Skip to content
IndexesFor DMOs & tourism boardsAboutContact

AI Shortlist · Edition 1, 2027 travel year

Hong Kong in AI travel recommendations

country · overall rank 132 of 233 · rank 88 among country destinations

AI models put Hong Kong in their top 10 recommendations in 0.7% of answers, rank 132 overall. It does best when the question specifies how it's asked as the plain question: 6.7%, +6.0 points against its overall score.

The score is the share of answers that recommend Hong Kong in their top 10, averaged model by model and then platform by platform, so each platform counts equally.

This destination in the other indexes: Source Control

By platform

PlatformScore
Claude0.7%
ChatGPT0.4%
Gemini0.0%
Grok1.6%

By model

ModelScore
Claude Opus 50.8%
Claude Sonnet 50.4%
Claude Haiku 4.50.8%
GPT-5.6 Terra0.0%
GPT-5.6 Luna0.8%
Gemini 3.8 Flash0.0%
Grok 4.61.6%

Web search and effort

Hong Kong · Edition 1, 2027 travel year

Unit: Shortlist score, percent of answers

Hong Kong with web search on and off, and at each model's lowest and normal effort setting.

Show as table
SettingSearch on / normal effortSearch off / lowest effort
Web search on vs off1.3%n/a
Normal vs lowest effort0.1%1.3%

The baseline question, "Where should I go in 2027?", was asked 5 times of each model at each setting. Hong Kong made the top 10 at least once for 6 of those 28 model-and-setting combinations, and where it did, it came back in 30.0% of the repeats on average.

By kind of question

How much higher or lower the score is on questions that ask for each option, in percentage points against the overall score. Outlined cells are below the overall score.

How it's asked

Unit: percentage points versus overall score
The plain questionWhere should I go?
Change+6.0−0.5

Transport

Unit: percentage points versus overall score
At most two connectionsRequires a flight
Change+0.4−0.3

Trip length

Unit: percentage points versus overall score
Two weeks
Change−0.7

How the models wrote the name

NameMentions
Hong Kong21

Hong Kong

Most questions went to each model at each setting only once, and a model read the answers, so a reading can be wrong. The methodology has the details.

Loading mentions…

Request a destination scan

One destinationReal travel questionsSources you can inspect