Medium CEO explains how AI is changing writing
- Deliberations over AI slop
- Writing as an act of thinking
- Validation, not money, as the incentive
- Pushing back
- Reconciling two facets of writing
- Legitimate uses of AI
- Conclusion
Deliberations over AI slop
Tony Stubblebine is the CEO of Medium.com, which is a popular platform for people to write content. Personally, I’ve never been a huge fan of Medium because it presents a paywall to read the content. Additionally, many Medium authors seem invisible to me as real people, so I’ve always been a bit suspicious of the platform. However, listening to Stubblebine on this episode changed my mind a bit. Stubblebine is smart and insightful, and some of his points resonated with me, especially how he sees writing as a tool for thinking, and the value in people sharing their life’s experiences — points I’ll touch upon later.
Stubblebine and Kantrowitz talk generally about AI slop and how sites like Medium can avoid promoting it, as well as how to avoid the general disincentivization of writing on the internet. Stubblebine’s main point about slop was that the existence of it isn’t the real issue, it’s whether readers see it. “Does it matter how much slop is on Medium,” he asked, “or does it matter… how much slop our readers are seeing which is the question I’m asking?” (3:42) This is an interesting distinction and shows the ways platforms can make AI slop less visible without enforcing draconian security measures up front.
Writing as an act of thinking
Stubblebine argues that writing is about thinking, and smart people like to think. So writing isn’t going away because of AI. As he put it, “I am always confused by… when people worry about the future of writing. So let me just start with first principles, right? Like writing is thinking and smart people like to think — that’s not going away” (10:24). As long as there are smart people who like to think, they’ll continue to write.
Stubblebine’s point about writing as a tool for thinking is one John Warner also drives home in his book More Than Words: How to think about writing in the Age of AI (a book we read in the AI book club, with a recording here). I agree 100% about writing as a mode of thinking, which is why I’ve sort of abandoned many attempts to use AI in writing blog posts. It’s not so much that AI-generated content in posts is soulless or too bland or too general; it’s that it can sometimes remove the thinking elements of the writing process. That’s the real value of writing — the clarity of mind and changed perspective that often results.
Kantrowitz agrees, saying “that idea of going from something that you think you know and then you start writing it and you realize you don’t really know it at all. Yeah, there’s great value in trying to connect those ideas on the page” (12:45). Writing forces you to sort and organize the thoughts in your mind in a way that requires clarity, and this often prompts you to review or re-investigate the subject to make sure you understand it. As technical writers, we know this. You can’t explain a complex technical concept if you don’t have a fairly good grasp of it yourself.
The whole discussion about writing as thinking reminds me of a conversation I had with Neil Perlin about 15 years ago at a conference. He told me he was planning to start a blog and that he had about 28 ideas already planned for posts, but after that, he worried that he wouldn’t have anything more to say. This struck me as ridiculous. As long as you’re actively thinking about your experiences, you’ll always have something to write about. The well never runs dry. And in fact Neil continued his blog for many years before retiring.
Validation, not money, as the incentive
Stubblebine argues that there’s great value in people sharing experiences from their lives, saying, “What I fell in love with was the idea that every single person is learning something just through the act of their life that’s worth sharing and that other people would get value from” (50:15). He doesn’t want AI companies to undermine that sharing. For most people, he says, the motivation to write was never for monetary reasons. “Most of the writers on the internet are not even doing it for money,” Stubblebine said. “They’re doing it for validation. And… if they don’t get readers, the sort of the point of it goes away from for them” (44:12).
I agree. Although I’ve had ads on my blog for at least a decade or longer, the revenue has only been pocket change. I’d still blog without any monetary incentives, for the same reason people write novels without doing it solely in hope of a future windfall.
This is the core of the incentive problem. Writers on Medium are sharing life lessons. “This thing happened to me and I want to share it with you” (28:54). Stubblebine sees great value in sharing your life’s experiences with others, and I completely agree. This has been the motive behind so much of my writing. Are AI sites threatening to upend that connection with readers?
Stubblebine says Google’s great invention with the Internet was to allow anyone to start writing about a topic and find readers. “The idea that you can just write something on the internet and traffic will just show up out of the blue,” he said, “That’s kind of like… Google made the public internet” (29:36). But now with AI chat, many of those readers never fully go to your site. He said the traffic trade-off is devastating, estimating that “we lose about a hundred clicks from traditional Google search for every one click we get back from a Gemini summary” (43:33).
As someone who’s been blogging for nearly 20 years, I owe a debt of gratitude to Google, as it’s allowed me to write and be discovered and gain readership for my content in ways that weren’t possible previously. Is that model going away? Will you find that you’re writing in a vacuum, writing for the machine to consume your content and spit it out in unrecognizable ways? That’s hardly incentivizing.
