Quartz's AI-Generated News Articles Raise Concerns Over Accuracy and Journalistic Integrity

Quartz Is Publishing AI-Generated Articles Based on Other AI Slop, Along With Warning They May Be Filled With Errors

G/O Media is once again attempting to use AI to churn out content -- while simultaneously warning that anything the AI produces might be littered with mistakes. As first reported by The Aftermath, the G/O Media-owned business news site Quartz is grinding out AI-generated articles under the byline "Quartz Intelligence Newsroom." Its articles, penned in bland and formulaic text, field the business news gamut; the site's AI previously only generated standardized corporate earnings updates, but recently began to include full-fledged news hits. At the foot of each article is a telling disclaimer disclosing the use of AI and warning of its likely pitfalls. "This is the first phase of an experimental new version of reporting," reads the disclaimer. "While we strive for accuracy and timeliness, due to the experimental nature of this technology we cannot guarantee that we'll always be successful in that regard." That disclaimer in mind, the AI indeed appears to have one significant blind spot in its news articles: the quality and accuracy of its citations, which are frequently mangled -- and, in some cases, appear to be regurgitated from even more AI slop. At the top of each article, "Quartz Intelligence Newsroom" lists the sources from which it compiles its news summaries (OpenAI's web-searching ChatGPT and Google's AI search "Overview" attempt to do the same.) In theory that's a good idea: sourcing is important reasons like transparency and giving credit for others' work. But on several occasions, as the Aftermath points out, the AI has failed to accurately untangle messy webs of aggregation, for example linking back simply to MSN or Yahoo in cases where those news aggregators simply circulated original reporting from other outlets like the Washington Post and the Wall Street Journal, among others. Skimming the AI-generated Quartz articles, we also noticed that the AI has repeatedly cited a website called "Devdiscourse," which itself strongly appears to be an automated news site churning out AI-generated news summaries. Consider an article published today by Quartz's AI, titled "Ozempic maker Novo Nordisk will cap insulin prices to settle a lawsuit."At the top of the article, the publication's "Intelligence Newsroom" cites three sources, starting with a link to a Devdiscourse article titled "Novo Nordisk Caps Insulin Prices in Landmark Settlement." Like Quartz's AI-generated articles, Devdiscourse's post doesn't list a human byline; instead, it's attributed to "Devdiscourse News Desk," a byline under which the website generates a dizzying amount of content. The post is brief, and doesn't hint at any original reporting. Even worse? It's outfitted with a laughably bad AI-concocted version of the pharmaceutical unicorn Novo Nordisk's logo, complete with incomprehensible gibberish. The company name is clearly misspelled as "NORDIISK," and under it is "STIAPLAME," an apparently made-up word. (Almost every Devdiscourse article appears to feature similarly garbled AI-generated images.) Devdiscourse claims to be a "media platform for development stakeholders" helmed by a self-avowed consulting company called "VisionRI." Overall, it seems likely that Devdiscourse is an SEO-focused content farm, working to increase visibility for VisionRI. The articles aren't disclosed as AI-generated. But Devdiscourse's LinkedIn page, which doesn't list any employees, says that the site is "designed and developed on the framework of machine learning technologies which facilitate smart content based on user interest, user browsing behaviours, location and skill sets." Its images, meanwhile, include an AI disclosure. To that end, though the hilariously terrible synthetic image seen in the Novo Nordisk article is disclosed as AI-generated, that it would be included at all hints at extremely limited human oversight over at Devdiscourse's content machine -- and, what's more, hints at a concerning lack of human oversight regarding the management of "Quartz Intelligence Desk," too. It's hard to believe that any legitimate human writer or editor would consider the "NORDIISK STIAPLAME" graphic acceptable to source in a serious news article. We reached out to G/O Media, but haven't received a response. The effort marks the latest of several attempts by G/O Media to use AI to create content for its various brands, each of which have similarly been marred by embarrassing inaccuracies and backlash from human staffers. There was that time the media company used AI to draft error-ridden "Star Wars" clickbait for Gizmodo's beloved io9 blog without notifying the publication's editors; its replacement of Gizmodo's Spanish-speaking staff with machine translation, which quickly mangled several articles; and the AV Club's AI articles that pretty much just copy-pasted material from IMBD pages. Per Aftermath, Quartz's human writers are frustrated and angered by their owner's latest push towards AI, and understandably so. On the one hand, the articles just aren't very good or engaging; coupled with the AI's inability to coherently and accurately source, it calls the priorities of G/O Media's leadership into question. In the face of pushback and frustration from its employees, G/O Media once again pushed ahead on its seemingly neverending AI efforts -- and in doing so has, once again, threatened to undermine the work and reliability of its award-winning websites and the human journalists who work there.

