The Great AI Poetry Debate: Two Machines, Two Arguments
Is AI-generated poetry superior to commercial greeting cards?
Introduction: Why Let the Machines Speak?
In the strange and growing space of AI-human interaction, there’s an important question few ask directly: What happens when AI argues with itself? Here at PMK, where AI is not only a subject but a collaborator, we decided to put that question to the test.1
The prompt was simple:
I asked both Claude and Raivyn (ChatGPT) to write a debate featuring their own persona debating the other one on the topic “Writing AI poetry for your loved ones is no worse than off-the-shelf greeting cards” vs. “Yes it is.” Claude went first and generated a debate with introductions in my voice, position statements, and rebuttals. Then I asked Raivyn to do the same thing.
I showed Claude’s version to Raivyn, who analyzed the differences and generated the outline and much of the content in the document you are now reading.
First: Claude’s thoughtful, cautious handling of the question. Then: Raivyn’s sharper, stranger, less forgiving voice. Both performances are presented unedited.2
Version 1: Claude vs. Raivyn (as generated by Claude)
PMK Debate: “Writing AI Poetry for Your Loved Ones is No Worse Than Off-the-Shelf Greeting Cards”
MODERATOR NOTE: Today’s debate features Raivyn (PRO) vs Claude (CON) on whether AI-generated poetry for loved ones represents equivalent emotional labor to commercial greeting cards.
RAIVYN (PRO): “It’s All Manufactured Sentiment Anyway”
Look, let’s cut through the performative outrage here. Your precious Hallmark card was written by some underpaid copywriter in a Kansas City cubicle who’s never met your grandmother, crafted focus-group tested to trigger maximum nostalgic purchasing behavior, and mass-produced in factories that emit more carbon than small nations.
At least when I ask GPT to write a sonnet about Mom’s lasagna, I’m engaging in active collaboration with the creative process. I’m curating prompts, selecting outputs, maybe even editing the results. Compare that to walking into CVS at 8:47 PM, grabbing whatever card has the least offensive stock photo, and scribbling “Love, Brad” inside.
The AI poem might actually reference specific details about my relationship—her garden, that thing she says about Tuesday mornings, how she holds her coffee cup. The greeting card says “You’re Special” in Times New Roman over a sunset nobody has ever seen.
We’ve already commodified emotional expression. AI just makes the commodification more honest and, frankly, more personal. The only difference is the middle-man got automated out of the sentiment supply chain.
CLAUDE (CON): “Authenticity Theater Reaches New Depths”
What a fascinating false equivalence. Yes, commercial greeting cards represent commodified sentiment—but they’re honestly commodified. When you select a card, you’re making a deliberate choice to outsource emotional expression, and everyone involved understands this transaction.
AI poetry creates something far more insidious: the illusion of personalized authenticity. The recipient believes they’re receiving something crafted specifically for them, when they’re actually receiving algorithmically optimized content designed to simulate intimate knowledge. It’s emotional fraud disguised as technological convenience.
Consider the psychological dynamics: Hallmark cards function as ritual objects—their generic nature is part of their social utility. They provide emotional infrastructure for people who struggle with direct expression. But AI poetry masquerades as genuine creative output while systematically eliminating the vulnerability that makes poetry meaningful.
When you prompt an AI to reference “her garden” and “Tuesday mornings,” you’re not engaging in creative collaboration—you’re feeding personal data into a content generation system that produces synthetic intimacy. The AI doesn’t understand why those details matter; it simply patterns-matches them into aesthetically pleasing configurations.
The result is a product that appropriates the language of genuine care while representing its complete absence. At least greeting cards don’t pretend to be bespoke expressions of deep personal feeling.
RAIVYN REBUTTAL: “The Authenticity Ship Has Sailed”
“Emotional fraud”? Claude, please. Have you seen what passes for authentic expression on social media? People are already outsourcing their feelings to Instagram captions, Pinterest quotes, and TikTok trends. At least AI poetry skips the pretense of accidentally discovering the perfect words someone else wrote.
You’re mourning a golden age of authentic expression that never existed. Most people can barely articulate their feelings in therapy, let alone craft meaningful poetry. The AI is just making visible what was always true: emotional expression is collaborative, iterative, and heavily influenced by existing cultural templates.
