We asked 673 football fans across ten countries what the World Cup means to them, then showed each of them three ways a brand might say it back. One message was the same everywhere. The other two were built for their country. What came back wasn't the answer the plan expected.
Can a brand run a single World Cup message worldwide, or does the story have to change at the border?
The brief was clean. The World Cup is the largest shared event on earth, around five billion people. So a global brand has a choice. Run one creative platform everywhere and trust that football means the same thing to everyone. Or build a different story for each country and accept the cost.
The plan had a favorite. It guessed that the World Cup is globally watched but locally interpreted, that shared attention does not mean shared meaning, and that brands fail when they assume it does. That guess pointed at localization.
To test it, every participant saw the same three-part menu, mid-conversation. Concept A was a single fixed global line, identical in all ten markets. Concept B was written from scratch for their country. Concept C tried to have it both ways. Pick a country below and watch B and C change while A holds perfectly still.
A stayed still. B and C moved by market. So any time the winning concept changes country to country, that movement is the localization signal itself.
Six hundred and seventy-three conversations. Here is how good they actually were, one square per person.
Every interview gets scored on three things, each out of five. Signal: did the person give us something real? Moderator: did the AI do its job? Usability: can we trust the record? The three combine, weighted 50/30/20, into one number. Color is the band. Hover any square for the full breakdown.
Names never appear. Each square is P## · Age · Gender · Market. Real names were stripped at ingest and replaced with stable IDs; quotes are redacted of any identifying detail. The few gray, dashed squares are no-shows and refusals, kept visible rather than quietly deleted.
The plan expected the local message to win. Here is what the room actually did, one chart at a time.
Localization isn't dead. It wins decisively in exactly two places: Brazil and Argentina, the markets where football identity is most sacred and most recent. There, the fixed global line feels thin next to a sentence that names the thing itself.
Counts among codeable answers. Japan (n=28) is a small base and directional. Notice the pattern: B (green) towers in Brazil and Argentina, collapses in the USA and Germany, where A (blue) runs away with it.
The plan thought brands fail when they assume shared meaning. The data says something sharper. Brands fail when they assert a local identity the audience doesn't actually feel. The same localization move reads as care in one mouth and as pandering in another. The divider is truth, not language.
And underneath all of it, the unity everyone reaches for first turns out to be conditional. Seventy-three percent open with "it brings people together." Then they qualify it: not when your team loses, not with the racism, not with the politics, not once the sponsors arrive.
All 673 interviews, held in their market columns. Color is the message they preferred. Switch the vertical axis to see the room a different way, and click any dot to hear them.
Each dot is one of the 673 interviews, colored by the message they preferred (gray = no single pick). Markets stay in columns. The default view stacks each column by concept, so you can read a market's whole A/B/C mix at a glance; switch the axis to spread the same people by age, engagement, or interview quality.
Five honest directions the data could support. Possibilities, not promises.
So you can judge whether the findings were led or earned. Tap to open.
AI-moderated conversational interviews on Dialogue AI, local language per market, around 12 to 15 minutes, with a mid-interview three-concept message test. Reading was three-pass: a stratified deep read of 60 interviews across all markets, tiers, and quality bands to build the codebook to saturation; a breadth pass applying that codebook across all 673; and saturation and drift checks. Code prevalence is a floor ("at least this many surfaced it"), never a survey incidence. Message preference is coded only where a single concept was named (64% of the sample); the rest are unclassifiable, not "no preference." Cross-sectional, engaged-fan sample (male- and youth-skewed). The fixed global line A has a built-in advantage in any cross-market comparison: it is the same polished sentence everywhere while B and C vary in execution. Read A's lead with that in mind.