[Critical Essay on Creative Writing] AI and Haiku Writing

The Capabilities and Limitations of AI

Artificial Intelligence as a Tool for Haiku Writing: Possibilities and Constraints

Abstract

This critical essay examines the merits and limitations of generative artificial intelligence as a tool for writing contemporary English-language haiku. Drawing on a series of interactions with ChatGPT, it evaluates AI’s understanding of haiku poetics, its ability to generate and revise haiku, and its potential to function as a critique partner and collaborator in linked-verse composition. The discussion demonstrates that AI possesses considerable knowledge of haiku conventions and can offer useful suggestions for revision, generate creative alternatives, and contribute meaningfully to collaborative forms such as rengay. At the same time, the essay shows that AI often produces formulaic or overly explanatory poems and may struggle to recognize or correct weaknesses in mediocre drafts without careful prompting. Ultimately, the essay argues that AI should serve as a tool that supports rather than replaces human creativity. Its value depends not only on the quality of the prompts it receives but also on the poet’s ability to evaluate, revise, and selectively incorporate its suggestions.

Keywords: Generative artificial intelligence; ChatGPT; Haiku; Creative writing; AI-assisted writing

Introduction

Generative AI has earned notoriety among some instructors of English at both the secondary and collegiate levels as a source of plagiarism. They are often frustrated by student papers written entirely or in part with ChatGPT and other AI tools. Meanwhile, many instructors now recognize the educational benefits of AI. One of them is Mari Ramler, a rhetorician and associate professor of English at Tennessee Tech University. Instead of resisting AI and prohibiting its use by students, she incorporates it into her instruction. She asks students to generate a piece of writing with AI and then evaluate it. She comments, “AI is exciting and terrifying, but there are no better instructors to teach a new communication tool than humanities professors—people trained to work with language, conduct cultural critique and explore what it means to be human” (qtd. in Trotter). This statement seems to reflect the sentiments of many instructors of English and the humanities today. The presence of AI is unavoidable, and it may be better to accommodate it and use it for educational purposes.

Considering the increasing use of AI as a tool for prose writers of all kinds, it is not hard to assume that it is also being used by poets. Indeed, it takes less than ten seconds for AI to produce a sonnet or a villanelle, which would often take humans more than an hour to write and revise. It goes without saying that generating a three-line haiku is even easier for AI. Given the easy and free availability of generative AI, it seems worthwhile to investigate how much AI knows about the conventions of contemporary English-language haiku and how it can serve as a helpful tool for haiku poets. After testing ChatGPT’s knowledge of haiku, this essay examines how AI can function as a critique partner and even a collaborator in the writing of haiku, while also discussing its limitations for poets.

AI’s Knowledge of Haiku Poetics

Drawing on AI’s vast store of knowledge across many fields, I asked ChatGPT to explain the basic principles of English-language haiku. Feigning complete ignorance of the poetic form, I used the following prompt: “One of my friends writes haikus. They sound really cool, but I know nothing about them. Can you define them in one sentence and list their stylistic characteristics using several bullet points?” ChatGPT responded as follows:

A haiku is a very short poem, traditionally of Japanese origin, that captures a fleeting moment of perception or insight through concrete imagery and suggestive simplicity.

Common stylistic characteristics of haiku:

· Brevity: Traditionally written in 17 sound units (5-7-5) in Japanese, though many English-language haiku do not strictly follow this pattern.

· Focus on a moment: Often presents a single, immediate experience rather than a narrative or argument.

· Concrete imagery: Relies on sensory details and images instead of explanation or abstract ideas.

· Seasonal reference (kigo): Frequently includes a word or phrase that suggests a particular season.

· Juxtaposition: Places two images, scenes, or ideas side by side, allowing readers to discover the connection between them.

· A “cut” or pause (kire): Creates a shift, contrast, or moment of reflection within the poem.

· Simplicity of language: Uses plain, unadorned diction and avoids excessive description.

· Suggestion rather than explanation: Invites readers to infer meaning rather than stating it directly.

· Attention to ordinary life: Finds significance in everyday experiences and seemingly small details.

· Emphasis on awareness: Often conveys a sense of wonder, insight, transience, or the interconnectedness of human beings and the natural world.

For example, the famous haiku by Matsuo Bashō illustrates many of these qualities:

old pond—
a frog jumps in,
the sound of water.

