“According to the articles I’ve read, AI has flawless spelling and grammar,” my wife volunteered a couple of months ago as she tidied up the kitchen. “Perhaps if you want people to believe that a human wrote their stuff, you should try misspelling a few words.”
The look that crossed my face could have earned me 25 years in a penitentiary. The thought that I might have to intentionally reduce the quality of my work to convince others of my humanity after years of honing my skills to a razor’s edge made me feel stabby. Understandably so.
Not to worry; I didn’t actually stab anyone—least of all, my beloved wife.
However, the anecdote reveals just how eager people are to determine authorship in the age of AI. And to do that, they’ll typically latch onto the first pattern they recognize. AI-generated content does have recognizable patterns. Fortunately (for me), flawless spelling isn’t the main one.
So What Makes It AI?
People have made all sorts of claims about the features of AI authorship. Some say it’s long sentences. Others say it’s short sentences. Some say that AI capitalizes hashtags for social media content. Others point out AI’s love for words like “leverage,” “delve,” “elevate,” and “unlock.” But, honestly, any human writer could check these same boxes.
Yet I can nearly always sense when a text has been generated by AI. How? Because, ultimately, it lacks a key component found in all great art and communication: intention.
Let me show you.
Thinking About Intention
Let’s say I wanted to write a poem that expresses the loneliness of working from home on a rainy day. I do it myself and come up with this:
Late September
The gray haze at my window
Approaches on the pitter-patter of soggy feet. Dog asleep, I’m
Sure the cat is preening somewhere. Just the clickety-clack,
Tap-tap of computer keys, I must write the words
Free of the abyssal echo that surrounds me. Today,
Socks on hardwood, hard chair at the back, I break
My soul into coffee-stained pieces for you to
Pick up and praise.
Now, let’s see how ChatGPT does it.
(Prompt: Write a poem that expresses the loneliness of working from home on a rainy day.)
Rain at the Window, Silence at the Desk
The rain taps softly, a distant friend,
Whispers on glass where the streetlights bend.
Grey shadows stretch in the afternoon’s fall,
I sit here waiting, but for what at all?
Keys click like echoes in an empty hall,
Words drift like mist, unspoken, small.
The glow of the screen, a cold, steady flame,
My thoughts drift away, they all feel the same.
A mug half-full, its warmth long gone,
The hours melt slow, the day drags on.
No footsteps above, no voices below,
Just the quiet hum of a world I don’t know.
The world outside is wet, alive,
But here I sit, as shadows dive
Into the spaces between my thoughts,
Tangled in tasks that time forgot.
I miss the hum of human breath,
The warmth that lingers close to death
Of this lonely day that will not end,
Where rain’s the only voice to send.
Now, ChatGPT wrote a perfectly serviceable poem. I even find some of its imagery delightful (e.g., Words drift like mist, unspoken, small). However, the style is juvenile, the rhyme scheme outdated, and the expression a bit predictable. While this poem may work for a greeting card or an email to a colleague, your kids won’t be analyzing it in English class next year. If anything, you’ll read it once and move on.
But why? I’d say it’s because the purpose of poetry is to express the depth of human feeling, and depth is what ChatGPT’s poem lacks. It has structure, rhyme, and even some metaphor—but very little abstraction. Compare I miss the hum of human breath to I break my soul into coffee-stained pieces. One makes you say, “Oh”; the other, “Woah.”
Why Intention Matters
Using AI to write music and poetry is like using a calculator to solve math problems. The calculator receives your input and spits out an answer. However, it doesn’t show you how it arrived at the answer. There’s no thought process to analyze, only a result that you’re intended to accept at face value.
But that thought process matters when we’re absorbing information, and it’s expressed through the intentionality of the author’s word choice and grammar. There’s a reason good math teachers require students to show their work on problems instead of handing them calculators. If you try hard enough, you can guess the answer to a math problem and accidentally get it right. But in that case, you lack both knowledge and intention. And that will become quite clear to others the moment you have to take an exam or apply the principles of problem solving to another situation.
Likewise, readers easily catch on to the lack of intention in written content. The text is too formulaic. Generic. Boring. The idioms and metaphors aren’t phrased quite right. The details seem vague, the conclusions weak (or nonexistent), the style slightly off. That’s because AI simply predicts which word should come next in the linguistic pattern for that topic—a pattern formed from all the good, bad, and virtually unreadable content created throughout history. It’s not actually thinking about the audience because it does no real thinking, and it falls short of expressing human empathy because it’s never felt anything, ever.
Whom Does the Content Serve?
A common complaint about AI-generated content is how unhelpful it is. “I’m so sick of AI articles on all these websites,” my wife fumed as she plugged away at her laptop last week. “They don’t give me any of the details I’m actually looking for.”
As someone who regularly researches niche subjects (like inspection turnarounds), I agree. AI-generated articles typically explain topics in a basic way, and they tend to be entirely devoid of insights and analysis from subject matter experts. In terms of depth, they usually rank somewhere around Yahoo Answers—(Remember Yahoo Answers?)—just summaries of info that a beginner might read. (Note: this is not the same as simplified content.)
In determining authorship, it helps to ask whom the content exists to serve. Is it the reader? Or is it the company looking to pad its website with strategically optimized keywords? These things matter.
Does the content feel helpful? Is it accessible? Would you tell other people about it? Is it something you’d read again if you needed a refresher on the topic? Would you quote it in your own content?
If the answer to all of these is “no”—and you’re reading how to “create solutions that will vault your esteemed business to the height of excellence”—AI was likely involved.
Prompt Quality (Sort of) Makes a Difference
Of course, the quality of AI-generated content depends, in large part, on the quality of the prompt that calls it into existence. Let’s return to the earlier poetry assignment and see if we can generate a better result.
Prompt: Try again to write a poem that expresses the loneliness of working from home on a rainy day. This time, write it in a more modern, non-rhyming style (like ee cummings) and with a darker, more cutting tone.
untitled (rain/work/loneliness)
rain presses its face
against the window,
smears grey across the glass.
i’m still in yesterday’s shirt,
coffee gone cold in the cup
(it’s always cold, isn’t it?)
the world is wet and loud,
but here—nothing.
just the clickclickclick of keys
that don’t speak back.
just the drone of the screen,
static light on my skin.
time doesn’t move,
it loops
stretches
breaks.
how many emails have i sent?
how many will i forget
before the day collapses
into another night
that feels like this one?
i think i heard a voice once
but maybe it was just the rain,
sliding down,
down,
down
the pane
like everything else
that’s out there.
inside,
there is only the chair,
the empty room,
the work
and me
Notice, this version clearly mirrors the earlier poem with the wet world, the clicking keys, and the coffee gone cold. Its imitation of ee cummings’ style is also a bit too on the nose. And its imagery displays the same (near zero) level of abstraction. But the tone is darker and more cutting, which puts it closer to the style of my poem. With additional prompting, I might get closer still.
Some writers and designers say that, with the right prompts, they can use AI to generate the kind of in-depth content they and their clients are seeking—which is how we got the term “prompt engineering.” Prompting for the outcome you want is a skill. You have to be able to describe what you want in specific terms and then adjust those prompts according to the AI’s limitations. You still have to understand audience, scope, style, and tone. And that probably still won’t get you the level of depth, abstraction, and SME insight that savvy readers crave. Most AI-enhanced copywriters confess they must edit the final output to add a human voice and make the piece interesting.
The problem is, too many companies looking to save marketing dollars aren’t using AI this way.
Psst! Need great intention for your marketing or internal comms? Let’s connect.
Leave a Reply