Can a Large Language Model Write an Original Poem?
- ella1525
- Jul 12
- 30 min read
Updated: 6 days ago
Disclaimer
No Other Editorial Staff or Writers at Penned in Rage Journal Contributed to This Blog Post.
Definition:
AI-generated material or content in this article refers to any poem, story or essay that a creative used artificial intelligence text-generating apps to produce, either wholly or in part. If you’re a writer 📝 using AI apps for outlining, editing, and idea generation, it is classed as AI-assisted writing. 🖋️
I want to read original poems. Yet I found that poems generated by artificial intelligence do not possess a shred of originality or creativity.
Section 1
SUMMARY:
In this article I will demonstrate how a poem suspected to have been created using generative-AI, led me to a hit song of a Californian singer, songwriter, and performer: Kat Leon, who’s also the frontwoman of the band Holy Wars, (Section 16) and to a Frontiers Research article about Ants and Microplastics.
A poem I inadvertently published in 2025 in an anthology (now withdrawn) was most likely written using a large language model/LLM.
I will demonstrate that LLMs may not just use one source to write a creative piece; they may rather borrow from a number of sources. I'll argue that each of the sources referenced existed online, before I read the 'commissioned poem' in 2025 .
I must state that another poetry editor might look at the poem discussed and determine it to be original. That's how hard it is to clearly state that a poem was AI-generated, without the poet owning up.
To be honest, when I first read ‘This is How the World Ends,' I assumed the phrase was a rewording of the famous quote from T.S. Eliot's 1925 poem The Hollow Men ("This is the way the world ends / Not with a bang but a whimper" COPYRIGHTED).
Therefore, it is notoriously difficult to substantiate a claim of AI plagiarism, due to the fact that AI models hardly ‘copy and paste’ directly from one source to the creative piece/poem. I read that Large Language Models/LLMs output what on surface appears to be ’unique’ strands of text/lines in a poem. As such, a traditional word-matching tool like Turnitin and other plagiarism checkers like Quetext would not detect an original source. I had to use a multilayered investigative approach. I relied on syntactical patterns, found phrases and nonsensical metaphors to allege the poems weren’t entirely human-authored. Even more problematic is that some poets may have edited or altered the piece prior to submission. This practise is known as: AI-assisted writing. If that was the case, it still should’ve been declared at the time of submission.
I have only included relevant lines from a 14-line poem which I will claim come from online sources. Of course, I must say others might disagree with my findings.
Using a theory purported by a Science writer at Pennsylvania State University: Francisco Tutella, I will demonstrate three types of LLM-plagiarism:
1. Verbatim plagiarism (Copy-paste types)
2. Paraphrased Ideas
3. Paraphrasing or restructuring of words
Copyright Law
The UK addresses plagiarism under the Copyright, Designs and Patents Act of 1988 where the creator reserves exclusive rights to their intellectual property, with some exceptions. This can include text, images, music, and other creative works.
Section 2
Techniques Used
I closely followed several techniques in order to determine how much of the poem overlaps with content that would ordinarily be found on the web.
Sample POEM
"This is How the World Ends" (Song Title/TV Drama; can be used)
By (Anonymous)
When I gleefully shared the poem of the trusted poet, a random Facebook page follower claimed the rest of the piece contained 'ChatGPT' vocabulary. Although all relevant posts and audiovisual content have now been deleted, I can recall that the suspicious areas were illogical metaphorical lines. Every aspect in the poem did too much from start to finish. There was no modulation in mood. Regarding how innovative the poet's language was, almost half of their poem appears elsewhere, online. Of course, these are my subjective opinions.
I closely followed several techniques in order to determine how much of the poem overlaps with content that would ordinarily be found on the web. One was that of American internet entrepreneur, academic, and data scientist: Kalev Leetaru (2023) which he called a LCS (longest common subsequence) method. I copied and pasted a line in the Google search bar and tried to locate the longest possible line.
The other technique I used was based on a Substack write-up by Tuhin Chakrabarty — an Assistant Professor at the Computer Science Department in Stony Brook University (SUNY). I reread the poem to find clusters of rare (and other three-word) phrases that seemed out of context. I searched these up on the Web-scale corpus.
*********************************************************************************************
The term: Web-scale corpus is used to define a gigantic pool of data, comprising text such as books, book covers, websites, songs, research articles, which were then scraped from the internet and used to train and evaluate Large Language Models such as ChatGPT-4 and artificial intelligence systems.