Then, out of curiosity, I asked ChatGPT the following question: “Do you like the word haikus?” It gave the following response:

I don’t have personal likes or dislikes, but I can say the word haiku tends to work better in its original singular form in English, and “haikus” is often considered nonstandard.

In English usage:

· haiku = singular and plural (most standard: “I read several haiku”)

· haikus = informal plural, common in casual speech but less formal in literary contexts

So if you’re thinking in a more academic or poetic register, “haiku” is usually preferred—but “haikus” isn’t “wrong” in everyday conversation, just a bit less traditional. (ChatGPT; boldface in original)

The answers above demonstrate that ChatGPT has a sound understanding of haiku aesthetics based on its analysis of relevant resources available on the Internet. There was little, if anything, to disagree with. It was also reasonable in addressing the common mistake of using the word haikus. I believe that the spelling haikus, which is widely used today, will eventually be accepted by dictionary makers, who tend to describe rather than prescribe language.

As a next step, I wanted to test ChatGPT’s ability to write haiku. Here was the prompt: “Create four haiku set in four respective seasons.” In response, AI yielded the following:

Spring
rain soft on new grass
cherry petals drift and vanish
the sky forgets cold

Summer
heat shimmers the road
cicadas stitch the afternoon
light leans on still air

Autumn
dry leaves in the wind
a fence line slowly dissolves
dusk gathers early

Winter
bare trees hold the snow
footsteps fade in white silence
breath writes and erases

Here, ChatGPT revealed one of its weaknesses. It generated four haiku in a 5-7-5 syllable pattern. Because Japanese haiku are based on 17 sound units (on) rather than syllables, the 5-7-5 structure is rarely used in contemporary English-language haiku. Although skilled poets can use the 5-7-5 format effectively, many English-language haiku written in that pattern tend to sound wordy and prosaic.

I then gave ChatGPT the following prompt: “All four of these haiku are written in a 5-7-5 syllable pattern. They sound wordy and prosaic. Can you make them more concise and less cluttered?” Here is the revision ChatGPT came up with:

You’re right that strict 5-7-5 in English often pushes haiku toward padding. Here are tighter, more compressed versions that lean into image and cut more aggressively:

Spring
rain on new grass
cherry petals gone midair
sky forgets cold

Summer
heat on the road
cicadas in the noon air
light presses stillness

Autumn
dry leaves in wind
a fence line fades from sight
early dusk

Winter
bare trees, snow held
footsteps vanish into white
breath, then nothing (ChatGPT)

ChatGPT knew how to make the lines more concise, but these revised versions still sound choppy and loose. Rather than creating sharp, juxtaposed images, they merely shorten the original poems. This suggests that, despite its considerable knowledge of haiku aesthetics, generative AI has a limited ability to compose poems that embody the characteristics of haiku it ostensibly understands.

Finally, I asked ChatGPT to generate one spring haiku and one autumn haiku that might be suitable for publication in a leading North American haiku journal. Here is what it produced:

Spring
first warm rain—
the dirt exhales
green light

Autumn
wind through dry fields—
a single crow lifts
the silence (ChatGPT)

These poems are much stronger than the previous ones: they contain sharper images, sound fresher, and end effectively. This series of experiments suggests that AI does not always seem to remember what it knows. Depending on the prompt and the guidance provided by the user, generative AI can produce weak, formulaic haiku, but it also has the potential to generate effective ones.

AI as a Haiku Critique Partner

Most guidebooks for creative writers include the benefits of sharing drafts with fellow writers before they are submitted. For example, in The Haiku Handbook, William J. Higginson and Penny Harter explain, “Sharing is one of the things we want most in life, to give something of ourselves to others, so that they might accept us and our experiences and perceptions as important. The haiku, while short, is long enough to give others an impression of who we are, some piece of the story of our lives” (47). Jane Reichhold is also enthusiastic about sharing one’s work with other poets online: “The ease with which someone new to haiku can find a compatible group online with which to share and study haiku is an opportunity so new that it almost makes one giddy” (105). These perspectives suggest that haiku is a deeply social form of literary exchange.

While many poets would agree with these statements, not everyone is comfortable sharing works in progress with other poets. Research on writing has shown that many writers experience anxiety about receiving feedback on works in progress because they often interpret criticism as a judgment of their abilities or self-worth, even though feedback is essential for learning and improving one’s writing (Badenhorst 106). Some may therefore be too self-conscious to share what they consider imperfect before publication. Others simply prefer to work on their poems alone, even though they recognize the benefits of sharing rough drafts.

