I’m not making that disclosure because I’m a solo investigative journalist and generative artificial intelligence (gen AI) makes it easier for journalists like me, or even micro-newsrooms, to report the news.
While generative AI can speed up the writing process, that’s the least important reason I use it. For The Big Drop, its real value is in finding data, analyzing it, and continuously updating news reports with that data in ways that were impractical, or simply impossible, before.
This one-person newsroom covers the collision of two energy buildouts — the natural-gas pipeline network, that is infrastructure, and the data centers now demanding its output.
Before 7 a.m. today, this site re-counted roughly 224,000 industrial-control devices that can be accessed on open ports, particularly Modbus port 502, on U.S. networks, re-read and re-categorized more than 340 eminent-domain court cases, pulled the newest public comments from federal energy dockets, and re-scored the day’s exposure meter against a live measurement rather than yesterday’s. A newsroom of one cannot do that work once. The Big Drop does it every morning.
The scale problem in covering critical infrastructure is not a shortage of information: It is the abundance of information.
The records are largely public. They include court dockets, regulatory filings, corporate disclosures and permitting records. But they’re in separate databases — one for the courts, one for the regulators, one for the companies — with no shared index and no way to search across them. With generative AI, they can be connected if the reporter is enough of a subject matter expert in the sector and knows the right questions to ask. That is the expertise I bring, and The Big Drop gives you the real deal about what’s going on.
No single reporter could analyze all of it without help, even a reporter who was also a data scientist. That is where generative artificial intelligence becomes consequential.
Generative AI can identify what changed. It can sweep venues a solo journalist could not practically monitor and surface the filings as they appear. It can analyze what changed, comparing thousands of pages or records for patterns that would otherwise consume days of staff time. Which company’s language begins appearing in a regulatory proceeding. Which jurisdiction saw a sudden increase in cases. Which measurement moved, and by how much.
But the third function is the most important, and that is constraint. The reporting system is designed so the model does not get the final word on the facts. Published figures must trace back to an underlying query, court record, regulatory document, or other defined source. Measurements are computed outside the prose-generation process. Software then checks the resulting draft against those measurements. If the language contradicts the underlying record, the draft fails before publication.
That last step is the point. Used this way, AI does not replace editorial judgment. It enforces it at a scale one person’s judgment could never otherwise reach.