I packed 67 words with every pattern AI detectors flag as machine writing — em dashes, "delve," parallel structure — and Pangram called it 100% possibly human. I've been writing that way for decades.

Tech Reader  ·  Opinion

Opinion  ·  October 31, 2026

I Write Like an AI. Pangram Says I'm Human. I've Been Writing This Way for Decades.

A 67-word experiment designed to look like AI — and the detector still called it human.

AI writing detectors have identified a set of telltale patterns that supposedly separate machine prose from human prose. Em dashes. The word "delve." Parallel constructions like "not X, but Y." I have been using all of them since before large language models existed. So I ran a test.

By Aaron Rose  ·  Tech Reader  ·  October 31, 2026

I have a problem. Not a big one. Not the kind that requires a support group or a strongly worded letter to my congressman. But a problem nonetheless.

I write like an AI.

Not because I am one. Not because I hand my drafts off and go make coffee while a model does the work. I write like an AI because, apparently, I have always written like an AI — long before AI existed to write like anything at all.

The em dash is my first offense. I love the em dash. Always have. It does something no comma can do and no period will do — it holds a thought at tension for exactly one beat before releasing it. I discovered this early in my career and never stopped. Now, in 2026, the em dash is Exhibit A in the case against machine-generated prose.

The parallel construction is my second offense. "Not X, but Y." "Not this — that." I use it constantly. It clarifies. It moves. It lands. Large language models use it constantly too. We apparently share the same taste in sentence rhythm. I am choosing to be flattered.

Then there is the word "delve." I use it. I have always used it. It is a perfectly good word that means precisely what I intend when I use it. The AI world has apparently run it into the ground, and now human writers are being told to scrub it from their vocabulary as a defensive measure.

I am not scrubbing anything.

The Experiment

I wrote 67 words and loaded them into Pangram's detection tool. The text was deliberately engineered to trip every wire. I packed in every pattern I had seen flagged as a machine fingerprint. It opened with "delved." It used the "if not X, then Y" construction twice. It substituted triple dashes for em dashes, since Pangram's text box does not render them cleanly.

Here is what I submitted:

In the confluence of the two rivers, we delved into the unknown. I might say our thinking went like this: if not X, then Y, if not A, then B. I thought deeply about this — too deeply. But then I thought back to my university studies with Professor Schultensteinnenbrennerenhagen at Koln. He would scoff at me while eating bon bons. He would laugh...and I would cry.

Pangram's verdict: "100% possibly human."

The patterns they claim prove AI authorship are so common in careful human writing that a detector cannot flag every signal I deliberately put into it.

What the Result Actually Means

Here is the thing nobody seems to want to say out loud. AI did not invent these patterns. It learned them. It learned them by reading millions of documents written by people who, like me, happened to write clearly, precisely, and with a preference for structure over rambling.

The AI wrote the way it did because writers like me wrote that way first.

So when a detector flags someone's work as machine-generated, it is not necessarily catching fraud. It may simply be catching a style that good writers developed long before any model existed to absorb it. The algorithm learned from us. The detector then turned around and used that learning to call us machines.

That is not a quality control system. That is irony with a software license.

I am not going to start writing badly to prove I am human. I am not going to retire my em dashes, ban "delve," or introduce deliberate typos so a probability engine feels better about my authenticity. My writing is my writing. It was mine before the models. It will be mine after them.

It is not AI.

It is just me —

Aaron Rose is the founder and publisher of Tech Reader Magazine, covering system architecture, cloud platforms, and AI policy.