How I used a research agent to make sense of millions of World Cup conversations

The 2026 World Cup featured more teams, more brands and a larger audience than ever before. For anyone trying to understand how brands performed during the tournament, that scale created a new kind of challenge: not finding more data, but making sense of all of it.

The World Cup is one of the few moments where sponsors, non-partner brands, broadcasters, creators, athletes and fans are all competing for attention in the same cultural space.

Every match creates thousands of new conversations across social media, news and digital channels. Narratives appear, evolve and disappear quickly. A campaign can gain momentum overnight. A reputational issue can emerge within hours. An unexpected brand can suddenly become part of the story.

The challenge is, therefore, not simply the volume of data. It is understanding what matters within it. A top-down approach works when you already know what you are looking for. The World Cup often rewards the opposite approach: staying open enough for unexpected patterns, emerging narratives and opportunities to reveal themselves.

That's why I used the autonomous research agent Saga.

Instead of building another reporting dashboard with predefined metrics, I used the autonomous research agent to build a daily layer of intelligence over the tournament. Each day, it traced the narratives gaining momentum, the moments that changed the conversation and the signals pointing to new opportunities or emerging risks for brands. Over time, those daily snapshots revealed how the tournament evolved beyond the matches themselves.

Here's how I built the workflow behind the World Cup Intelligence Dispatch designed to surface the stories and signals every day that brands needed to pay attention to throughout the tournament.

1. Getting the data right

First and foremost, every insight depended on getting the boundaries of the conversation right.

So, I started by building a Boolean search around the World Cup in Pulsar TRAC. Alongside official tournament terms, I layered in keywords around brands, partners and sponsors, then kept refining the query as new campaigns, hashtags and unexpected brands started appearing. That meant the dataset evolved with the tournament, rather than being locked into what we expected to find on day one.

From there, I brought together around 4 million World Cup conversations from across social and news platforms including X, YouTube, Instagram, social video, Facebook, Threads, blogs, forums, podcasts, TV, Radio, online news and more. Starting before the opening match meant we could capture the anticipation as well as the action, revealing which brands were already shaping the conversation before a ball had even been kicked.

2. Developing the analysis frameworks

One of the most interesting things about using a research agent this way was how it changed my role in the process. I spent less time thinking about the mechanics of building the analysis and more time thinking about the questions I wanted to answer, the signals I wanted to follow and the relationships I wanted to understand.

Rather than manually constructing every analytical framework, I could focus on the research itself. Saga translated those questions into the right methodologies and analytical structures, making it easier to test ideas, refine my approach and explore different ways of looking at the data.

That came in particularly useful, when looking to capture as complex a dynamic as brand reputation. Reputation cannot be understood through a single metric like volume or sentiment alone.

To understand brand reputation throughout the tournament, Saga helped me apply Pulsar's Brand Reputation framework to build a single Reputational Health Score (RHS).

The score combined five weighted measures:

• Brand-directed sentiment
• Brand prominence within conversations
• Tier 1 outlet share
• Share of voice
• Conversation volume stability

Saga helped operationalize the framework by turning it into a comparable score out of 100 and creating different visualizations to track how brands were evolving over time. This made it easier to benchmark performance, identify changes in reputation and understand the factors driving movement in the score.

3. Interrogating the data and finding the story

Once the data and frameworks were in place, the challenge shifted from collecting information to deciding what was actually worth exploring.

Saga helped reduce the time and effort required to do the heavy lifting manually, but the strategic work was still in interpreting what the data was pointing toward.

So, the next step was interrogating the signals and narrowing down the stories behind them.

Rather than accepting the first insight that appeared, I used Saga to go deeper into the data: testing whether a pattern was consistent, understanding what might be driving it, identifying when it happened and exploring the context around it.

For example, if a brand suddenly saw a spike in conversation, the question was not simply "what happened?" It was:

• Was this driven by a campaign, a cultural moment or an external event?
• Did the conversation happen across one platform or multiple channels?
• Was the attention positive, negative or simply high volume?
• Was this a short-term spike or part of a wider trend?

This process helped move from individual data points into a clearer narrative: understanding not just what was happening, but why it mattered and what brands could potentially do about it.

4. Creating a reusable intelligence system

Finally, once the analysis was developed, I moved the outputs into the World Cup Intelligence dispatch built using Claude Code.

This became a standing brief that I could return to across different projects, providing a consistent starting point for new research questions and analysis.

Traditionally, a large part of research is spent gathering information, cleaning datasets and organizing evidence before you even reach the interesting questions.

By taking on more of the groundwork, Saga gave me more room to focus on the parts where 'human' judgment is most needed: asking better questions, exploring unexpected signals and deciding what was actually worth investigating.

Frequently asked questions

To stay up to date with our latest insights and releases, sign up to our newsletter below: