‘Automation software and machine learning’ to ‘breathe life into struggling local news industry,’ new SFist ownership claims


SFist — the snarky news-aggregation site launched in 2004 after its founders met at El Farolito on 24th and Alabama streets and finalized the deal over drinks at Doc’s Clock — is coming back.

This is a heartening development for both purveyors and consumers of local news, distraught after SFist and its fellow Gothamist websites were abruptly shuttered in 2017 by malevolent billionaire Joe Ricketts.

Less heartening, however, is the language used to describe SFist 2.0. Its new owners, Impress3 Media, seek “to revitalize local news through software-assisted reporting and writing,” per a press release put out today. From that statement:

Impress3’s has acquired the SFist assets to bolster its mission to breathe new life into the struggling local news industry. It uses automation software and machine learning to assist writers and editors as they produce the most interesting content for readers in specific geographic regions or business verticals.

So, this is an unmitigated good — provided you think what’s missing in the local news world is “automation software and machine learning.”

Not everyone does; veterans of sites already using such software describe the results as “Artificial Intelligence Mad-Libs.” Eve Batey, a co-founder of SFist, wished her successors well, but noted that “machine learning can be tremendously helpful with respect to data reporting. But I don’t think that alone is a solution for anything other than getting on Google News and getting some short-term clicks.”

Zack Chen, 30, is the CEO of Impress3 and a San Francisco native. He describes himself as a fan of SFist from way back when, who initiated this sale via Twitter. This deal closed around three months ago, he said. Chen claimed he was not permitted to explain why he was required to wait this much time to announce this transaction, nor was he allowed to discuss SFist’s sales price.

Chen said the purpose of the “automation” inherent in “our system” is not to “create content automatically. That would never be our goal.” Rather, he says, his system “makes sure our content creators find the most interesting topics faster than, generally, a human reader would be able to find them. It pulls from various websites and content online in order to help our editors and writers see things faster.”

Asked to clarify this, he continued “without saying too much, the purpose is not just to see what’s trending but to verify some of the content on its own so the process moves along faster. To cut out a little bit of the necessary steps that go into it — an attempt to speed up and streamline everything, so content writers will be able to move it along more quickly.”

This is a fairly loaded statement, as “the necessary steps” and “process” Chen is referring to are what we in the business refer to as “reporting.”


Asked for an example of how this has worked out, Chen pointed to a story on Impress3’s extant news site, Up to Boston, about bomb threats made to the Boston Globe. Chen says his system determined online claims regarding the bomb threats and evacuations to be reputable, enabling Up to Boston to publish a story some 18 hours prior to CNN.

This story, however, does not indicate that any proactive reporting took place. If a phone call or e-mail or any other message was made to the actual Boston Globe (or anyone), it is not indicated within this article.   

Chen describes Up to Boston as “our testing ground for local news” after several shots at crafting-industry magazines via software and machine learning. Up to Boston is “a good general idea of what we plan to do on SFist, but on a smaller scale.”  

A quick glance at Up to Boston reveals a very small staff aggregating local news. Video of a deadly local fire is gleaned from the Twitter feed of a TV reporter. Video of a family’s car, laden with holiday gifts, being driven off by thieves, is reposted from a local TV station. If there is any proactive reporting, it is hard to discern.

But SFist’s prior iteration wasn’t known for proactive reporting, either. Its articles often resembled book reports of other outlets’ proactively reported articles.

But they were funny.

“A second-hand approach to news coverage would not be out of line with the blogging model of Gothamist and all the ‘-ist’ sites were originally based on,” notes former SFist assistant editor Caleb Pershan. “But the idea of automation definitely strikes me as the opposite of what SFist was, which was an extremely online voice. Snark was occasionally deployed. I occasionally deployed snark to great effect. Snark is sometimes the order of the day.”

Pershan, Batey, and fellow former SFist editor Brock Keeling all said they hoped the site succeeded, but expressed various degrees of misgivings with its professed operating model. None had heard about this development prior to a cryptic Dec. 28 tweet from SFist’s long-dormant account; Chen mentioned nondisclosure agreements during his discussion with Mission Local.

Chen aims to have SFist up and running in a matter of months; he confirms that he has been in contact with former site editor Jay Barmann, but the size of the staff is “a work in progress.”

Barmann confirms he had his first discussion with Chen today, “but there’s nothing firm just yet.”

Chen hopes to have “more mixed media” on SFist than on the Boston site, with original podcasts and videos: “We’ll be sending someone out to do some original video production of a new restaurant or something going on in San Francisco.”

Will there also be original written reporting? “That’s the goal,” he continued. “But we have to build up to that.”

Also, he conceded, “that’s something that really falls outside the realm of our software assistance.”