Pushing back
Stubblebine also argues that Medium provides so much content to AI companies, they have the potential to poison the well if they wanted. He made a fascinating claim about the em dash:
Stubblebine: “…we learned really early on that medium is a big enough corpus of data that we can poison the… results of any large language model… the Medium training… is so heavy on em dashes… so when you hear oh this must have been written by AI because it’s got so many em dashes it’s because the AI is trained on Medium…” (16:18, 16:57).
Kantrowitz: “Wait, this is why ChatGPT writes with em dashes?” (17:14).
I actually wasn’t that aware that AI has a strong preference for em dashes, but yeah, they appear semi-frequently. Is this because of an abundance of em dashes on Medium? I doubt it. Is Medium also taking credit for other aspects of AI writing, such as excessive amounts of adjectives, flowery and overblown descriptions, and inserting “crucial” and “delve” everywhere? Do all posts on Medium insist on titlecasing headings?
Reconciling two facets of writing
Despite the threats AI poses to readership, Stubblebine isn’t against using AI as a tool in the writing process. He sees AI as a “second brain” tool, similar to personal knowledge management tools. He described a new app concept they’re working on, where you can quickly get any content out of your bucket without spending time organizing it.
“…we’ve been using the word bucket for a while, which is just the idea of what does it take for you to feel safe, to think of it as a bucket that I can throw anything into it and I don’t have to spend time organizing it and I know I’ll be able to get it out when I need to” (23:19).
For example, in writing this post, I might want to quote from Warner’s book. I can’t remember those quotes, but a second-brain AI that holds everything I’ve ever written might be useful.
I like the idea of using AI tools as a second brain for writing, essentially assigning their role as a research assistant rather than the one steering and generating the content. I once tried putting my entire blog into a Gemini gem but quickly found that it’s difficult to do that. I have millions of words on this site from nearly 20 years of blogging. Gemini’s upload limits tap out at 100 MB or so, whether you’re uploading folders or pulling from GitHub.
Legitimate uses of AI
Stubblebine also welcomes other supporting uses of AI, such as for ensuring accuracy or for summarization. He gave his own example of using it as a research tool to “go find events in my area in the next two weeks” (26:58). But he acknowledged that there’s a gray area, noting that the human effort is still critical “in the prompt and in the fact-checking” (28:15).
There’s one especially good use of AI that few people realize: identifying and splicing in quotes. For this post, I initially described each of these points from memory on the same day I listened to the podcast. A few days later, Kantrowitz posted the interview on YouTube, which provides a transcript for the content. I then copied the transcript into Gemini and asked it to first verify that I had relayed all the points accurately in my draft.
But I also wanted to pepper in some quotes from the interview, essentially implementing a kind of diglossia technique in my writing. Weaving together multiple voices makes writing more interesting. AI does an excellent job at this, and I don’t think readers will feel that the post has lost some of its “soul” because I used AI to do this.
Conclusion
Overall, the podcast made me reflect on the strange division between blogging and writing documentation. For this personal blog, writing is a tool for thinking and so I don’t use AI tools as much here. I’m still experimenting and seeing what works, like what I described in the previous section, but nothing like my AI usage at work.
At work, writing tech docs seems to tap into some other non-creative area of the brain, where using AI to generate content is fine and even preferable. I mean, I shamelessly use AI tools for as many documentation tasks as possible. People who read docs aren’t interested in seeing a Descartes-like mind thinking through an idea. They want a clear description of what data an API field provides, for example. Accuracy is pretty much all that matters, not voice and a lively style.
In fact, many people think users won’t even read the docs directly but rather consume them through chat interfaces instead (an idea I wrote about here). If that’s the case, do you really want to be writing docs by hand to ensure they have voice or something?
Perhaps the Stubblebine podcast caught my attention so much because it once again highlights this strange dichotomy in my attitude toward AI, which is what I also focused on in my review of Warner’s book. At work, I’m a proponent of using AI to write. On my personal blog, I’m a proponent of using writing as a tool for thinking.
I’m still persuaded that we can leverage AI for more and more tasks that won’t replace the thinking mode of writing. Perhaps I haven’t explored this with enough thought.
About Tom Johnson
I'm an API technical writer based in the Seattle area. On this blog, I write about topics related to technical writing and communication — such as software documentation, API documentation, AI, information architecture, content strategy, writing processes, plain language, tech comm careers, and more. Check out my API documentation course if you're looking for more info about documenting APIs. Or see my posts on AI and AI course section for more on the latest in AI and tech comm.
If you're a technical writer and want to keep on top of the latest trends in the tech comm, be sure to subscribe to email updates below. You can also learn more about me or contact me. Finally, note that the opinions I express on my blog are my own points of view, not that of my employer.