How Much of Substack Is Actually AI?
I analyzed thousands of posts from the top newsletters to find out how much of Substack is AI. Some of the biggest “writers” on Substack aren’t writing at all.
In March 2025 an alleged debate between Elon Musk and Keanu Reeves went viral on Substack. The post amassed 25,000 likes and nearly 5,000 reposts. Even today, it continues to spread. But the debate between Musk and Reeves never happened. The Substack post was entirely generated by AI.
Lately it feels like I’m encountering more and more AI-generated content on Substack. This mirrors patterns across the broader web. Last fall, media outlets declared that over 50% of all articles online were AI generated. According to some pundits, that number could rise to 90% by the end of this year. As a human writer, I think a lot about the prospect of being drowned out by a deluge of AI generated slop, and I began to wonder how much top content on Substack is really just AI.
Pangram is a tool that can detect AI generated text, even if it’s been “humanized“ or put through tools to evade AI detection. (They have a fantastic Chrome extension that allows anyone to check if the content they’re reading is AI.) I leveraged Pangram’s API to build a program that would analyze the 10 most recent posts from the top 25 Substack Bestsellers in every category, and reveal how much top content on Substack is AI generated.
The first thing I found is that the majority of Bestseller content on Substack is not AI. Two-thirds of top publications, 384 out of 575 newsletters, did not have a single trace of AI generated content in their recent archive, according to Pangram.
Certain Bestseller categories are much more likely to contain AI-generated content. Bestseller content in Technology had the most AI-generated writing. 28% of writing in top newsletters in Technology is fully or partially AI generated. That means that more than 1 in 4 posts in Technology have a substantial level of AI content. This is somewhat expected – people writing about technology usually love to leverage new technology, and some well-known tech writers on Substack have been open about using AI to draft their stories.
Other categories that contained high levels of AI writing were more concerning. 23% of top content in the Philosophy category and 22% of top content in the Health category is partially or fully AI generated. After that, the percentages drop precipitously. 13% of writing in Culture has been shaped by AI, and only 5% in sports, 3% in food and drink, and 1% in music, according to Pangram.
It seems like the more analytical the content is, the more it’s likely to be generated by AI. People don’t appear to want to get their cultural commentary, musings about restaurants or travel, or music reviews from AI, and these categories are heavily built around personal, human voices.
When it comes to technology, business, health, or philosophy, though, Substack readers do seem comfortable getting information from AI, or at least allowing AI to shape the information they’re consuming, whether consciously or not.
Another trend I discovered is that publications that are using AI use it on almost every post. There are some newsletters that are publishing 100% AI generated content, according to Pangram, seemingly with no human editing whatsoever. These newsletters are warping the broader category trends.
This is especially true in News, where one AI-generated newsletter makes up 35% of all majority-AI posts in the category. Culture and World Politics also had outliers. Just two large Culture publications account for roughly three-quarters of AI content in that category. Same for World Politics, where, if two major AI-generated newsletters were removed, the category mean would drop from 8.4% to just 2% AI content.
The success of these large AI-driven newsletters reveals that a handful of people appear to be profiting significantly from running high-volume AI-operated newsletters in otherwise very human-dominated categories. Just 29 Substack publications produced half of all majority-AI posts.
Overall, the top-ranked publications in each Bestseller categories use slightly less AI, but in Business, Art & Illustration, Parenting, and Sports, the top 10 publications are actually more likely to use AI than the rest of the leaderboard. In Business, the top 10 publishers average 14.4% AI, whereas publishers ranked 15 and below average 0.8% AI.
As I scrolled through the data, I was surprised that some categories contained either low or high amounts of AI-generated content. I used Pangram’s Chrome extension to easily spot-check posts from publications across categories. While some posts were clearly 100% AI generated and some were fully human, Pangram also identifies a mix of human and AI-authored content within a single post. Pangram lets you scroll through a post as it highlights what text is likely AI-generated or contains signatures of AI-generated writing. The tool also provides a graph of AI usage across the document.
There are some limitations to this analysis. Some top newsletters paywall a portion of their posts, so Pangram only processed the free preview text available. Pangram’s model requires 50+ words to determine outcome, and the model leans towards a “Human” outcome when it is unsure to preserve the writer in question. Substack’s full Bestseller lists contain 100 publications each, however, for the purpose of this research, my program only analyzed the top 25 publications in each category.
