Semafor Uses AI to Extract 4,900 Claims from Its World Economy 2026 Conference

Semafor built an AI tool that parsed 4,900 distinct claims from more than 300 speakers at Semafor World Economy 2026, with every claim anchored to a specific quote in the transcripts. The event took place over five days on three simultaneous stages. Semafor Intelligence is a new AI-enabled editorial insight product built on Semafor's global convenings.

Reed Albergotti created a prototype using OpenAI’s Codex the Sunday morning after Semafor World Economy 2026 wrapped up, with the goal of identifying the central themes across conversations. Alastair Clements worked with Reed Albergotti over the next 36 hours to turn the prototype into a robust analytical tool.

The tool analyzed every transcript and pulled out every distinct claim each speaker made.

It turned each claim into a numerical fingerprint that captures meaning rather than wording. This technology is called “embedding” or “vectorizing”. The proximity map produced by embeddings was used to help refine the report.

The tool used multi-agent reasoning to surface direct quotes from speakers that support or push back on the central themes. Semafor’s journalists reviewed every theme, stress-testing the premises, interrogating the supporting quotes, and editing down to the ones most clearly supported by what was actually said. The report is a product of that editorial process.

Current AI systems aren’t capable of generating insights on their own more reliably than journalists. AI systems can allow building tools that expand the scope of what journalists can discover and analyze. The vector database runs on Google’s BigQuery.

The fingerprints were produced by an embedding model from Voyage, now owned by MongoDB. 7 models helped with text analysis. A second-pass ranker from Cohere helped surface the relevant evidence for each query.

The cluster map came out of an open-source library called UMAP that compressed 1024-dimension vectors into two-dimensional coordinates. The whole pipeline was wired together using Claude Code. The API calls and new database cost a few hundred dollars.

Reed Albergotti and Alastair Clements built the tool in a matter of days. Four years ago, an analysis like this would have required a data science team, weeks of scoping and implementation, and a six-figure budget.

"The tool surfaces patterns across thousands of claims that no journalist could hold in their head simultaneously," Alastair Clements added. The system only captured what was said on stage and missed behind-the-scenes conversations or discussions held under Chatham House rule. Semafor’s journalists augmented the readout with reporting that wasn’t captured in the transcript.

Semafor reported that the technology determined what was possible to surface while the journalists determined the framing and what was worth publishing. The tool doesn’t just help journalists work more efficiently, it extends and expands what can be done with proprietary data and notes. -centric framings.