This startup is automating investigative journalism. Is that a good thing?

What happened to investigative journalism?

Lots of things, none of them good. The conglomeration of publishers, revelations concerning Project Mockingbird, the re-legalization of government propaganda in the Obama era, drastically truncated attention spans, and the super-saturation of sheer information all came together to diminish its once-proud tradition.

Whistleblowers blew, the government failed to protect them, and a great many journalists (so-called) operated less as the brakes on power and more as easily purchased shills.

A new AI-integrated operation, however, seeks to put things back on track.

The turnaround time is already approaching lightning speed.

Effort.news is the creation of entrepreneur Brian Chau. The endeavor brings together high-power, customized artificial intelligence with an evolving human competency/responsibility component that signals the best of what’s possible in the era of AI integration.

Speaking to Blaze Media on the human component of his operation, Chau said, “When we hire, the single most important human skill is the ability to anticipate which stories, ideas, and issues will matter to potential audiences. That’s true across the publication. What findings are newsworthy? What headlines and prose read well? What will readers remember or infer about the graphs? When do we include information in prose vs. a table vs. a graph vs. a footnote? All of these are questions the AIs don’t answer well right now. I also don’t think it’s the type of skill that can be pulled out and separated. The people who are good at knowing what matters tend to be good writers, editors, and researchers.”

Hard-hitting coverage

Readers notice a few things immediately with Effort.news — first off, its leaders aren’t playing around in terms of applying those brakes on power. Recent investigations have focused varyingly on EU machinations in the Ceuta invasion-migration story, California’s race-based pregnancy funding, and the old-money families behind the “degrowth” meme.

Secondly, the investigations are not surface scrapes. They are going deep, isolating patterns in vast amounts of information (often intentionally obscure and hidden information) to cull what otherwise simply wouldn’t be done in either mainstream or, lacking the tools, grassroots journalism outfits.

RELATED: This app censors your phone in a stunning new way — and people are using it on purpose

Juliane Sonntag/Photothek/Getty Images

And, finally, they’re moving extremely fast. “Our complete production time used to be 4-10 days from first looking into a lead to finishing the completed article. Now it’s closer to 1-4 days,” says Chau. “I think the research techniques we use would just be impractical for human use.”

For example, Chau explains, “the main thing we did to uncover donation rows for the degrowth article was that we used AI to look up every single donor to Swiss Philanthropy Foundation and search the donor’s financials for whether they donated to P4NE. For context, SPF hides P4NE’s donations by mixing 106 different funds into one account. SPF had 249 donors in 2024, so depending on how thoroughly people searched each donor’s records and tried to validate matching financials, that table could take around a month at least.”

A glimpse of the future

So the product is high-level. The turnaround time is already approaching lightning speed and getting better. The key with AI, that it’s too late to turn back, appears to be finding the proper balance between human and machine. Chau may not have all the answers to how we get there as a whole, but he’s starting to fill in some blanks. No question, it will require immersion into the messy, and dangerous, process of trial and error, but it’s also going to require a return to standards and someone honest and/or determined enough to implement them.

Chau says Effort.news is still refining. “Our AI workflows are extremely customized, and our team updates them several times every day,” he says. “There’s a lot to optimize around cost and speed of retrieving data, mostly around cost and breadth of information covered. We also make a lot of qualitative improvements to how AI models present information to us. It’s still not clean enough to directly publish, but it made reading the AIs 95% more readable. We’ve improved the models enough to cut our production time by more than half on average, and I think we can cut our production time by half multiple times more.”

Come what may, Effort.news is just the beginning.