One thing AI can do for those who do not want to show their rough drafts is provide feedback in a nonthreatening, impersonal setting. The process is simple: haiku poets can complete a draft, share it with AI, and request feedback before revising it. As with human critique partners, AI can offer honest and supportive comments, and authors are free to accept or reject its suggestions. Based on my experience, AI is more often helpful than unhelpful in revising promising rough drafts, making it a faster, more economical, and more convenient critique partner.

To test AI’s ability to revise a haiku, I provided ChatGPT with a rough draft. I was satisfied with the concept and imagery of the poem, but it still seemed to need further refinement. Here was my prompt: “This is a rough draft of a spring haiku. Revise it for flow and imagery.”

coming out of winter
squirrel and robin
together

The poem was based on my observation of a spring scene in my backyard. Line 1 felt wordy, and line 3 seemed to lack impact. ChatGPT responded as follows:

Because your draft emphasizes emergence from winter, you might strengthen the seasonal transition and the relationship between the two creatures.

A few possibilities:

out of winter
a squirrel and robin
share the yard

Or, with a bit more movement:

winter’s end
a squirrel and robin
cross paths

Or, if you want to keep the sense of togetherness:

coming out of winter
squirrel and robin
foraging together

Of these, “winter’s end / a squirrel and robin / cross paths” feels the most haiku-like to me because it shows rather than explains. The reader senses spring through the animals’ activity without being told too much. (ChatGPT)

Compared with my original version, all three alternatives seemed to flow better. For line 1, “out of winter” and “winter’s end” sounded better than “coming out of winter,” which ChatGPT retained in version 3. In my draft, line 3 consisted of the single word “together,” which seemed vague. Together doing what? All three versions generated by ChatGPT—“share the yard,” “cross paths,” and “foraging together”—portray what happens in my backyard in early spring, a season of renewal and renewed energy, more effectively. Among the three versions, I liked the second best because of its succinctness, followed by version 1 and then version 3.

Inspired by ChatGPT’s alternatives, I came up with the following revision:

out of winter
squirrel and robin
together again

Here, line 1 omits the unnecessary word “coming,” and line 3 adds “again” to suggest that the squirrel and robin have reappeared in the backyard they share each spring. This revision process shows that AI can serve as a virtual critique partner, and, as with in-person critiques, it can offer ideas on how to improve a haiku. The poet has no obligation to accept all the suggestions, but AI’s comments and alternative versions can be helpful in revising a rough draft. Of course, submitting ChatGPT’s version verbatim to a journal would constitute plagiarism.

Compared with poems that require minimal editing, it is more challenging to revise a draft that needs substantial improvement. As an experiment, I gave ChatGPT a three-line poem with a 5-7-5 syllable structure that portrays the beauty of a summer lake:

Morning sun warms lake
gentle ripples kiss the shore
herons rise in air

This is the kind of poem that those new to haiku might create—beautiful but verbose, choppy, and one-dimensional. Too many things are happening in the poem. It also consists of three short clauses that can be arranged into a single sentence: “Morning sun warms lake, gentle ripples kiss the shore, [and] herons rise in air.” I asked ChatGPT to assess this three-line poem, noting that it was intended as “a haiku.” Below is the response it generated:

This works very cleanly as a 5-7-5; the syllable count is correct in all three lines, which already gives it a solid formal base. Imagistically, it’s strong and coherent. The scene moves smoothly from morning light to water movement to bird flight, so there’s a natural progression and widening of space.

A few things you might refine if you want more haiku intensity:

The second line is the most vivid (“ripples kiss the shore”), but the first line is slightly conventional in phrasing (“sun warms lake”). It states the scene rather than sharpening it through a more specific or unexpected detail. The final line (“herons rise in air”) is clear, but a bit abstract in its last phrase—“in air” is implied by “rise.” You could tighten it by letting the verb carry the image more fully.

Overall, it’s a balanced, readable summer haiku with good motion and atmosphere; the main opportunity is to reduce explanatory phrasing and let one or two images carry more weight.