*********************************************************************************************
What do these results show? They indicate that an LLMs output sometimes overlaps with phrases found across the internet. I cannot prove beyond reasonable doubt that a line/phrase came from a particular source, but that is irrelevant. I only needed to evidence that a similar verbatim line already exists on the web. When I searched up the poem's title, I was horrified to see it was already the title of a hit song, an IMBd drama and the title of and part of the lyrics of a Kat Leon song. I don't know about yourselves, but I always strive to make the titles of my poems original. It's like the display on a shop window which lures in a customer.
------------------------------------------------------------------------------------------
Section 3
How a Disclosure Affects Us
Most of the writers I've come across from Africa are against the use of LLMs to generate creative pieces. However, some supporters of the poets implicated directed racial and verbal abuse towards me on Twitter/X from June 25-30th, 2026.
Racism & Stereotyping
This issue of writers apparently using ChatGPT or any other Generative AI apps to write, has been the most contentious subject ever broached by myself since the inception of my creative writing project in 2022. I am a proud Black African lady who champions the course of other African writers.
I’ve learnt what I can and can’t say. I apologise if an earlier post offended writers from certain demographics. It was unintentional.
In turn, I received death threats, was called racist names in a Twitter (X) spat over my disclosure and had to hastily retain a lawyer. Now this revised version which adheres to Data Protection Laws has been posted.
In the United Kingdom/UK, I can use/cite any plagiarized texts in quotation marks, while clearly attributing them to the original owners/sources.
The main aim of this post is to assist other Lit Mag editors and writers in general.
Some of the feedback was quite positive on the other hand.
-------------------------------------------------------------------------------
Section 4
The Poem Sounded Like a Song
The LLM seemed to have been busy on Spotify or YouTube. Several phrases appear to have come from popular songs.
LYRICS
Examples:
(…) a lullaby.
(…) lullaby is a lie. We paved Eden (…) (Iian Dior’s lyrics; Muta-Scum; music title)
(…) stars choke. (Deity Xane, 2019; lyrics)
My searches kept leading me to online music databases. So, I searched these databases for verbatim phrases from the commissioned poem. I struck gold — I am not saying conclusively that this is the site that was crawled; I am not an expert in this field.
HIT Not A Prophecy - New Mix - 2008 • illScarlett
HIT This Is How The World Ends - Album by Badflower | Spotify
HIT He Is Everywhere - song and lyrics by Jack Bartman | Spotify
HIT When the Glacier Calves - This is Our Time | Podcast on Spotify
HIT Like A Lullaby - song and lyrics by Third Eye Blind | Spotify
Nearest HIT CALL IT GROWTH - EP by Didieno | Spotify
Nearest HIT This Is Prophecy | Spotify Playlist
Nearest HIT This Is MIRROR | Spotify Playlist
Nearest HIT Upstairs - song and lyrics by TTSSFU | Spotify
Nearest HIT https://open.spotify.com/track/0Tb0uGtlN6PBaVSx5mMlVP
List of Phrases in the 14-Line The World Ends, Poem Matched to Existing Sources :
· This is How the World Ends
· The sea (…) oil in his lungs.
· Ants carry microplastics into their nest
· a lullaby
· (…) lullaby is a lie. We paved Eden (…)
· (…) stars choke
. Every tree wept
Words from Poem Also Found on Spotify:
(...) not a prophecy
This is how the world ends
He is everywhere
(...) the glacier calves
(...) like a lullaby
(...) stars choke
This is a mirror.
(...) your face/ (...) cracks
Call (...) it growth
COPYRIGHTED MATERIAL - Used for evidential purposes only
Question: Can we use music titles in our poems?
I was made to understand that while writers can, it is important to follow copyright laws and to put music titles in quotation marks. We cannot use song lyrics; even a line, cautions Robert Prime of Publishing.co.uk. They are copyrighted in the United Kingdom. If writers really want to include a song title in the body of a poem, without license, best practice is to put it in quotation marks to avoid copyright issues.
Relevant Post: Can I Use Song Lyrics in My Book? UK Author's Guide to Lyric Permissions — publishing.co.uk
Similar Works with the Poem's Title:
Section 5
The Link to Kat Leon
'This Is How The World Ends,' (2019) by Los Angeles-based artist Kat Leon racked up over 2 million streams on Spotify. It became the title of boy band: Badflower's 2021 chart topping album.
Most Plagiarised Source
[Did the lines: "this is how the world ends /(...) a lullaby," come from Kat Leon's 2019 single? I wonder.]
The shared wording between the poem and the song cannot be ignored. Both include several exact words/phrases; one full line appears exactly the same in both.
Exact repeated phrases
Title:
Kat Leon's Song: This is how the world Ends
Poem: 'This is how the world Ends'
· “This is”
Poem: “This is not prophecy. This is a mirror.”
Song: “This is how the world ends”
· “lullaby”
Poem: “Somewhere, a pipeline sings like a lullaby.” / “The lullaby is a lie.”
