Wired has revealed that a request for an unusual method of payment was what gave the game away after it commissioned and published a piece by a “freelance journalist” going by Margaux Blanchard.
Press Gazette revealed on Thursday that Wired and Business Insider were among publications in the UK and US to have removed news features written by a journalist calling themselves Blanchard over concerns they appeared to be AI-generated works of fiction.
Wired removed the article, which was about couples getting married in online spaces like Minecraft but contained quotes from multiple people that do not appear to exist, in May. It has now said it has “taken steps to ensure this doesn’t happen again”.
Business Insider and art and culture title Cone Magazine removed articles with similar issues after Press Gazette contacted them for comment about Blanchard this week.
Naked Politics, a site for young people in the UK, and freedom of expression non-profit Index on Censorship also began investigating their own articles after Press Gazette got in touch.
Naked Politics has since taken theirs down, saying the article about children’s mental health services contained “likely false or unverifiable information from experts cited that were not able to be traced”.
Since the publication of our story on Thursday, a story written by Blanchard about Disneyland influencers for California-based, Hearst-owned news outlet SFGate has now also been taken down.
The link that previously hosted the SFGate story now contains an editor’s note stating: “In light of questions raised about this article, it has been removed pending an internal investigation.”
Original story: Wired and Business Insider remove ‘AI-written’ freelance articles
Wired: ‘We made errors here’
Wired responded to Press Gazette’s article with a post bylined “Wired management”, which explained that Blanchard’s pitch, about “the rise of hyper-niche internet weddings”, had “all the hallmarks of a great Wired story: a quirky internet subculture, fun visual potential, and a larger point about ‘love, community, and identity in an era where physical spaces often feel less ‘real’ than our digital ones.’”
“You couldn’t make a better Wired pitch if you built it in a lab,” the post said. “Or in this case, with the help of a large language model chatbot.”
There were “no alarms” during the edit process, it went on, with the writer responding to notes “promptly and amicably”.
But in the days after publication, Wired said, “it became clear that the writer was unable to provide enough information to be entered into our payments system. They instead insisted on payment by PayPal or check. Now suspicious, a Wired editor ran the story through two third-party AI-detection tools, both of which said that the copy was likely to be human-generated.
“A closer look at the details of the story, though, along with further correspondence from the writer, made it clear to us that the story had been an AI fabrication. After more due diligence from the head of our research desk, we retracted the story and replaced it with an editor’s note.
“We made errors here: This story did not go through a proper fact-check process or get a top edit from a more senior editor. First-time contributors to Wired should generally get both, and editors should always have full confidence that writers are who they say they are.”
Press Gazette understands that red flags have been raised in other instances by Blanchard asking to be paid in a non-standard way.
Editors now receive ‘a ton of AI slop’
Other journalists have also now spoken about being pitched by Blanchard.
A section editor for a well-known magazine that publishes long-form reportage, who asked to stay anonymous, told Press Gazette they received a pitch from Blanchard and ultimately decided it felt wrong.
They said the pitch was “strongly aligned with what we do” but at the same time was very general so they wrote back asking for the reporting plan and who the sources would be.
Blanchard’s responses were not much more refined but were parroting words and phrases the editor had just written.
Finally the editor turned down the pitch but explained in their email what they had been looking for, giving a hypothetical example to demonstrate what they meant.
Blanchard then responded saying she understood what was needed now, sharing a reporting plan and stating that she would speak to the fake example the editor had just shared – as well as several other people that a quick Google search showed did not exist.
The editor said they receive a “ton of AI slop stuff that’s super easy to detect” but some are more sophisticated, especially as they are often so specific and targeted to their magazine.
One writer sent a relevant and interesting pitch but claimed to have written for the likes of The Guardian and The Washington Post and could only send a link to their bylines on blog site Medium. When questioned, they claimed to have a “handshake deal with the Post for a story later this year,” the editor said.
Eventually, the editor added, the writer said they had already written it on spec and asked to send it over.
The editor said it was “well-written, definitely not the kind of thing that you could plug into ChatGPT or a reverse search engine and just automatically be like ‘oh right, this is AI dribble’.
“But it was absolute and utter bullshit.” The article was on a subject in which the editor is an expert and they immediately spotted several things that could not be true.
When they pointed it out, the author sent a second draft of their 3,000 word article within about three hours.
Asked what the giveaways are that something has been written by AI, the editor said there are often bullet points, a lot of “AI slop language, so the kind of thing that sounds more like what you’d read in website copy or advertising copy than journalism”, and vague sourcing.
But the difficulty, they added, is that it can be hard to distinguish between a mediocre pitch that might nonetheless have some potential, and an AI-written one.
“For me, though, the biggest thing, I think, for editors in assessing these is just, really, doing a little bit of extra due diligence. It’s maybe Googling a couple of the sources that they say they’re going to talk to and if stuff doesn’t stack up, walking away.”
“It’s getting harder and harder to find journalists who can do interesting reported stories, because there’s not a lot of money out there for it, and so there’s fewer freelancers nowadays who can really do the good stuff. And so when you get something that is so tailored and feels so exciting, it’s really easy to want to give people the benefit of the doubt.”
Editor says ‘I might have fallen’ for Margaux Blanchard once she had major bylines
A commenter named Mike wrote underneath Press Gazette’s previous article saying he was pitched by Blanchard and “thankfully I didn’t fall for it”.
“Her pitch seemed AI-generated,” he wrote. “I asked her for clips and she sent me two PDFs of scholarly articles with her name on them.
“However, as I began paging through one of them I noticed another surname at the top of the pages. I Googled the titles of both papers and found that they had been authored by other people and Margaux had simply pasted her name over their names on the first page. Naturally, I didn’t reply.
“Imagine my surprise when I saw this article! I wonder if I was one of her first marks – I consider myself lucky because if I had seen her bylines in legitimate publications like those discussed in this article, I might have fallen for it. (But the pitch wasn’t that good anyway.) I was also suspicious about her near-total lack of an online presence.”
Sophia Epstein, former magazine editor at tech publication Digital Frontier which stopped publishing in July, wrote on Linkedin that she had also rejected Blanchard and that she received many pitches that were clearly generated by AI.
“In some AI-generated cases it just took a quick Google to find the companies listed as examples didn’t exist, but these kind of human stories are so hard to vet in pitch form; I started asking writers for identification from their interviewees.”
Lou Covey, chief editor at Cyber Protection Magazine, wrote that she is sent “AI-written content all the time. It is vapid, empty and formulaic. Completely indistinguishable from content produced by mediocre writing. Don’t even need an AI checker to reject it.”
The post Wired says ‘AI-written’ story did not have ‘proper fact-check process’ appeared first on Press Gazette.