Considering its supposed knowledge of haiku conventions, ChatGPT’s overall positive assessment of this three-line poem disappointed me. So, I asked whether it could be revised for submission to a top haiku magazine. Its revision is below:

Morning sun on the lake
gentle ripples at the shore
herons lift into blue sky (ChatGPT)

This version is an improvement in that the first two lines are now phrases rather than complete sentences. It also avoids using the phrase “kiss the shore,” which reflects a technique—personification—that is not very common in haiku writing. However, the revised poem still consists of three separate lines that do not cohere effectively. This demonstrates that ChatGPT often struggles to revise a draft with substantial room for improvement. In other words, AI can be a helpful tool for experienced haiku poets, but it may struggle alongside those who are unfamiliar with haiku aesthetics.

AI as a Collaborative Haiku Poet

AI can do more than serve as a virtual critique partner; it can also function as a collaborative writing partner. In addition to offering feedback on individual poems, AI can co-write haiku-related sequences such as renga and rengay. Some poets have many colleagues with whom they can collaborate comfortably and productively, but others do not have that luxury. Some may find collaborative writing difficult because of scheduling conflicts, differing writing speeds, or personal shyness. AI, by contrast, is always available and can respond intelligently without making poets feel self-conscious. I once composed a renga with ChatGPT. When I wanted to give AI co-author credit, however, it explained that it could not serve as an author and suggested instead that I acknowledge its contribution in a note accompanying the poem.

While writing this essay, I invited ChatGPT to compose a rengay with me. A two-person rengay consists of six linked stanzas arranged in a 3/2/3/2/2/3-line pattern. Poet A writes stanzas 1, 3, and 5, while Poet B writes stanzas 2, 4, and 6. In the following exercise, I provided the opening three-line stanza as Poet A, ChatGPT served as Poet B, and we continued alternating stanzas until we completed a rough draft. We came up with the following sequence under the title “From Window to Hillside,” which provides a general theme:

March morning
soft light brightens
the window

a robin pauses
on the fence rail

a bluff path
faces emerge
from winter

buds loosen
along the trail

beneath a clear sky
greening hillside

red-wing blackbirds
stitch the distance
with song

In this sequence, the focus moves from the indoor scene to the outdoors, to signs of spring growth, and finally to birdlife on the hillside. In terms of linking and shifting, the sequence seems to have a smooth flow. However, I noticed that each of ChatGPT’s stanzas—2, 4, and 6—consisted of a clause, which is acceptable but often does not sound as poetic as sentence fragments. I asked ChatGPT whether it could provide a break in its stanzas by using fragments. In agreement, it generated the following alternatives:

a robin
on the fence rail

loosening buds
along the trail

red-wing blackbirds
over the valley
morning chorus

These versions looked better, but the last word in Stanza, “chorus,” sounded generic. I encouraged ChatGPT to revise the three lines. It came up with five possible options, including the one below, which I decided to use for Stanza 6:

red-wing blackbirds
over the valley
rising calls

The last line here is less abstract but still musical. The word “rising” also seemed to fit better for springtime.

Now I asked ChatGPT whether it had any suggestions for my stanzas: 1, 3, 5. Regarding Stanza 1, it commented that “brightens” sounds abstract and duplicates what “soft light” already implies. Then, it suggested three options, one of which is below. I liked the fact that lines 2-3 are composed of a concise and imagistic phrase instead of an explanatory sentence:

March morning
soft light
at the window

Meanwhile, instead of the original Stanza 3, which consists of another clause, ChatGPT offered three options. I accepted the following choice:

a bluff path
stones exposed
from winter

Instead of people, these lines focus on landscape, thereby deepening the sequence’s movement. Lines 2 and 3 also sounded more creative and naturally visual.

ChatGPT found the final stanza strong but seemed somewhat generalized. Out of three possible options, I chose the following one:

beneath blue sky
greening hillside

I liked the contrast in color, the alliteration, and the compactness of the lines.

Now that ChatGPT and I finished giving virtual peer critique, I arranged the revised stanzas. It flowed better than the original version, and it sounded tighter and more imagistic. Next, knowing that ChatGPT cannot serve as an “author,” I asked how I could give credit for Stanzas 2, 4, and 6. It suggested several different options, and I chose “John J. Han, with ChatGPT (OpenAI) contributing stanzas 2, 4, and 6.” Below is the final version of the rengay with proper credit.

