The 100 most famous updates on Substack were scanned by the AI-detection business GPTZero to see if 25 to 30 of their most recent posts had AI-generated information. According to the report, 10 papers” considerably rely” on AI in some way, compared to seven that” significantly rely” on it in their published output. ( GPTZero paid for heavily paid subscriptions to Substack newsletters. ) Four newsletters that GPTZero identified as using AI thoroughly confirmed to WIRED that the writing process includes using artificial intelligence, whereas the other three did not respond to requests for opinion.
Some GPTZero newsletters, which were flagged to be AI-generated, concentrate on sharing investment reports and financial guidance. The analysis suggests that hundreds of thousands of people are currently regularly consuming AI-generated or AI-assisted content that they are specifically subscribing to read, even though no AI-detection service is perfect ( many, including GPTZero, can produce false positives ). In some cases, they’re also paying for it.
” It’s difficult not to be a little shocked”, says GPTZero director and CTO Alex Cui about the results. ” These are all important authors”. In a way that is comparable, Cui cited another research that GPTZero ran earlier this year on Wikipedia, which estimated that roughly one in 20 of the site’s papers were good AI-generated, which is roughly half the speed of the Substack content GPTZero looked at.
Not everyone is surprised by the frequency with which specific platforms use relational artificial intelligence. Makes” complete sense to me,” says Max Read, the author of the Read Max online magazine. He describes some financial information sources as” the somewhat upmarket version of hustle-culture YouTubers.”
Helen Tobin, Substack’s head of communications, declined to comment immediately on GPTZero’s results. ” We have several mechanisms in place to find and alleviate inauthentic or coordinated email actions, such as copypasta, record articles, SEO email, hacking, and app activity—many of which can include AI-generated content”, Tobin told WIRED in an email. ” However, we do n’t proactively monitor or remove content solely based on its AI origins, as there are numerous valid, constructive applications for AI-assisted content creation”.
A number of the Substack authors who spoke to WIRED spoke to asserted that they had used AI to polish their prose rather than writing entire posts by themselves. David Skilling, a sports agency CEO who runs the popular soccer newsletter Original Football ( over 630, 000 subscribers ), told WIRED he sees AI as a substitute editor. ” I proudly use modern tools for productivity in my businesses”, says Skilling. ” AI-detection tools may detect the use of AI, but there’s a huge difference between AI-generated and AI-assisted”.
Subham Panda, one of the writers of Spotlight by Xartup ( over 668, 000 subscribers ), which covers news about startups around the world, said that his team uses AI as an “assistive medium to help us curate high-quality content faster”. He emphasized that writers are accountable for the “details and summary” contained in their posts and that AI is primarily used to create images and to gather information for the newsletter.
Max Avery, a writer for the financial newsletter Strategic Wealth Briefing With Jake Claver ( with over 549, 000 subscribers ), claims to polish his rough drafts with Hemingway Editor Plus. He says the tools help him “get more work done on the content-creation front”.
Financial entrepreneur Josh Belanger says he similarly uses ChatGPT to streamline the writing process for his newsletter, Belanger Trading ( over 350, 000 subscribers ), and relies on the chatbot Claude to help him copyedit. ” I will write out my thoughts, research, things that I want included, and I will plug it in”, he says. Additionally, Belanger develops customized GPTs (versions of ChatGPT specifically designed for specific tasks ) to help polish technical writing that includes particular jargon, which he claims reduces the number of hallucinations the chatbot produces. ” For publishing in finance or trading, there are a lot of nuances … AI’s not going to know, so I need to prompt it”, he says.
Compared to some of its competitors, Substack appears to have a relatively low amount of AI-generated writing. For instance, two other AI-detection companies recently discovered that nearly 40 % of Medium blog posts were created using artificial intelligence tools. However, powerhouse accounts are publishing the AI writing on Substack, while a large portion of the suspected AI-generated content on Medium had little readership or engagement.
Some readers either do n’t notice or are unconcerned by the use of AI by authors they adore. According to GPTZero’s findings, many people are reading and enjoying AI-generated newsletters, and other authors may soon try to emulate their success by adopting the technology as well.
But that does n’t mean there wo n’t be backlash or pushback. A free “certified human” badge for bloggers is being released by GPTZero, which points to the future where ensuring that you do n’t use AI becomes a key selling point. In other creative sectors, this disclaimer is already common. The credits of the new A24 horror movie Heretic, for example, included a disclosure:” No generative AI was used in the making of this film”.
Similar badges and seals that state that creative works are entirely human may become widespread over the coming years. They appear unlikely to stop the steady flow of AI into the media and film industries, but they may give worried consumers a sense of moral choice.
