Skip to main content

Canlah Research

How AI picks its answers, measured in the open

Canlah Research is the research arm of Canlah AI, a Singapore GEO agency. We publish studies, not opinion posts: raw data and rerunnable methods with every release, conflicts disclosed in place, losses published with the wins.

Publications

  • 2026-09-28 · Open dataset · method book · DOI

    Singapore GEO Citation Probes (2026-09) and the GEO Playbook

    110 Singapore buyer questions put to ChatGPT and Google AI Mode: the engines shared 19 of 507 cited URLs on the same question, and only 22 of 307 ChatGPT citations were in Google's top 10. Open data (CC BY 4.0), and the GEO Playbook built on it: a blueprint for each of 46 cited-page types.

  • 2026-08-30 · Case study · open data

    Same Question, Seven Engines

    One commercial question, seven search/answer engines, one night: no page cited by 5 or more engines, mean top-10 overlap 0.171 (CN) / 0.091 (EN). Per-engine source profiles, targeting hypotheses, and a logged-out ChatGPT replication.

  • 2026-08-21 · Open dataset · DOI

    Agent-Readiness of 50 Cross-Border DTC Brands (2026)

    25 of 47 reachable storefronts served a UCP manifest — and all 25 were issued by Shopify, none by the brand. Agent-readiness, today, is a platform switch, not a brand decision.

  • 2026-08-11 · Whitepaper · free to read

    Marketing in the Agent Era

    When AI becomes the first reader and the first buyer: 28 chapters, free to read in full, built on the original dataset above.

How we work

  • Questions come from practice: every study starts as something we actually hit while doing GEO work, not a content calendar.
  • Data before opinion: a release means raw data + probe specs + classification code, so every derived statistic can be recomputed without trusting us.
  • Limits in the body text: how small n is, what is hypothesis, what went unverified — you should meet a number's limits before you can quote it.
  • Conflicts disclosed in place: we are a GEO vendor measuring the market we participate in; our own brand's failed measurements are published with the wins.

Corrections, replications, collaboration: admin@canlah.ai