Brand placement audit · World Cup highlights · internal

Who owned the goals?

You'd expect the brands paying FIFA the most to own the moments fans replay. In both World Cup finals we measured, they mostly did: FIFA's top-tier partners held 51% of the seconds around the goals in 2022 and 60% in 2026. The exception was a 2026 group game. In Brazil v Morocco, regional supporters, FIFA's lowest tier, held 30% of screen time, and three of them took 72% of goal-moment seconds.

60%of the 2026 final's goal-moment seconds went to top-tier partners
72%of Brazil v Morocco's goal-moment seconds went to regional supporters
≈ $0to measure all three games, on a laptop and inside Google's free tier

The finals went to the top tier. The group game didn't.

Top-tier partners' share of goal moments: 51% in the 2022 final, 28% in the group game, 60% in the 2026 final
FIFA Partner (top tier)World Cup SponsorRegional supporter
Share of sponsor brand-seconds (one frame sampled per second, perimeter boards only). Goal moments run from the build-up to the end of the replays. 2022 final: 638 brand-seconds, 194 of them in goal windows. 2026 group game: 491 and 103. 2026 final: 117 and 10.

With a third game, a final can be compared with a final. That is the fairer comparison: same stage, and both edits built around the decisive goals. On it, the top tier held on to the goal moments. Our first version of this page compared the 2022 final with the Brazil v Morocco group game and read the drop as a shift. The 2026 final doesn't support that. So far the group game is the outlier.

In every game, the cause is the same: which boards were behind the goal when the ball went in. Both Messi goals in 2022 came in front of Budweiser and Crypto.com, both World Cup Sponsors. In Brazil v Morocco, Vinícius's equaliser was framed for 80 seconds of build-up, celebration and replays by Kraken (a crypto exchange), Betano (a betting brand) and Inter Rapidísimo (a Colombian courier). In the 2026 final, the LED boards behind the goal switched from Hisense to Aramco about four seconds before Spain scored. Each game page shows this goal by goal.

Chinese brands' share of screen time was 24% in the 2022 final (Wanda, Vivo, Hisense, Mengniu), 12% in the group game (Lenovo, Hisense, Mengniu) and 19% in the 2026 final (Lenovo, Hisense). Both 2026 games are below 2022, in line with the drop in Chinese official sponsors from seven in 2018 to three in 2026 (Caixin).

This is three matches, with one highlight edit each. The 2026 final is a 2-minute edit with one goal, and its goal window holds only 10 detected brand-seconds. Which board is lit when a goal goes in is partly luck. Treat these numbers as what to measure across a tournament, not as a trend.

2022 final · Argentina v FranceThe banned beer brand got the most goal timeBudweiser: 11% of screen time, 19% of goal-moment seconds. The most of any brand. 2026 group stage · Brazil v MoroccoThree regional supporters got Brazil's goalKraken, Betano and Inter Rapidísimo: 16% of screen time, 72% of goal moments. 2026 final · Spain v ArgentinaThe boards switched to Aramco four seconds before the goalAramco: 12% of screen time, 60% of goal-moment seconds. Hisense had the build-up. 2022 round of 16 · Brazil v South Korea · full matchEvery Brazil goal had a different sponsor behind itFive goals, five sets of boards. Two Qatari brands held 1.3% of screen time and 21% of goal moments, all in one stoppage. 2018 group B · Portugal v Spain · full matchThe most-seen board got one second of goal timeQatar Airways had the most screen time of any brand and 1 second inside a goal window. Six goals, six different leaders. Cost at scale · 100,000 hours$0.26–$0.31 to audit an hour of videoApple's on-device text reader on a rented Mac, 12 to 15 minutes per hour, start to finish. Cloud APIs: $1.48 to $9.00. At 100,000 hours: $26k to $31k.

How well can cheap tools see sponsors?

Two free methods, checked by hand

We sample one frame per second. Each frame goes through two methods:

  • On-device OCR that reads board text: the full frame, zoomed tiles for distant boards, and a Chinese-language pass.
  • Google Video Intelligence's logo recognition, keeping only names on the sponsor list.

A brand counts as present in a second if either method finds it.

Good: recall ≥ 0.80, precision ≥ 0.95Fair: recall 0.50–0.79, precision 0.85–0.94Weak: recall < 0.50, precision < 0.85
Recall: the share of sponsor boards a person could identify that the method found. It comes from hand-labelling one frame every 10 seconds in the two longer videos and every 5 seconds in the 2-minute final: 72 frames with 76 boards in the 2022 final, 60 frames with 66 boards in the group game, and 24 frames with 34 boards in the 2026 final. The 95% range is in brackets.
Precision: the share of detected brand-seconds that were real, from a by-eye review of every detected appearance. The false ones are removed before any number on these pages.

The review caught errors a raw run would have published. Examples:

  • "UNITED STATES OF AMERICA" read as Bank of America.
  • Aramco's "Global Energy Partner" read as Globant.
  • "AIRWAYS" read as Lay's.
  • The player name "UPAMECANO" read as Betano.
  • "Powered by" on a Lenovo AI graphic read as Powerade.

It also caught one of ours. In the first version of these pages, ADI Predictstreet's boards were counted as Adidas. They are now a separate brand, and the earlier pages' numbers are kept in the benchmark log.

The biggest remaining gap is small, distant boards. We catch about 4 in 10 in the 2022 final, 6 in 10 in the group game, and 1 of 8 in the 2026 final.

0:01, 2022. Found: Qatar Airways, QatarEnergy, McDonald's and Vivo. Still missed: the tiny Visa, Coca-Cola and Adidas boards to the left. Watch on YouTube at 0:01 ↗
8:36, 2022. BYJU'S is real: a large board for 10 seconds, between QatarEnergy and Hisense. Nine sponsors in one frame. Watch on YouTube at 8:36 ↗
Per 10 minutes of videoTimeCostVerdict
On-device OCR (Apple Vision, three passes)≈10–13 min on an M4 laptop$0Keep: best recall, reads any board text
Google Video Intelligence logo recognition≈15–20 min, runs in the cloud$0 to 1,000 min/month, then $1.50Keep, filtered to the sponsor list: adds script and symbol logos
Hand review of detections≈20 min—Keep: it is what makes the numbers trustworthy
Google Cloud Vision logo detection (single frames)—$1.50 / 1,000 framesDropped: found 0.6% of 2,469 labelled logos in a soccer benchmark
OWLv2 open-vocabulary detector≈0.75 s / frame$0Dropped: finds logo-shaped areas (recall 0.47, precision 0.26) but cannot name the brand

Scope: perimeter and LED boards only. Kit and jersey logos (Nike, Puma, Adidas on shirts) and on-screen graphics (such as "Powered by Lenovo AI") are excluded from every figure and kept in a separate file, as are drinks coolers.