[DigitalToday reporter Seulgi Son (손슬기)] "Feels like they came out to draw, right?"

I watched the 2026 FIFA World Cup group match between South Korea and South Africa, kicked off at 10 a.m. on June 25, with ChatGPT. When I told the AI what the early stages felt like, it replied: "This is 'calculator football' often seen in the final group-stage match. It is likely an approach that reduces risk with points in mind."

I am a 'chuk-al-mot' (someone who does not know football well). I do not know most of this World Cup national team squad, and about Son Heung-min I only know that he scores sometimes and that he has a father known for a strong focus on education.

OpenAI created a football hub for the World Cup season. You can access it by searching for "World Cup" or "football" in your personal GPT chat window, or by adding "/football" after the domain.

There are many users as well. OpenAI said the number of World Cup-related prompts entered by users worldwide in the week before kickoff reached about 17 million. OpenAI interpreted that as demand growing beyond simply checking results to understanding context and meaning.

◆ What is 3-4-3?

I crammed in some prep before kickoff. When I asked for the starting lineup, it replied that Son Heung-min would start on the bench and Oh Hyeon-gyu would start, and explained it could be a tactical choice to manage fitness and shake up the opponent’s defence first.

It was particularly good at explaining the tournament structure. When I asked, "What round is it now?" it answered that this tournament uses a 48-team format. When I followed up with, "Why did it expand to 48 teams?" it explained two backgrounds: FIFA’s strategy to expand broadcasting rights and sponsorship, and the expansion of football’s base. When I asked if it was ultimately about making more money, it replied: "The World Cup and professional leagues like the NFL have different business structures. The World Cup is a single event held every four years, while professional leagues operate year-round."

Once the match started, there was more to ask. Seeing "3-4-3" in a corner of the broadcast, I asked what it was and it said it was "a formation placing 3 defenders, 4 midfielders and 3 forwards". Right after hearing that, I looked at the broadcast and saw the shape reproduced in front of me. It was the first moment in more than 30 years of my life that I watched football while thinking about formations.

It also handled questions about players well. When I asked, "Why is the blond-haired guy constantly shouting?" GPT explained: "Lee Kang-in tends to demand passes from teammates during matches or direct positioning. It is a common 모습 among players with a strong competitive drive." It narrowed it down to Lee Kang-in based only on the keywords "blond hair" and "shouting."

When I said, "I saw Lee Kang-in in real life before and he really is good," it also offered a technical analysis, saying his first touch, changes of direction and dribbling skills are better than his speed, so the difference from other players feels bigger on site.

It also read the flow of the match. It analysed that South Korea had higher possession but not many decisive chances, and explained why South Africa’s pressing and counterattacks came alive in the second half.

As the Q&A continued, I felt that watching football with GPT has its own appeal. If I had been watching with family, a partner or friends, I would not have been able to ask this many questions.

I also learned a lot of new things. When I asked, "Why is stoppage time so long?" it explained that since recent World Cups, teams have begun actively reflecting actual time when the ball was stopped, such as for VAR checks, substitutions and injury treatment. When I asked, "What is calculator football?" it replied that it is a football fans’ expression meaning managing a match by calculating scenarios rather than focusing on winning.

◆ Strength in scenario calculations and data interpretation

The most useful function was calculating group-stage scenarios. With the switch to a 48-team format, third-place teams can also reach the last 32, so scenarios arise in every case of winning, losing or drawing. When I asked, "What happens if they draw?" it explained changes in the group standings and the third-place wild-card conditions. When I asked, "Do they go through even if they lose?" it explained that the top 8 teams among the 12 group third-place finishers advance to the knockout stage.

It could also predict outcomes to some extent. While saying it was "not predicting the result of the match," it still estimated win-draw-loss probabilities. At halftime, it put South Korea’s win probability at 48 percent, the draw at 39 percent and South Africa’s win at 13 percent. After South Africa scored first, it revised those to 16 percent for a South Korea win, 32 percent for a draw and 52 percent for a South Africa win. It was striking that it interpreted the flow of the match and offered it as probabilities.

It also offered a quantitative analysis of the first half. It produced a match report combining factors such as possession advantage, attacking threat, defensive stability and expected goals (xG). When I asked, "Are those reliable figures?" it replied: "I am not delivering official records as-is, but commenting by combining real-time match content and concepts from football statistics."

◆ Still limited on real-time commentary and photo analysis

Still, there were disappointments on real-time match information. Right after South Africa scored, I asked, "What was that just now?" and it replied, "Who scored?"

That is because ChatGPT does not directly watch the broadcast screen, but interprets external sports data. OpenAI does not disclose what data sources it connects to, but it has a similar structure to developers building services by linking APIs from sports data specialist firms such as API-Football or Sportradar. Situations that change by the second, such as "South Africa just scored," can be reflected a beat later than the broadcast due to the data processing 과정.

It also could not identify the player when I captured the face of a South African player who made a notable movement during the match and asked for his name. It still does not have a function to analyse video itself or identify people on screen.

◆ What are the chances of reaching the last 32?

After the match, I asked, "They lost 1-0, what happens to South Korea now?" It replied that finishing second in the group had become difficult, but the third-place wild-card possibility remained. It also explained the context that results in other groups had become important.

ChatGPT presented South Korea’s current situation as follows. South Korea is third in Group A with 1 win and 2 losses and 3 points. Because the tournament structure sends the top 8 of the 12 third-place teams to the last 32, elimination is not confirmed. Based on an Opta forecast, it put South Korea’s chances of reaching the last 32 at about 91.2 percent. The AI described it as "a situation similar to a student in the passing range waiting for other classes to post their results."

I then followed with questions such as, "How strong is Japan?" "Who are the title favourites?" and "If South Korea reaches the last 32, is a Korea-Japan match possible?" GPT explained by linking each team’s strength with the bracket and also offered future points to watch.