Substack Now Lets Any Reader Score Your Writing for AI. That Changes How You Use It.

Any reader who sees your Substack post can now run it through AI detection — one click, free, on anything over 100 words. Substack launched this on July 21, powered by a tool called Pangram, and it doesn't require a subscription or technical knowledge. What used to be an HR department's concern or a professor's headache is now a button any reader can press. If you've been publishing with heavy AI assistance — or if you're planning to — that changes the calculation.

What Substack actually launched

Pangram's model returns a percentage estimate — human versus AI-written — broken into sections. It's not a binary pass/fail. It highlights paragraphs. The patterns it catches are the ones that characterize AI output: hedging qualifiers stacked two deep, over-structured explanations that anticipate every possible misunderstanding, transitional phrases that exist to signal effort rather than move an argument forward.

Ethan Mollick, whose work on AI in professional settings is closely watched across business and academia, noted publicly this week that he's relieved he has always written his own posts rather than delegating drafts to AI. That's a meaningful signal from someone who uses AI tools daily for serious research work: he draws a clear line between AI as a thinking partner and AI as a writing stand-in.

The distinction Mollick is pointing at is the same one Pangram is trained to detect.

What to change, and what to keep

The problem isn't AI assistance. It's substitution.

Using AI to brainstorm before you write, to push back on a weak argument, to identify gaps in your reasoning — that produces human writing that reflects how you actually think. The AI shaped your preparation. You wrote the post.

Using AI to generate a first draft and editing lightly — that produces text that sounds like AI, because it is. You cleaned it up. The structure, the transitions, the hedges, the characteristic rhythms: those came from the model, and Pangram is tuned to find them.

The Monday morning version of this: if someone forwarded your last published piece to a reader who doesn't know you, would they hear your voice or a model's? If you're not sure, you have a detection problem and a more fundamental writing problem at the same time.

Danielle — the right way to use AI for writing

Danielle runs a 3,400-subscriber Substack covering operational finance for early-stage startups. She uses Claude on every post: she voice-notes her rough thinking, pastes the transcript, and asks Claude to find the weak points and push back on anything that doesn't hold. Then she writes from scratch, using Claude's questions as a prompt rather than its draft as a starting point.

Her posts score 88–94% human on Pangram, which she checked after seeing the Substack announcement. She's never asked Claude to write a section for her — not primarily because of detection risk, but because the posts that earn the most replies are the ones that read like how she actually talks to founders in a room. The AI is part of her preparation. She is the writer.

Ben — the harder problem

Ben is a senior project manager at a 200-person construction technology company who started publishing on LinkedIn in March to build a professional profile before a job search. He asked ChatGPT to draft most posts, editing for accuracy before publishing.

His posts score 71–82% AI on Pangram. He didn't find out through a reader — he ran his own posts through the tool after reading about the Substack launch. The detection risk wasn't the thing that stopped him. It was a different problem he noticed while checking: he couldn't remember the argument in a post from six weeks ago, because he hadn't constructed it. If a recruiter or a prospect asks him about something he published, he's working from the AI's version of his thinking, not his own. The content exists. The thinking isn't there.

The one thing

Substack put a scanner button in the hands of every reader. That's the news. The deeper thing it reflects is that content authenticity is becoming measurable and visible, and Substack won't be the last platform to surface it. Professionals who have built any kind of following through their writing should check that the writing is actually theirs — not because readers will run the scanner on every post, but because the thinking behind the writing is what the reputation is built on.

<BlogPracticeSection />