Substack has added a new tool to its app that lets users check whether a post, comment, or reply was written by a person or generated by AI. The tool, called Pangram, went live in July 2026 and works on any post, note, reply, or comment longer than 100 characters. (TechCrunch)

Pangram gives an estimate of how much of a piece was written by a human versus how much was generated by AI. It does not block or punish people for using AI. Instead, Substack says the goal is transparency. Writers can add an optional note to their posts saying they used AI, and the platform encourages creators to explain their process through a "how I make this" statement. (TechCrunch)

Writers keep control over how the tool works with their own content. They can run Pangram on their drafts before publishing to check the results themselves. If they think a scan on their work is wrong, they can report it and ask Substack to remove it. Substack CEO Chris Best called the feature "good use of AI." (TechCrunch; Substack)

The way Substack set this up matters. The platform is built on the idea of direct relationships between writers and readers. Putting AI detection in readers' hands means anyone can scan someone else's newsletter right inside the app and see an estimate of how much was human-written versus AI-generated. That is a choice that could affect trust between writers and their audiences.

Substack's decision not to penalize AI-assisted writing fits with how the platform has always handled content. It tends to favor disclosure and letting readers choose, rather than making strict rules. The Pangram tool follows that same thinking: it gives people information and lets them decide what to do with it.

How accurate Pangram turns out to be is still an open question. AI detection tools in general have a known problem with false positives, meaning they sometimes flag human-written text as AI-generated. This is especially common with writing by people who learned English as a second language. Substack lets writers contest scans they believe are wrong, but because readers can run scans without telling the writer, an incorrect result could shape someone's opinion of a post before the writer even knows about it. In my view, that imbalance, where readers can scan freely but writers have to actively find and fix false results, is the biggest tension in this new feature, and it will be worth watching how it plays out.

The 100-character minimum is a practical cutoff. Shorter pieces do not give the tool enough text to make a reliable judgment. Longer ones give Pangram enough to work with. The feature covers Substack's shorter social posts and replies, not just full-length newsletters.

The broader context is that AI-generated content has flooded publishing platforms over the past two years, and platforms are trying to figure out how to help readers understand what they are reading. Substack's approach — detection plus optional disclosure, with no bans — is one possible model. Whether it works over time depends on whether the detection stays accurate as AI writing tools keep improving, and whether readers actually change their habits based on what the tool tells them.

For writers who use AI in their work, whether for drafting, editing, research, or translation, this adds a new factor to think about. A reader scanning a post that used a lot of AI but was not disclosed could create an awkward situation the writer did not expect. On the other hand, writers who are upfront about their process may find that honesty itself builds trust in an environment where people are paying more attention to how content is made.

Substack's bet is that readers want this information and that writers will adjust to it being available. The platform is not saying whether using AI to write is good or bad. It is handing people a tool and letting the community figure out what to do with it.