My feelings about what I discovered are mixed. Competing on Substack is hard. Out of the top 20 Substack Bestsellers in Technology there is only one woman: me. I spend hours on my newsletter every week. And so learning that a significant portion of my competition in Technology is leveraging LLMs has made me wonder if I should experiment more with automation. For example, using AI to auto-generate my link round ups at the end of my newsletter. (If I do this, I will disclose my AI use).
At the same time, the results were also validating. Clearly, there is still a strong desire for human-written content and analysis. The fact that two-thirds of Substack Bestsellers don’t appear to use AI at all, proves that there’s still value in producing human-written content and that human-written content can result in financial success. Substack seems to be staving off the broader slopification of the internet for the time being.
But, I am concerned about the group of high-volume, fully-AI publishers that are pulling entire categories in a direction that readers probably haven’t consciously chosen. These people seem to be using AI to fully automate newsletter production on a massive scale. And, based on the available paid subscription data, appear to be making a lot of money doing it.
The best advice I have for readers is to pay close attention to what you’re reading on Substack, especially if you’re consuming content about Technology, Philosophy, Health, or Business, where AI writing is most prevalent. Pangram can also help. Download their Chrome extension to easily check for AI-generated text as you scroll the web. It can show how much AI content is in your feed on LinkedIn, X, Reddit, or Substack. Using Pangram’s Chrome extension regularly has helped refine my own natural ability to identify AI generated text.
User Mag readers can get a free month of Pangram using the code: USERMAGPANGRAM
👉 The most effective way to fight back against AI slop is to support real, human creators! You can buy a paid subscription to my newsletter here. 👈
Thank you to Pangram for sponsoring this post.
What I’m reading
Dwarkesh Patel’s Podcast Lets You Eavesdrop on the A.I. Elite
Dwarkesh Patel was a bored college sophomore looking for intellectual stimulation. Now he commands interviews with Jensen Huang and Mark Zuckerberg and holds his own with deeply nerdy A.I. researchers. - NY Times
With A.I., Anyone Can Be an Influencer
TikTok and Instagram made it easy to monetize the physical self. Now the social-media-savvy can use A.I. to play with their identity, or overhaul it entirely. - New Yorker
Are on-screen relationships normalizing settling?
From Beef, to Materialists, to The Drama, a new wave of on-screen drama questions whether motivations for marriage can ever be wholly ‘pure’. - DAZED
Sergey Brin, his "MAGA Girlfriend" and a $57 million bid to kill the California billionaires tax.
The inside story of how a Google cofounder and a gut-health influencer ghosted Matt Mahan and confronted Gavin Newsom — even if Newsom is "handsome." - NY Times
What Happens If Trump Seizes AI Companies
The administration could exert much greater control over the industry—but just how far would it go? - The Atlantic
Democratic Governance of AI Is the Real Solution
The potential for catastrophic effects from the AI boom demands robust deliberation and real democratic governance. Localized initiatives like data center moratoria won’t get us there. - Jacobin
Techno-fascist fashion: Why Silicon Valley is moving into menswear
Palantir’s new ‘chore coat’ merch is just another way for the spy tech firm to worm its way inside our brains – but why is the tech industry now so obsessed with ‘taste’? - DAZED
The Men Who Want Their Foreskins Back
Some try to manually stretch their skin into place. Others are turning to experimental surgery. - The Cut
User Mag news rundown
Technology
560+ Google employees signed a letter urging CEO Pichai to reject military use of Google AI, warning it could enable mass surveillance.
Brands are releasing behind the scenes videos of their ads to prove they don’t use AI.
Tinder is making users scan their eyeballs for “proof of humanity.”
A cafe in Sweden went viral after claims that it is run entirely by Gemini AI, but it turns out it actually still needs humans.
A YouTuber built an anti-surveillance backpack to dodge America’s 82 million surveillance cameras.
Every new car in the U.S. will be required by law to have tech that puts constant surveillance on the driver by 2027.
Samsung’s first smart glasses, which are launching later this year, have leaked
A ‘hyperscale’ data center project in Utah is expected to generate and consume more power than the entire state.
Tech entrepreneur Ben Pasternak has been charged with felony strangulation of his girlfriend.
China says it will reverse a major AI acquisition by Meta.
A town of 7,000 planned so many data centers, it’s like adding 51 Walmarts.