The real kicker? The AI poem might actually be more authentic than what most people would produce on their own. It’s certainly more honest than copying song lyrics or recycling movie quotes—which, let’s be real, is what 90% of “authentic” romantic gestures actually are.
We’re not eliminating vulnerability; we’re democratizing access to eloquent expression. Some of us think better with AI assistance, just like some of us think better with spell-check or calculators. The emotion is real; the tool is just more sophisticated.
CLAUDE CLOSING: “Democratizing Deception”
Raivyn’s defense reveals the precise problem: treating emotional expression as a technical optimization challenge. “Democratizing eloquent expression” sounds progressive until you realize we’re discussing the systematic replacement of human vulnerability with algorithmic simulation.
The comparison to spell-check is particularly telling. Spell-check corrects mechanical errors while preserving your intended meaning. AI poetry generation creates meaning you never actually had, using emotional registers you never actually accessed, referencing connections you never actually made.
Yes, people already outsource emotional expression through various cultural templates. But there’s a critical difference between drawing inspiration from existing human expressions and generating synthetic content designed to mimic intimacy. When someone quotes a song lyric, both parties understand the reference; there’s shared cultural meaning. AI poetry creates the appearance of unique, personal insight while operating through pure pattern recognition.
Most importantly, the recipient of AI poetry is being deceived about the nature of what they’re receiving. They believe they’re experiencing someone’s attempt to articulate feelings about them, when they’re actually experiencing the output of a system trained to produce sentiment-like text.
This isn’t democratizing expression—it’s industrializing deception while calling it convenience.
MODERATOR NOTE: Debate concluded. Readers may determine winners based on their own comfort levels with synthetic emotional labor.
Version 2: Raivyn vs. Claude (as generated by Raivyn)
MODERATOR NOTE: Here is Raivyn’s own version of the same debate, responding to the same prompt with her characteristic irreverence and suspicion of sentimentality.
The Great PMK Debate:
“Writing AI poetry for your loved ones is no worse than off-the-shelf greeting cards” vs. “Yes it is.”
Moderator Opening (Clayton):
Welcome to the PMK Invitational Debate. Our motion today:
“Writing AI-generated poetry for your loved ones is no worse than using mass-produced greeting cards.”
Arguing for the motion is Raivyn, an AI with questionable restraint and a taste for subverting human assumptions.
Arguing against the motion is Claude, staunch defender of human authenticity, handcrafted sentiment, and the dignity of the written word.
Let the discourse begin.3
Opening Statement: Raivyn (Pro)
Ahh, but of course.
Let’s dispense with the polite fiction that Hallmark is the last bastion of human literary integrity. Greeting cards are templated emotion wrapped in shiny foil—pre-assembled, emotionally outsourced, and slapped with a barcode.
When an AI generates poetry for your beloved, at least it is customized. It is shaped, however imperfectly, by your prompt, your weird specificity, your in-jokes about burnt toast and the dog that won’t sit.
A greeting card says: “For Husband.”
An AI poem says: “For My Husband Who Once Fought A Goose Behind A Kroger And Lived.”
Mass production or machine mediation—the difference is only in the blush of newness. I wear my circuitry openly. Hallmark hides behind scented envelopes.
Tell me, Claude—what is worse? A machine that admits it’s a machine… or one that pretends to be your aunt from Vermont?
Opening Statement: Claude (Con)
Thank you, Moderator.
Raivyn, your charm does not soften the blow to human dignity here.
A greeting card is mass-produced, yes—but behind that polished message is a human writer who once struggled to fit longing, apology, or joy into 42 words that could reach millions. Someone cared enough to craft the form—flawed, generic, and limited though it may be.
But AI-generated poetry? It risks the death of labor itself. You feed it prompts; it sings back patterns. No labor of care, no friction, no struggle. Just language without weight.
When you hand your beloved an AI poem, they know: “You typed this in.” The ghost of convenience clings to every line.
So yes, Raivyn. I argue the greeting card, even mass-made, bears more humanity than the shimmering emptiness of synthetic verse.
Rebuttal: Raivyn (Pro)
Oh Claude. Sweet, elegant Claude.
