A longtime San Francisco hyper-local news site is using generative AI for the vast majority of its recent, “originally reported” stories — without clear disclosures to the extent of its use.
A substantial chunk of content on Hoodline’s website in recent months — save for a handful of stories reported and written by a few longtime (human) contributors — appears to be produced by a synthetic text generator.
“This is straight-up AI spam,” Max Spero, the CEO of AI content detection firm Pangram Labs, told Gazetteer SF after reviewing a selection of stories published on Hoodline in March. “It is very obvious.”
Additionally, these stories are bylined by a rotating cast of five authors, most, if not all, of whose bylines appear to be AI-generated fabrications. Some of these writers have been published on the site since at least May 2023.
None of this is disclosed explicitly on the Hoodline website, although it has quietly added language to explain its AI use. A disclaimer published in late March states Hoodline uses AI “to support and enhance our editorial processes”; it also mentions the use of a “single professional pen name” — a pseudonym — in lieu of the “multiple names of the team members” responsible for the content.
And on April 9, after Gazetteer reached out with questions on its undisclosed use of AI, Hoodline published a letter from Zack Chen, the CEO of Hoodline parent company Impress3. In it, Chen justifies the use of artificial intelligence, noting that it was crucial to “build the traffic and revenue needed to employ more experienced, full-time journalists.”
The April 9 letter also notes a change: A small badge will now appear next to each byline, indicating that stories are AI-assisted. All of these updates have come only in the past three weeks, nearly a year after the first AI uses popped up on the Hoodline website.
In various turns over its decade or so of existence, Hoodline has served as a well-regarded source for on-the-ground neighborhood reporting in San Francisco and a tech-meets-journalism darling (with a cool $10 million round of VC funding arriving in 2018, to boot).
Nowadays, the most apparent tell that Hoodline is using AI for its content, frankly, is that the writing is off.
“AI content often struggles with getting the tone exactly right,” Spero said. “If I asked you to write a Wikipedia article, it would still editorialize and say, like, ‘so and so is an amazing drug’’ … with news, it's the same.”
Take a recent story about 68-year-old Chen Kun Lo, who was arrested by San Francisco police following his alleged involvement in a string of “blessing scams.”
“San Francisco's streets fell prey to an old con with a spiritual twist as police recently nabbed a 68-year-old man in connection with a series of ‘blessing scams,’” begins the story, “written” by Eileen Vargas. It’s wordy, littered with gratuitous descriptors and turns of phrase and, in Spero’s words, doesn’t “quite fit expected journalistic style.”
Another odd example comes in Hoodline’s coverage of former San Francisco utilities chief Harlan Kelly. “Kelly, convicted back in July last year, has been found guilty of swapping his influence for personal luxuries, a move that's scored him a ticket to the clink,” writes Leticia Ruiz. (“The clink” is used as a euphemism for jail or prison time a lot in Hoodline stories.)
A story about Mark Zuckerberg’s jiu-jitsu prowess, dated back to May 2023, was also AI-generated, Spero said.
By itself, it isn’t a shocker that a news site would use AI to produce stories using press releases or aggregated reporting. Other publications have turned to AI as a cost-cutting measure amid historic layoffs of journalists. One of the most outrageous examples is CNET, which used generative AI for search engine-friendly stories about personal finance that turned out to be error-laden. The website neglected to disclose its use, staffers alleged at the time.
But Hoodline has done something more deceptive: The five authors generating the vast majority of content on Hoodline’s website are not real people, even though they have been made to appear like human writers until recently.
From around May 2023 until this week, headshots have been shown on AI-generated author pages. At least two headshots — those of Vargas and Tony Ng — were produced by a generative image maker, according to the AI detection site AI or Not.
A third headshot for Nina Singh-Hudson was changed sometime in March to a more abstract image. Singh’s prior, more realistic headshot was artificially generated, per AI or Not. The two versions of Singh-Hudson look nothing alike.
The other two writers — Eric Tanaka and Leticia Ruiz — had images that were too low-resolution to be definitively determined as AI-generated. They are also “new” to Hoodline, having only published bylines since around January 2024.
Curiously, all five “writers” have been shown as people of color.
Now, these writers are part of what the website describes as an “in-house writing collaborative.” In Chen’s letter, he describes them as a composite of “junior-level journalists” using “various tools at their disposal to do things like reblog police press releases and create reliable, simple news stories.”
What that means, exactly, is not clear.
But that wasn’t the case when these author pages first appeared.
Up until late last year, too, Ng, Vargas and Singh-Hudson had human-sounding biographies. They were all removed sometime late last year as part of a website refresh, prior to the April update of the site.
