For writers, two platforms these days stand out above the rest. Medium, and Substack. I’ve been enjoying Medium quite a lot and have been writing there in full swing since early 2022. (Crazy that it’s been that long, actually.)
In that time, I’ve joined the Medium Partner Program and am making decent money there.
But Substack intrigues me. Not as a direct Medium competitor. It’s not. But rather as a platform that lets writers grow an email newsletter around their favorite subjects. A great concept. And lots of writers have already found it to be an excellent place to publish.
Personally, I’ve focused more on Medium than on Substack, with a definite eye toward building on Substack in the near future. But Substack has made a change that greatly concerns me.
And it should concern you too.
What’s changing on Substack
Starting July 21, 2026, Substack is partnering with Pangram, a tool that is designed to evaluate text and determine whether it’s AI-generated or written by a human. By default, this is on for all Substack notes and all posts on the platform.
Chris Best explained the change in his post, “Against Claudefishing.”
Substack readers will now have the option, on any note or post, to click the three-dot menu and select “Scan for AI text.”
At that point, Pangram steps in and grades the content with a breakdown of the percentage it thinks is AI-generated, AI-assisted, or written by a human.
In order to work, Pangram needs at least 100 words, so a short little Substack note isn’t going to be successfully graded by Pangram, but a longer article absolutely can be.
The idea behind this change is to help readers better distinguish between AI slop and human-generated content on the web.
A worthy goal
To be clear, I very much appreciate the intent behind this. AI slop has overtaken the Internet in ways I think only a few people could have imagined a few years ago. Especially now that the vast majority of Internet traffic is actually from bots, we have a problem.
Personally, I want to read and enjoy content that I know was created by real, human authors. Not ChatGPT or Claude. I’m not interested in what a LLM thinks about a subject. I want to know what a real, flesh and blood human author thinks. Even if I disagree.
So I applaud the desire to make the divide between human-generated content and AI slop more visible. I do. In a world that has become flooded with AI-generated garbage, human content is growing more important every day. Content that was written and created by humans and for humans. Not by machines for humans or by machines for machines.
But there’s a big problem that I think Substack is ignoring, and we need to talk about it.
AI detection tools aren’t perfect
In a perfect world, a reader could select “Scan for AI text” and get an accurate, fair assessment of a block of text. Either AI slop or human-written.
But that’s not always what happens.
AI detection tools aren’t new. They’ve been around for several years now and have great usefulness. Especially in academia for teachers and professors trying to weed out AI-generated papers and assignments.
The problem? They’re often wrong.
Content that is entirely human can be and often is wrongly flagged as AI-written even though it was created entirely by a flesh and blood man or woman.
Potential damage to writers’ reputations
The bulk of my concern with this is with false accusations. Authors being wrongly accused of using AI when in fact they did not. When an author portrays his or her writing as human generated but Pangram through Substack proudly, incorrectly labels the text as AI or at minimum AI-assisted, that hurts his reputation.
It just does.
When I write on Substack and Medium, I never ever have AI write for me. Never have. Never will. I don’t understand or support those who let AI write for them. So you can imagine how I would feel if Pangram thought my content was written with AI.
Horrible. Disgusted.
My concern is that with this new “Scan for AI text” feature, readers are going to start scanning unsuspecting writers’ content for kicks only to have the system wrongly label that content as at least partially AI-generated, leading readers to suspect that these authors are taking the easy way out and feeding them AI slop instead of taking the time to write well-thought-out content that matters to their readers.
Even if Pangram is incorrect. Even if the writer used no AI whatsoever. The damage would already be done.
And that’s a problem.
A way to opt out
The good news is that right now, there is a way to opt out of allowing your posts to be scanned for AI text.
At the bottom of the Publish page, there is a button to “Disable AI detection.”
The bad news is this has to be done on a post by post basis. There is no way that I have found to turn it off across the board. The good news is it exists, at least.
Even so, I do have to wonder. If writers turn off AI detection on their content, will that just lead readers to be suspicious? Will that make people wonder if writers are hiding their use of AI if they don’t allow detection?
An unwelcome change with a pure motivation
Some have surmised that this change is the foot in the door for Substack to begin training AI with writers’ content. That’s a very fair and valid concern.
Personally, I don’t think that’s the reasoning behind this move. I say I don’t think it is because, obviously, I have no way of knowing for sure one way or the other. I will say, however, that I highly doubt that Substack would go to that extreme.
I’m certainly not endorsing or agreeing with this Pangram partnership. I think it’s a terrible move for writers and for the Substack platform as a whole. But I’m choosing to believe that Substack is doing it out of the pure desire to keep writing human. I can stand by that.
But that doesn’t mean this is the right direction.
The AI witch hunt we didn’t ask for
It’s been common in online discourse for people to jump to an accusation of using AI when addressing someone with whom he or she doesn’t agree.
“You sound like AI” or “Did AI write this?” have become the common retorts of the intellectually vacuous.
Look, to be clear, I’m absolutely against the practice of using AI to write on your behalf. If you didn’t write the words yourself, you didn’t write it. And to claim ChatGPT’s vomit as your own writing and post it on Substack or Medium or even your own site is tantamount to lying in my opinion.
I think AI-written and AI-generated content is slop and should be clearly labeled as such. Period.
But it’s sad that we’ve reached a point in the devolution of the Internet that we’re now quick to ask everybody and his dog if the content in front of us was AI generated. We’ve entered the phase where it’s easy to accuse anyone and everyone of using AI, even if no AI was used at all.
Substack integrating Pangram into the Substack platform only serves to fuel that witch hunt in my opinion. Now writers are at risk of being accused of using AI even if they used none simply because they use style elements that match the data sets LLMs were trained on.
The machine judging our humanity
It’s not lost on me that through all of this, real human work is being evaluated by AI models that were created by stealing years of human work in order to sound human. The creation is attempting to lecture its creators on humanity.
Poetic? Sure. Ironic? Most definitely.
The reality is, human writing predates AI by millennia, and humans have been using style elements like em dashes, idea grouping and the like for hundreds of years. If not longer. Those elements and features are no more proof of AI than they’re proof of purple gummy bears living on the Moon.
But somewhere along the line, we decided that these are infallible evidences of AI generated text and ran with it.
Now we’re to the point where we have to put up with tools like Pangram determining whether or not our content has a right to be labeled human or not. And that? That just doesn’t sit well with me.
At all.
A grim landmark for Substack
As I said, above, I choose to believe that Substack’s motivation here is pure. That they simply want to help promote human content and help people recognize and avoid AI slop. I can understand and get behind both of those goals. But integrating Pangram into Substack feels like the wrong direction.
It opens legitimate authors up to false accusations of using AI and it threatens the reputations of writers who want nothing to do with AI slop but who will inevitably be accused of producing it. I don’t know about anyone else, but that bothers me. Greatly.
In time, I think Pangram will continue to improve and perhaps be more accurate in its detections, but the underlying problems still remain. It’s still going to judge content based on stylistic choices that may have nothing whatsoever to do with AI and it’s still going to be used to incorrectly label content that’s 100% human as being slop.
That’s a problem. Will Substack acknowledge those problems? I hope. I’m not abandoning the platform, and I do plan to continue using it for the time being. But to say I’m concerned would be an understatement.