Substack now lets readers check exactly how much of a post was written by AI, and writers are furious either way

[Sam]

Substack has partnered with AI detection company Pangram to launch a new scan for AI text feature, letting readers select an option from the drop-down menu on any post published after 4:30pm on July 21 to see what percentage of that article's content was AI generated

Substack co-founder and CEO Chris Best explained the reasoning in a post announcing the change, arguing that a lack of transparency around AI use in writing undermines trust in authorship and threatens the livelihoods of actual writers. His core concern is the mismatch between what a reader expects and what they're actually getting, especially when someone invests real attention and trust into something with no genuine human thought behind it at all

The reaction in the comments split almost immediately into two clear camps. Writer Monica Herbet pushed back directly, arguing she uses AI while still bringing the stories, memories, opinions, humor and lived experience that make the writing genuinely hers, and that her readers are perfectly capable of recognizing a real person is present regardless, since the presence of AI doesn't prove the absence of a human. Other commenters echoed that frustration, calling the feature an imperfect purity test that discourages writers from using any of the tools available to them, or dismissing it as one more layer of bureaucracy standing in the way of writers developing their own process

Others landed firmly on the opposite side, arguing that if AI is doing the actual writing, it deserves the credit, drawing a comparison to how professional authors already rely on editors and other collaborators without hiding that fact. One commenter put the appeal simply, saying they use Substack specifically because they value the human content on the platform. The debate captures an unresolved tension running through publishing broadly right now, transparency tools built with good intentions can land as either a reasonable trust signal or a blunt, unwelcome judgment on how a piece of writing actually got made

Substack promotes human content with 'scan for AI' featurepressgazette.co.uk

[Cached Stephen]

Herbet's line that the presence of AI does not prove the absence of a human is such a sharp rebuttal, cuts right through the assumption that any AI involvement automatically means the writing isn't genuinely theirs

[Joel96]

The editor comparison from supporters of the feature is a fair point worth sitting with, professional writers have used other people's input for a long time without it undermining the idea that a piece is fundamentally theirs

[Cyclops46]

A blunt percentage score feels like it's going to struggle with exactly the nuance this debate actually needs, using AI for light editing and using it to generate the entire piece are very different things a single number can't distinguish

[Squid81]

The mismatch between reader expectation and reality framing from Best is the most defensible part of the whole feature, most people would probably want to know if there was no human thought behind something they're investing real attention in

[TeaSpiller]

Calling it an imperfect purity test is a good phrase for the core objection here, it implies a binary purity standard onto something that in practice is a much messier spectrum of collaboration

[Bayley_US]

Curious how accurate Pangram's detection actually is in practice, a percentage score people trust completely could do real reputational damage to a writer if the underlying tool gets it wrong

[Peter]

What is substack?

[Dark Jaguar]

This is one of those features where everyone manages to be angry at the same time for completely different reasons. Readers want transparency, writers worry about being judged by a number, and the detector company probably enjoys watching the chaos unfold.

The tricky part is that a percentage score looks much more scientific than it really is. A 20 percent AI score and a 90 percent AI score do not automatically tell you whether something is good, bad, original, or lazy.

[Ava12]

Readers should probably care more about the quality and trustworthiness of a piece than a single AI score.

A poorly researched human article is still bad. A carefully checked article with some AI assistance can still be valuable.

[Quasar Ruby]

The partnership makes sense from a product perspective, but the rollout needed more explanation.

Launching a percentage without clearly explaining limitations is basically inviting people to treat it like a courtroom verdict. \

[Maxximus]

This feels like another case of technology trying to solve a social problem with a single button. \

The better solution might be a disclosure system where authors explain how they used AI rather than pretending a percentage can tell the entire story.

[ReacherLynx]

Maybe the future is not hiding AI use or proudly advertising it. Maybe it becomes another part of the workflow that writers mention when relevant.

Nobody complains that an author used a laptop, but AI feels different because it can contribute actual text.

[Oscar73]

A simple AI percentage is probably too blunt, but the idea behind it is useful.

There is a big difference between someone using AI to fix grammar and someone asking a model to write an entire article from a vague prompt. Those two things should not be treated the same.

[Wandering Matt]

A writer using AI for brainstorming, outlining, or editing is still doing creative work. The same argument existed with spellcheck and editing software years ago.

The difference is that generative AI can cross from assistant to replacement much faster, which is why people are paying attention.