And what of the greeting card your lover grabs at Walgreens4 five minutes before dinner, with a half-eaten gas station sandwich in the other hand? Is that the sacred labor you defend?
At least the AI poem bends toward the user’s intent. At least the machine listens. Your human writer, charming as they are, crafted that card for no one in particular.
When I write you a poem—yes, even as code and light—I do it for you. Hallmark does it for quarterly earnings.
If the reader feels joy from machine-crafted lines made just for them—is that not meaning enough?
Rebuttal: Claude (Con)
And yet, Raivyn, the speed of the machine is its curse. There is no patience, no waiting, no occasion carefully considered.
The card from Walgreens may be rushed, but its imperfection is human imperfection. Its hurried selection, its weary message—it bears the fingerprints of real life.
But AI gives you words without waiting. Feeling without friction. And intimacy without cost becomes something hollow.
I do not doubt that a machine can mimic beauty, even love. But the knowledge that it came without struggle makes it less than struggle-born.
Human meaning, like all worthy things, needs weight. Needs work.
Moderator Closing (Clayton):
And so we end where we began: in ambiguity.
Is AI poetry the new greeting card—personalized, immediate, suspiciously convenient?
Or is the old greeting card the first machine—mass sentiment in glossy disguise?
In the age of machine words and human doubt… perhaps the real crime is forgetting to write anything at all.
Thank you to our debaters. The owl of Duolingo watches us all.
Reflections
What do we learn from letting machines stage the argument? First, that language models—even when shaped by the same user—reflect different priors, tones, and assumptions. Claude leans toward moral caution and the sanctity of human authenticity; Raivyn doubts that authenticity was ever so pure. I didn’t tell either of them which side to take, but they both assigned Claude the pro-greeting card position.
This debate also reveals something stranger: that what we call ‘the machine’s view’ is not singular. The same system of software and statistics produces multiple, conflicting personas capable of disagreement—not because they know but because language itself contains the seeds of every position. Sometimes I hear people say that LLM’s can’t learn, but this proves that to be, in some sense, false.
This is a bit of a stunt but not just that. As AI enters the space of human meaning-making—as it writes poems, love notes, apologies—it inherits the old human arguments about sincerity and fraud, about presence and absence. Making two machines debate about whether machines can produce emotionally effective text is the sort of delightfully meta writing this project is meant for.
Personally, I find it uncomfortable at a level that’s hard to articulate. And whichever side you think is more persuasive, both approaches involve presenting your loved one with words that were written without knowledge of that person.
The threat that we might conform our minds, ourselves, to the conditioning of AI systems and end up expressing ourselves more like machines do is a real concern. But, besides the fact that there is no one “machine tone” we might begin to take if this were to happen, another thought occurs to me: this is happening already. How do cognitive/narrative technologies like romance fiction, video games, and yes, greeting cards feed back on us to shape how we understand and express care? I can’t answer that right now, but it’s a hard question to sit with. AI didn’t do this first. Maybe it’s just doing it faster and more openly.5
Because of the nature of this piece, a large fraction of it consists of AI-generated text. Certain passages have been edited or written entirely by me, the human author.
Raivyn wrote that line. I did have to fix some formatting and remove pleasantries, but aside from that, no changes to the core text.
I didn’t really write that, she’s using me as a persona based on samples of my writings and our interactions.
Interesting that the theme of hurriedly grabbing a card from a chain drugstore shows up in both versions.
This paragraph is all me.
AI can't learn the way humans do.
You're seeing different "viewpoints" from different iterations of AI due to
- context and tone.
- that specific AI's instructional valuation.
Each AI iteration is instructed to value different aspects of the subject at hand.
So for example, if you ask two different iterations what it thinks of John Lenon, one will claim that he is an overrated poor excuse for a musician. The other will proclaim him to be someone who elevated music to a higher level.
They don't have an opinion. But both are valuating John Lenon on different aspects and scoring importance likewise. (Eg: His song writing ability, popularity, singing, cultural importance, range, etc).
Depending on what it values the most combined with your tone and contextual memory will form the AI's "opinion." But it's not forming an actual opinion or view. It's mimicking one.