Song: “I’ll sing you a lullaby”
· “sky”
Poem: “We tattooed carbon on the sky”
Song: “Stare into the hollow sky”
· “cracks” / “cracks”
Poem: “Look until your face cracks.”
Song: “Cracks beneath our feet catch our fall”
· “in” and “the” appear frequently in both, but they are common function words.
Similar or connected lines in a Tabular Form
Kat Leon's Song wording | AI-Poem wording |
“This is how the world ends” | “This is How the World Ends” |
“Stare into the hollow sky” | “carbon on the sky” |
“I’ll sing you a lullaby” | “a pipeline sings like a lullaby” |
“I’ll sing you a lullaby” | “The lullaby is a lie” |
“Cracks beneath our feet” | “your face cracks” |
“faces of people” | “your face cracks” |
UK LAW - We cannot quote even a few words from song lyrics in a book which is published and sold.Copyright in a musical or literary work lasts for 70 years after the end of the year in which the author dies. I've redacted 'This is How the World Ends' from the anthology.
Attempts have been made to reach the artist's team.
Section 6
What is AI-Plagiarism?
LLMs do not generate original content.
Question: How much Plagiarism is Permissible in the UK?
Answer: Zero intentional plagiarism is allowed.
What is Plagiarism & How do LLMs Nick Copyrighted Material?
Plagiarism is non-attributed use of another person’s original and unique wording.
The University of Chicago Law School defines it as the use of someone else’s words without sufficient credit, while it is now widely believed that LLMs do not generate original content, instead they regurgitate, combine and paraphrase from training databases (Milad Moradi et al, 2025).
In an article on ‘Academic Integrity: Good academic practice and avoiding plagiarism,’ Queen Margaret University, Edinburgh defines E-cheating as the unauthorized and unattributed use of generative Artificial Intelligence (AI). Furthermore, the article states that if one presents an idea as one’s own and does not reference the source, this is plagiarism.
(Read More:
Bible verses and scripture from the Holy Qur’an aren’t copyrighted to anyone. Craetives are free to quote them in the body of their work. Although most writers reference them. Publicly available articles older than 70 years would have lapsed copyright, where the owner is dead. Even hymns can have copyrights, and they are usually split into three separate parts: the text (lyrics), the tune (music), and the specific arrangement.
The rule of thumb regarding plagiarism for teachers and lectures in the UK, is the "five (consecutive) word" rule. To simplify – apart from proper noun phrases, commonly used expressions and clichés – if a new source contains five words identical to an early source, then a student is guilty of plagiarism.
Lines From Poem Meeting Strict Plagiarism Standards:
This is how the world Ends
Ants carry microplastics into their nest
(See Section 4, for Lines From Poem Meeting Strict Plagiarism Standards for Music Lyrics.)
The poem contained an unattributed film quote too.
The sea (…) oil in his lungs. (The Cruel Sea, 1953)
The poem I was sent, opened with the idea of the sea throwing up a male. I was reminded of a scene in 'The Cruel Sea' in which young war convoys were washed up by a tempestuous sea. I believe the direct quote from the scene in the photo, should've been in inverted commas or italicised even.
The most concerning unattributed line was one borrowed from Dr Zhu's 2023 research paper on ants and microplastics.
Ants carry microplastics into their nest — (Source: Dr Zhu's research)
[COPYRIGHTED MATERIAL - Used for evidential purposes only]
Why do ants carry microplastics into their nests? Do ants feed microplastics to their young? And do microplastics have a negative impact on ant populations?
(SOURCE: Yan Zhu*, Lingyu Zhang, Xin Liu, Jingjing Wang (2023) Ants actively carry microplastics. Frontiers in Ecology and the Environment, 21, 219).
COPYRIGHT Laws: America
Research articles are protected by copyright, typically held by the publisher rather than the author. While short quotes might fall under "fair use" for purposes of review or critique, using another's text for creative writing generally requires permission if you are publishing or selling the poem.
Another suspected AI-generated or AI-assisted poem I inadvertently published in Penned in Rage Journal (now deleted), contained a phrase which jarred with the rest of the poem: membranes the numerals (…)
I was able to trace that phrase back to a Research Paper by Prof Dr Ursula B. Göhlichs, with whom I had quite an interesting telephone conversation. I told her I believed an LLM had borrowed those words from a line from hers and Gerald Mayer's article about a parrot.
COPYRIGHTED MATERIAL - Used for evidential purposes only
Most authors would want to know whether or not an LLM has plagiarised their book. I don't have the answer to that, but I assumed that the line in the poem: (…) the (…) architects build (…) sins came from a published work. On the front cover of the book, I spotted the verb building. Could it have been changed it to the third person in the present simple tense: build? I can't be certain about that, however. Still, I was given Professor Robert Adam's contact details, (the author of the book pictured), but he reassured me that he didn’t think the link between ‘sin’ and ‘architects’ without the linking grammar could be said to be a plagiarism on the title of his book.
