It is a simple but fascinating question.

To answer it, I spent some time yesterday messing around with the new Notion AI features. I was particularly interested in this newly added nugget:

I imported a few excerpts from several writers with distinct styles, including yours truly. Before delving into my results, a quick legal note is in order.

In doing these experiments, I wasn’t worried about violating fair use or engaging in self-plagiarism à la John Fogerty. (Yeah, that really happened.) My own blog posts are fair game. If you upload anyone’s entire book whose copyright belongs to a traditional publisher, you may need a lawyer one day.

Example #1: Impenetrable Academic Writing

I fed Notion AI a four-paragraph excerpt from the horribly written book I describe in this post. (Thank you, Shottr.) I then prompted it to write a 400-word post about careers emulating that author.

Here’s the result. (I’m including it as an image because I didn’t do the writing.)

Click to embiggen.

As instructed (and that’s the operative phrase), Notion accurately mimics Author X’s clunky prose and overly complex sentence construction in its AI-generated post.

You should never write in such a soporific manner.

Pro Tip

As an aside, you should never write in such a soporific manner. Notion’s excerpt is rife with passive voice, long sentences, use of leverage and synergistic, and other literary crimes against humanity. But I digress …

Example #2: Writing Like Me

When it comes to writing, one size has never fit all—not even close. To that end, consider the following questions:

• What would happen if I fed Notion AI a very different (read: actually coherent) chunk of my words?
• Could it capture the essence of my writing from that block?
• If not, could it at least come close?

With generative AI, outputs should reflect the inputs. Even with a single blog post as an input, I should see a markedly different result than the morass of text in the first example. After all, the two writing styles are night and day.

Feeding the Beast

I fed Notion this 464-word post of mine from August:

The Curiosity Dividend

I then prompted it to generate some career-related written advice based on this input. Below is the first part of its output—again, as an image:

Click to embiggen.

AI Pretending to Write Like Me: Initial Thoughts

My immediate reaction: Meh.

Notion AI kept its promise above. It again generated author-specific text output. My AI doppelganger doesn’t sound like an academic blowhard. Point: Notion.

Next, I immediately noticed the heavy use of the word curiosity and blatant references to some of the specific words I used in my original, human-generated post. The date-related headers above aren’t coincidences, either. AI sensed a pattern from my post on curiosity and, for better or worse, replicated it.

See the Scott Adams’s reference? Its inclusion led me to believe that AI seemingly picked up on my quasi-unhealthy quotation fetish. Maybe it even knew that I wrote a blog post about Dilbert a decade ago. (I can’t get under the hood of large language models, but that’s a post for another day.)

Notion AI didn’t bat 1.000, though. In its attempt to ape me, it put way too many words in quotes. Experienced scribes use them sparingly, opting instead for italics when accentuating specific words. Not surprisingly, Notion AI also hallucinated in some of the tests I conducted in separate experiments.

Wrapping Up

Notion AI’s attempt to write like me was synthetic.¹ It reminds me of the scene from the 1986 remake of The Fly. In it, Jeff Goldblum’s character teleports a steak across the room and asks Geena Davis’s character to taste it.

Finally, I’d bet that Notion AI would have earned higher marks if I fed it 500 or so of the 1,500 posts on my site. I just didn’t feel like putting the time in.

What You Need to Know

Go play with genAI writing tools beyond Notion’s Clippy-inspired effort. I suspect that your reaction will resemble mine. Yeah, you could eat the teleported meat, but it’s nowhere near as good, clean, original, accurate, and compelling as the real thing. As I wrote in The Nine, there are plenty of ways for all sorts of people to use AI. Experiment away, but I don’t rely on its as-is output for blog posts, never mind books.

What say you?

1. But not in the sense that The New York Times describes here.