Elon Musk and OpenAI CEO Sam Altman head to court in a high-stakes showdown over AI
Technologist Riley Walz made a time machine for Wikipedia
Canva apologized after its AI tool replaced the word ‘Palestine’ with ‘Ukraine’ in designs.
There’s a wall of shame for vibe coded apps, ranking the worst ones in terms of security practices.
Two college kids raise a $5.1 million to build an AI social network in iMessage.
Anthropic is paying up to $400,000 a year for an events role. They’re looking for someone to own the execution of brand experiences that translate Anthropic’s values into physical moments.
Silicon Valley execs are embracing a new breed of bodyguards after the Altman attack
Internet Culture
Gen Alpha boys prefer “AI girlfriends” over real ones.
Police are detaining Scientology speed runners as the viral trend ramps up.
Silicon Valley OnlyFans creator Aella, who organized the pro-Billionaire march in SF, has launched an AI doomer creator residency with Grimes as a mentor. They will pay influencers thousands of dollars to post content about how AI could destroy the world.
Hasan Piker is now the 3rd most-watched Twitch streamer in the world.
The Casino charity Duel paid $500k to have Andrew Tate stream blackjack, but the agency sent Bottom G, an Andrew Tate impersonator. They only realized when “Tate” started slut dancing and doing the splits on the table.
Matcha nails are the latest Gen Z beauty trend.
Footwear is getting more eccentric in what fashion insiders have dubbed “freaky sandal summer.”
A Polish influencer broke the world record for charity fundraising during a livestream, raising ~$69M for children’s cancer treatment over a nine-day stream.
Two Florida teens were arrested for driving lawnmower into Target & using leafblower in Culver’s to go viral
A man won $500k after TikTok inspired him to play the lottery.
Pop Culture
The $50 movie ticket has arrived.
There is a Rotten Tomatoes war happening right now over the Michael Jackson movie.
Madonna is using Grindr to promote her new album.
Cynthia Erivo completed the London Marathon as her iconic Wicked song played at the finish line.
Taylor Swift has filed a trademark on her voice and likeness to protect against AI.
The owner of Letterboxed is exploring a sale.
Politics
Everyone online seems to think the White House Correspondent’s Dinner shooting was staged
Democratic super PAC Priorities USA is launching an effort to make inroads with voters in “conservative-adjacent YouTube niches,” spending six figures to target voters watching sports, comedy podcasts, MAHA health/fitness content and more.
In leaked audio recordings, Peter Thiel suggested that the pope was a tool of the Antichrist.
Israel has hired a Trump surrogate to lead an AI influence campaign, getting pro-Israel messaging into AI products like ChatGPT
The USDA just signed a $300M deal with Palantir to consolidate the data of every farmer
FOIA documents show Palantir is helping the IRS target criminal investigations using “massive-scale” data mining across federal databases. This comes as the agency has reportedly pivoted toward targeting left-leaning political organizations
Gov. Janet Mills of Maine vetoed legislation that would have blocked new data centers in Maine until November 2027
Media
Ronan Farrow is on Substack now.
Poog, a podcast hosted by Jacqueline Novak and Kate Berlant, briefly had to rebrand after a dispute with their parent company, but now they’re back to the name Poog.
49% of adults say they mostly get news because they happen to come across it, not by actively seeking it out, an increase from 39% in 2019.
In the 1950s, the Iowa Writers Workshop received funding from the CIA to push a writing style that centered the emotional life of the individual while decentering the social and political forces that shape human behavior
Lifestyle & Health.
Pope Leo is trying to get the Catholic Church to focus less on sex.
Academics are studying what generates “food noise” aka relentless thoughts of food.
A third of Gen Z believe they’re psychic.
100 Italian chefs broke a world record with a 1,445-foot tiramisu
The weed killer Roundup, which has been linked to a host of horrific health issues, is being sprayed across every U.S. forest at record rates.
A man has been jailed after climbing the Las Vegas Sphere and telling police he’d get away with it.
For more great content follow my meme page, subscribe to my YouTube channel, and join my Instagram broadcast channel where I share links all week long.
Listen to my podcast Power User on Apple Podcasts:
Or follow Power User on Spotify:
I would be totally fine with the links section being AI generated. I don't think that would take away from it at all. Especially if that AI was trained on your content and reflected it.