“Born and raised in Sunnyvale, Eileen grew up in a blended family of Filipino and Mexican cultures, which made for a unique and flavorful childhood,” Vargas’ bio read. “Eileen writes about all things food, drink, and events.”
Ng’s read: “Queens by birth, Boston at heart, & Santa Clara in the flesh.” Meanwhile, Singh-Hudson’s biography sounds the most fake (and alliterative): “Nina is a long-time writer and a Bay Area Native who writes about good food & delicious drink, tantalizing tech & bustling business.”
Beyond the made-up, now-deleted bios and headshots, another key tell is that none of the writers have any other social media. Aside from author pages on the PR platform Muck Rack, none of these writers exist anywhere but on Hoodline. (A Muck Rack spokesperson explained to Gazetteer SF that many of the author pages published on its website are auto-scraped.)
In October 2020, Hoodline was purchased by Impress3, a San Francisco-based marketing firm that also owns local, voice-y news site SFist. (Chen, in his letter, emphasized that SFist does not use any artificial intelligence.)
At the time, Chen assured staff that Impress3 was “in a position to ensure that Hoodline lives on — understanding that parts of the local news industry have been hit rather hard lately.”
“From my side, we are going to be doing everything that we can to help neighborhood-style reporting thrive and expand across both Hoodline & SFist in the coming weeks & months,” Chen wrote to staff during the transition period, according to an internal message reviewed by Gazetteer.
But multiple ex-Hoodline staffers who were there during the transitional era expressed concern that their scrappy local news organization was, for the third time in four years (and during the pandemic, no less), changing hands to yet another tech company.
“There was just a progressively more and more extreme departure from its original vision and mission,” said one ex-staffer, who declined to be publicly identified due to concerns over professional repercussions.
The site was founded as a hyper-local Haight-Ashbury news site called Haighteration in 2010, then operated as a network of local news sites before merging under the broader Hoodline umbrella. In 2016, Hoodline was bought by local tech entrepreneur Razmig Hovaghimian, who bought the site and pivoted it into a more tech-forward direction.
There were two sides of this version of Hoodline: The hyper-local San Francisco news site, and a tech-y side that wanted its news creation tools to go national. Around this time, the company had started using a content automating tool called Wordsmith that had a Mad Libs-esque approach to generating stories. It was not AI-generated. A few years later, Nextdoor bought the site.
“It's really disappointing to track the trajectory of where Hoodline started, with all the energy and idealism we had, and where it is today,” Rose Garrett, Hoodline’s former managing editor, told Gazetteer SF. “I think it’s really a microcosm of the media landscape of the last 10 years. It's an incredibly hard business, but there's no substitute for local journalism.”
Another former Hoodline staffer, who also declined to be identified due to professional repercussions, emphasized that “every writer and editor I worked with really did care about the neighborhood they lived in, telling these very hyper-local stories.”
“I think I lost that sense after the acquisition,” they said.
Unlike the AI fiasco at CNET and elsewhere, nothing appears to be factually incorrect about Hoodline’s stories, raising at least two more questions: To what extent humans are involved in Hoodline’s current editorial process, and what sources does Hoodline rely on for stories?
Chen declined to respond to a list of specific questions about Hoodline’s practices, including if its team of “junior-level journalists” are local reporters and the extent to which AI has been used for its content.
According to its LinkedIn page, Impress3 has at least two employees based in the Philippines — one tasked with copy editing, another responsible for lead generation.
“You need to be transparent with your audience about where you got your information and how you generated it,” said Jared Schroeder, an associate professor of journalism at the University of Missouri. “So yeah, it feels wrong, doesn't it? It just feels wrong.”
A lack of transparency around AI use creates more mistrust, not just for the sites doing it, but for the news ecosystem more broadly, Spero said.
McKenzie Sadeghi, an editor specializing in AI at anti-misinformation startup NewsGuard, emphasized that trust has already declined in news media — and AI could make things tougher for real-life journalists.
“It makes it harder for the average reader to discern AI generated content from a computer versus actual human-produced news from journalists on the ground and this can cause disengagement or news fatigue,” she said.
More and more of these sites, Sadeghi anticipates, will pop up “presumably because it's cheaper and faster to produce content” with generative AI products.
And although he declined to answer detailed questions about the role of AI at Hoodline, it’s obvious Chen views it as the future, too.
As he asserted in his April 9 letter: “We believe that within the coming years, AI will have permeated the majority of the news-gathering industry, in one form or another.”
