for a category and language, AI cites it — and the more commercially incentivized the category, the more comparison media matures and gets cited.

  2. Insight 2

    Where no comparison media is available, AI falls back on authoritative sources — official tourism sites, international travel media.

  3. Insight 3

    In categories with little comparison media or authoritative coverage, AI cites a business's own site directly — when that site states concrete facts about the business itself.

Leading citation source type / July 2026 observation
Category JapaneseEnglish SimplifiedTraditional
Tea Ceremony Public52.2% Media46.0%#2 gap: 6pt Self-run40.6%#2 gap: 3pt Self-run56.8%
Kimono Media53.7% Media49.1% Self-run54.8% Self-run66.7%
Zazen Public50.0% Self-run80.2% Self-run49.5% Self-run43.4%#2 gap: 3pt
Conditions
3 categories × 4 languages = 12 cells / n=20 per cell / 240 runs total
Threshold
Noise floor for aggregate metrics is ±10pt. Cells where the #2 gap falls within that (○) are not treated as a confirmed leader
Conditions of execution
ChatGPT Web / Temporary chat / Kyoto
Source
Monthly Report #000 / data/summary_by_cell.csv

A single-point, single-engine cross-sectional observation. Causation is not established.

Read more

Scope
ChatGPT Web only. I measure the actual consumer product experience, so I don't use the API.

02

Reading

Read more

03

About

New data, not new names.

Kääselä (also written Kaasela) is an independent research project that continuously records and analyzes — based on actual verification results — how generative AI recognizes and describes businesses.

GEO, LLMO, AIO.

New names for this field keep appearing. Kääselä isn't interested in the names.

What does AI draw on, and why does it mention a given business? What does a business need, in the age of AI, to be properly recognized by its customers?

I keep testing that question through measurement.

04

Contact

If you'd like to know more about the methodology, want to try measuring your own industry, or have thoughts on the findings, feel free to reach out.
kamogawa.ai.lab@gmail.com

I'll email you when the latest Monthly Report is published.

Privacy Policy

On paid work
I'm currently employed full-time, so I don't take on paid individual engagements.
For anything in line with the spirit of this research*, I'm happy to talk, at no cost. *Explaining the methodology and findings, discussing the AI search field, advising on how to run a similar study.
Affiliation
This is an independent research and publishing project. It isn't affiliated with any organization or university