In an article, AI Assist versus AI-generated Poetry, dated 20/10/2024,* I had the following to say in the conclusion of a rather detailed piece:
‘First the poet needs to learn the rules before breaking them, ie the design requirements of particular poetic forms. Even the seemingly free Free Verse has some restrictions placed on it. How else would the poet be in a position to sift the chaff from the corn? AI has its uses, but it is no substitute for hands on experience. As such, AI as a teaching tool for poetry falls way short of expectations.
‘AI robs poets of their innate spontaneity and stunts the development of their own poetic voice, ie their distinctive turn of phrase and creative expression. To my mind, the only course of action left is to be a better poet than AI. Support your fellow poets, and ignore AI.’
Bamboozling AI
As those who have been following me for the past few years here at PoetrySoup would recall: I have made a point of it that my contests are not readily supported by AI-generated poetry. In other words, I purposely set out to bamboozle AI. Even more so with my new poetic form, InfraMetron (I’M poetry). Here is why:
Computers, even when trained to generate poetry, fundamentally lack intonation, emphasis, and the nuanced ebb and flow of human delivery—because they process language as a structured sequence rather than as an embodied experience.
Their default poetic output often prioritises formal elements like rhyme and metre, precisely because those can be mathematically mapped and predicted. But rhythm in poetry isn’t just about metrical precision—it’s about the organic interplay of pauses, breath, and emotional inflection. Humans instinctively adjust pacing, modulate tone, and lean into resonance, whereas AI-generated verse—without vocal dynamics or real-world sensory experience—remains comparatively flat.
Even when a computer constructs a poem with visual rhythm (like spacing and formatting), it still lacks the physical presence of voice—the subconscious choices we make when reading aloud, shaping sound to fit emotion rather than rigid rules. The ‘monotone’ effect is a direct result of AI approaching rhythm mechanically rather than experientially.
I’M Poetry & AI
With I’M poetry, the open-ended finale is where it stumbles—likely because AI models are hardwired to seek resolution. They tend to complete rather than extend, missing the intuitive hesitation, ambiguity, or lingering resonance that makes a poem breathe beyond its last line.
This reinforces my stance on oscillation as an organic phenomenon—one that a poet must feel rather than merely construct. AI can follow patterns, but it lacks the instinct to leave space for interpretation, allowing a poem to remain fluid rather than closed off. Open-ended finales require restraint, an awareness of when to pull back rather than conclude—something that AI struggles with because its primary function is to generate answers, not leave room for endless contemplation.
See my latest contest, Elements of Nature, where I challenge poets to bring their creativity to the table for all to admire—being better than AI is not that difficult … It does not have to be a lengthy poem – just your own.
Happy quills!
Su
*AI Assist versus AI-generated Poetry | PoetrySoup.com