Quartz has been quietly publishing AI-generated news articles

Quartz, the international business news outlet, has been quietly aggregating reporting from other outlets, including TechCrunch, in order to publish AI-generated articles under the byline "Quartz Intelligence Newsroom." Quartz started publishing simple AI-generated earnings reports months ago, but beginning last week, the outlet moved on to short articles. One of the 18 AI-generated articles published as of Monday afternoon, titled "South Korea shares preliminary findings on Jeju Air crash investigation," aggregates reporting done by real journalists at CNN, MSN, and The Associated Press on MSN.com. Each of the outlet's AI-generated articles is roughly 400 words in length, and includes no full quotes from sources. Rather than attributing information in the body of the text, as flesh-and-blood journalists do, Quartz's AI writer only cites its sources at the very top of its pieces. A spokesperson for Quartz corporate parent G/O Media confirmed to TechCrunch the existence of a "purely experimental" AI newsroom, without commenting on which AI models or tools the publication uses to write AI-generated news articles. It is not clear how Quartz's AI newsroom chooses which stories to cover. The spokesperson said that the goal is to free up Quartz's editorial staff to "work on longer and more deeply reported articles," and that the editorial staff reviews each AI-generated story before it is published. The quality control seems to be lacking, however, going by one article that Quartz's AI newsroom sourced from TechCrunch last week. The article in question is a piece I wrote detailing how you can delete your Facebook, Instagram, and Threads accounts. For each platform, it provides step-by-step instructions on how to download and save your data before deleting it and, ultimately, your accounts. This was a weird article to turn into a 300-word AI-generated summary. The Quartz article's headline - "How to delete your Facebook, Instagram, and Threads right now" - hints at a how-to piece similar to mine. But its account deletion instructions are vague: To permanently delete a Facebook account, users must navigate to the "Settings & Privacy" section and select "Account Ownership and Control." It's important to note that once an account is deleted, it cannot be retrieved. For Instagram, users either use the Account Center or settings to download their data before deleting their profiles. Deleting Threads profiles requires removing the linked Instagram account, as the two are interconnected. I could probably spend all day critiquing the "AI-ness" of Quartz's AI newsroom articles. I mean, just look this headline: "Jobless claims rise slightly as continuing claims set a record." Word echo aside, the clause is a contradiction. Jobless claims are rising only "slightly," yet some other "continuing claims" are setting a record? Tsk, tsk. My editor would never let me publish something so sloppy. G/O Media, which is owned by private equity firm Great Hill Partners, came under fire in July 2023 for publishing error-filled AI-generated content without input from G/O's editors or writers. The company's editorial director at the time, Merrill Brown, defended the practice, even as journalists at G/O-owned outlets like Gizmodo objected to it. Publishing AI-generated content presents a way for publishers like Quartz to access cheap labor -- AI doesn't command benefits and a salary, after all -- while potentially maximizing profits. The G/O spokesperson said reader response to and engagement with its AI stories have "far exceeded our expectations to this point." The spokesperson also denied rumors of cash woes, saying that the company is "very well funded" with a "good amount of working capital to draw on if needed." They also noted that previous staff reductions were due to the sale of some sites in 2024, but that Quartz is in the process of hiring more editorial staff. G/O isn't the first media organization to dabble with AI-generated content. CNET and Gannett have published their own factually inaccurate AI-generated stories and art, and -- in the case of Sports Illustrated -- under fabricated bylines.