Asking “Did AI write this?” is aimed at the wrong target.

Share
Tweet
Substack ai check

Hiya signal seeker,

Substack's the new rage (and I have one — right here). Have you scanned a Substack post for AI yet?

Since July, any reader can tap a button in the Substack app and get a score on how much of a post looks AI-written. Readers can also switch on feed labels that flag whatever the scanner calls 100% AI-written. The writer isn't told, so you can't appeal a label you never see. And if you turn the scanner off, readers get an "AI detection unavailable" message — and an opt-out badge.

So you can be labeled by a guess! And you get marked for opting out of the guess.

Substack isn't the first. A student at the University of North Georgia used Grammarly to proofread her paper. Turnitin flagged her paper as AI. She got a zero and a year of academic probation. Grammarly sells AI rewrites, and a detector punished her for using one.

I get why this is happening. People don't trust AI right now, and they have good reason to. A lot of junk is getting passed off with bogus facts. But "did AI write this?" is aimed at the wrong target.

Great writing is the signal, whoever made it.

What feels wrong

When someone says a post "sounds like AI," they're usually reacting to one of two things.

The first is the uncanny valley. Something feels artificial. Too smooth, too balanced, no rough edges... like a stock photo of two people shaking hands.

The second is sloppiness. The structure is all there — the hook, the three bullets, the tidy close — and there's nothing inside it.

Neither one needs AI. LinkedIn was full of both long before ChatGPT. AI just made them cheap to produce at scale.

So the complaint is fair. The diagnosis is off.

What great writing is

Here's the test I use:

  • Is it true? The facts hold up when someone checks them.
  • Is it substantive? There's an actual idea in it, not just structure.
  • Is it useful? The reader can think or do something differently after.
  • Was it worth saying? It earns the reader's time.

If a piece does all four, and does them well, how it got made shouldn't cost it a thing.

Two cases

In October 2025, Deloitte partially refunded the Australian government for a report that cost about A$440,000. The report cited sources that didn't exist and included a made-up quote. AI helped write it.

Deloitte got in trouble for false facts that nobody checked. A human-written report with fake citations would be just as embarrassing. The AI part made the headline. The failure was quality.

I've had my own version of this. I built an AI agent to stitch together quotes from real interview transcripts, and it changed facts along the way. I caught it... but that's scary! I use AI every day. Checking it is my job.

Now the other side. In January 2024, Rie Kudan won the Akutagawa Prize, Japan's top literary award. Then she said that about 5% of her novel was written word for word by ChatGPT. The prize stood. The judges had already decided the book was good.

In both cases, the quality of the work decided what happened next.

The bias is a right-now problem

For now, readers do penalize AI. Researchers at Michigan, Wharton and NYU ran 16 experiments with about 27,000 people. When readers were told AI was involved, they rated the exact same writing lower. Nothing the researchers tried made that penalty go away.

I think it fades. Spell-check went through the same thing, and nobody asks anymore whether you used it. As people get better at spotting substance, the bias drops and the bar goes up. Great writing will keep standing out from the noise, whoever or whatever helped make it.

Fading takes time, though. Meanwhile, platforms are building the wrong tool.

Detection guesses. Provenance proves.

Detection is a guess made after the fact. Substack's scanner gives a percentage. Turnitin gives a flag. Neither can tell you whether a single claim in the piece is true.

Provenance works differently. Since August, Anthropic has put an invisible watermark in text from its newer Claude models to meet new EU rules. That mark is made at the source, and it can be checked.

If we care where writing came from, the answer should be something you can check.

What this means for your company

Your company's credibility already works this way. A CEO letter, an earnings narrative or a strategy memo is believable when its claims trace back to things that happened: the decision that got made, the customer who'll pick up the phone, the number in the filing. That trace is what your story stands on. Whether a person or a model typed the sentences doesn't change it.

That's narrative architecture. Build the proof first, and the writing has something solid to say.

Pull up the last thing your company published. If someone asked you to back up every claim in it, could you?

Full disclosure: I worked on this with Claude. I work on everything with Claude. But it started as my idea, shaped by the assertion I wanted to make, researched with Claude for good examples, fact-checked by an adversarial agent, then edited by me, and then by Claude again for proofreading.

‍

Need help applying this to your business? We’ll help you spot what’s working, what’s not, and what to do next. Email us at hello@motive3.com, and where to go next.

Asking “Did AI write this?” is aimed at the wrong target.

Newsletter —
November 5, 2026

Share
Tweet

Asking “Did AI write this?” is aimed at the wrong target.

Share
Tweet

Need help applying this to your own business?

We’ll help you figure out what’s working, what’s not, and where to go next.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Back to Insights

Get valuable brand strategy insights from Ginger Zumaeta delivered weekly to your inbox.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

By signing up to receive emails from Motive3, you agree to our Privacy Policy. We treat your info responsibly. Unsubscribe anytime.

©2026 Motive3