Journal · 28 Jan 2026

Thai–English keyword collision

Notebook, laptop, and coffee during a planning meeting

Most catalogues in Thailand ship at least two metadata locales. The Thai listing carries the words people actually type. The English listing is often a leftover from a global template, a founder’s preference, or a hope that tourists and expats will search in Latin script. Those two locales do not politely share query share. They compete, sometimes with sibling titles in the same publisher account.

App Store Performance Analytics that averages locales into one conversion number will hide the fight. Page views can look stable while the Thai query you wanted is being eaten by an English subtitle that sounds clever in a slide and clumsy in the store.

The audit we run before a freeze

We print both locales side by side — subtitle, keyword field where it exists, short description, and the first three screenshot captions. We highlight exact and near overlaps, including transliterations of the same Thai term. Then we ask a rude question: which locale is allowed to own this query for the next six weeks?

If the answer is “both,” we treat that as a decision to cannibalize. Sometimes that is rational: a wallet might want “bill pay” in English and the Thai equivalent in Thai. More often it is inertia. English still says “all-in-one super app” while Thai has already specialized.

Sibling titles get the same treatment. Two casual games with near-identical English subtitles chasing “puzzle offline” will leak impression quality even if Thai metadata diverges. We saw this in a Catalogue Residency; the write-up is on the voices page.

What we do not do

We do not dump a keyword tool export into the field until it hits the character limit. We do not promise that differentiating subtitles will restore rank. We freeze the losing locale’s overlapping string, date the freeze, and watch page-view share for that cluster. If nothing moves in three weeks, we say so. Silence is data.

Stemming and Thai tokenization deserve respect. They also deserve not to become the whole class. In Keyword Collision Lab we keep a checklist an intern can run, then offer a slower recap for teams without NLP comfort. That recap exists because alumni asked for it, not because we enjoy linguistics theatre.

Where this sits in the studio

Module 03 of the ASO Measurement Studio produces a cannibalization note. The standalone Keyword Collision Lab is three weeks and narrower. The Store Signal Atlas keeps the bilingual risk page updated when consoles change labels.