GoKawiilProject turns 45,000 Enron emails into AI-generated wiki with cited facts
A team fed 45,047 emails from the Enron corpus into an AI system that produced 214 wiki-style pages on companies, regulators and banks Enron dealt with, totaling 49,401 facts each linked to a source email. The pages use timelines so facts that stopped being true, such as merger terms, automatically expire on the date the emails show they changed. The builders say no human edited any page or fact directly.
GoKawiil's interpretation of the reporting above, not reported fact.
The project is a test of whether AI can reliably extract and organize structured, verifiable history from messy raw text at scale, which could matter for journalism, litigation discovery, and corporate archives. A second AI model audited a sample of 500 facts and reportedly found 93% fully supported and none wrong at their core, though that self-grading method and sample size leave open questions about broader accuracy across all 49,401 facts. The approach could offer a template for mining large email or document troves, but also raises concerns about trusting AI-generated claims without human review.
- 45,047 Enron emails were processed into 214 entity pages with 49,401 cited facts.
- Facts are time-stamped and 'expire' automatically when later emails show they changed.
- A second AI audit of 500 sampled facts found 93% fully supported and none wrong at their core, per the builders.
Source: thisandthat.chat, 2026-10-23
Published there as: “Show HN: We had AI write a wiki from 45k Enron emails, every fact cited”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.