From Window to Hillside: Rangay
John J. Han, with ChatGPT (OpenAI) contributing stanzas 2, 4, and 6

March morning
soft light
at the window

a robin
on the fence rail

a bluff path
stones exposed
from winter

loosening buds
along the trail

beneath blue sky
greening hillside

red-wing blackbirds
over the valley
rising calls (Han and ChatGPT)

The process for creating this collaborative sequence shows that AI and a poet can work together effectively, giving and receiving feedback and coming up with a product that can satisfy both parties.

Conclusion: The Possibilities and Constraints of AI

AI is intelligent and creative, making it unrealistic to dismiss or despise its capabilities. If used carefully, it can be a valuable tool for haiku poets. As this essay has shown, AI can define haiku, explain its characteristics, generate sample poems, suggest revisions, and even participate in collaborative forms such as renga, rengay, and renku. For poets who do not have the opportunity to take a haiku walk or otherwise seek inspiration outdoors, AI can also serve as a source of creative stimulation. By generating sample poems and offering alternative approaches to a topic, it can help poets discover new possibilities and inspire poems of their own.

At the same time, AI should be used with caution and restraint. Because it is always available and more than willing to provide answers, it can be tempting to over-rely on it. Haiku poets should write their own poems and use AI as a source of inspiration, a critique partner, an editor, or an occasional collaborator. Asking AI to generate poems and then presenting them as one’s own work would be both ethically questionable and contrary to the spirit of haiku as a personal response to lived experience and observation.

Discernment is therefore the key to using AI in haiku writing. As we have seen, AI can provide intelligent and useful suggestions when given clear, specific prompts. However, it may struggle when prompts are vague or when users lack the knowledge necessary to distinguish strong advice from weak, insufficient, or even incorrect information. Ultimately, the quality of AI-assisted haiku writing depends on the judgment of the human poet. Those who teach haiku writing can play an important role by demonstrating how AI may be used productively and responsibly—as a tool that supports human creativity rather than replaces it.

Works Cited

Badenhorst, Cecile M. “Emotions, Play and Graduate Student Writing.” Canadian Journal for Studies in Discourse and Writing/Rédactologie, vol. 28, 2018, pp. 103-126.

Han, John J., and ChatGPT. “From Window to Hillside: Rengay.” Collaborative poem composed through ChatGPT, 24 June 2026.

“request to generate one spring haiku and one autumn haiku suitable for a high-quality North American haiku journal.” ChatGPT, OpenAI, 20 June 2026, https://chat.openai.com/

“responses to prompts on composing and revising four seasonal haiku.” ChatGPT, OpenAI, 24 June 2026, https://chat.openai.com/

“response to prompts on haiku and the term ‘haikus.’” ChatGPT, OpenAI, 24 June 2026, https://chat.openai.com/

Reichhold, Jane. Writing and Enjoying Haiku: A Hands-on Guide. Kodansha International, 2002.

“revision of ‘coming out of winter.’” ChatGPT, OpenAI, 24 June 2026, https://chat.openai.com/

“revision of ‘Morning sun warms lake.’” ChatGPT, OpenAI, 24 June 2026, https://chat.openai.com/

Trotter, Megan. “Tennessee Tech’s English Faculty Examine AI’s Impact on Writing.” Tennessee Tech University News, 6 Apr. 2026, https://www.tntech.edu/news/releases/25-26/tech-english-faculty-examine-ai-impact-on-writing.php.

About the Author

Professor of English and Creative Writing and Associate Dean of the School of Humanities and Theology, Missouri Baptist University

John J. Han, PhD, is the author, editor, co-editor, or translator of 35 books. His forthcoming book, Echoes from the Hills: Critical Essays on Ozarks Literature (co-edited with C. Clark Triplett), will be published by the University of Arkansas Press. He has also published nearly 3,000 poems in a wide range of journals and anthologies. In addition to his long tenure at Missouri Baptist University, he has taught at Kansas State University, the University of Nebraska–Lincoln, Nebraska Wesleyan University, and Washington University in St. Louis. He has also served as a visiting scholar at Georgia College & State University and the University of Ljubljana in Slovenia.

© K-Global Scholars and Professionals Forum. All rights reserved. Content published in the K-GSP Forum may not be reproduced, distributed, or transmitted in any form without prior written permission from the K-GSP Forum, except for brief quotations with full attribution.