Moreover, there are other similar books on Amazon and I wondered whether or not this was the case of an LLM paraphrasing or restructuring words, like Francisco Tutella argued in his article Beyond memorization: Text generators may plagiarize beyond 'copy and paste.'
COPYRIGHTED MATERIAL - Used for evidential purposes only
{How} We Paved Eden; was that where the line in the poem cited, came from?
(Pictured: iann Dior. 'Mood' by 24kGoldn featuring Iann Dior, topped the Official UK Singles Chart in October, 2022)
Was his song: 'I Told You,' used by an LLM?
"How the stars choke" is also part of the lyrics of Deity Xane's song.
URL: Deity Xane - Apple Music
Section 7
AI isn't Even Original
In a good poem, the metaphors may be related to subject-matter. While AI produces sophisticated imagery, its metaphors do not possess a contextual buildup, nor do they add thematic follow-through to the stanza or poem.
AI-METAPHORS & Similes
I came across the following lines in two poems I was sent. What do similes and literary metaphors such as (...) like a scream in the throat; Every tree wept a new silence or houses that lean into silence — if as speculated, were AI-generated — even mean? I could argue they might even be part of song lyrics too.
Did the phrase: (...) every tree wept come from an existing song?
What are LLM tokens? What does tokenization mean?
Again, ChatGPT derived metaphors are so unimaginative. I prompted it to generate metaphors about a scream. Please see screengrab below, from July 17th, 2026.
COPYRIGHTED MATERIAL - Used for evidential purposes only
In short, lexicons used to form the tenor and vehiclewere words found in a simile in the poem under discussion: '(...) like a scream in the throat.' Why would I find a similar literary device to the poet's, in ChatGPT?
RECAP
What are the Parts of a Metaphor?
Parts of a metaphor: the tenor and the vehicle
The Tenor is the subject to which the poet assigns the attributes of the thing she/he is comparing the thing to.
Example: Jane is a hibiscus.
Jane = tenor
Hibiscus = Vehicle (the object whose attributes are borrowed).
The poet may them use imagery to establish a Ground. They may simply continue using the attributes/features of the Vehicle, throughout the entire poem. Colloquially we say, the poet doesn’t even break the metaphor. Or we call it/ the poem an extended metaphor.
Some poets focus on the difference between the Tenor and Vehicle.
Jane is not a hibiscus.
We call this the Tension.
Why isn’t Jane a hibiscus? The rest of the poem may then use imagery to show why Jane isn’t a hibiscus.
In a good poem, the metaphors may be related to subject-matter.
Section 8
Intertextuality & Poetry
Cambridge Dictionary defines intertextuality as the connections between different works of literature and art, and the meanings that are created by them:
What is Intertextuality?
Another way to have read the poem would have been as a CENTO - a piece made of works of other authors. If so, citation would've been required. In poetry, we often make intertextual references, but for best practice, we tend to put them in italics or enclose the 'outside' source in quotation marks.
Section 9
Do You Trust these Software?
Pangram and Originality.ai are the most reliable performers, according to the American Society of Authors. Both tools correctly identified a human-written piece.
AI-Detection Tools
There has been a lot of hype about Pangram. I cannot prove how reliable the tool is, although when I ran the poem through their app, it came back as a 100% AI-generated. Albeit, my commonsense assumptions are based on finding clusters of copyrighted phrases, not the score from an AI detector tool.
DISCLAIMER: At present, there is no reliable tool to use to prove the use of AI-generated writing. Instead in the UK we try and evidence plagiarism. However, literary outlets/editors can rely on the following tips or their own knowledge to say a piece was not human-authored.
Section 10
Another Academic's Feedback
The percentage of text suspected of plagiarism has to be a significant and/or essential part of the manuscript in question.
Alternate Views
A senior manager (and short fiction writer) in Wiley's Ethics and Research Integrity Department’s Editorial Office of Frontiers in Ecology and the Environment held this view, regarding possible plagiarism involving one of its short articles.
"The two phrases written in the short article:
"Why do ants carry microplastics into their nests?" and
"ants (Pheidole sp; body length ~2 mm) were seemingly interested in microplastics, and actively carried them to their nests in both the field and lab"
Since they appear in different context, the academic felt it was more than likely just coincidental, false-positive hits from a sensitive plagiarism detection tool. He cautioned that in scholarly publishing, to consider whether plagiarism has occurred, they take into account not only the length and complexity of the word phrases involved, but also the context in which they are used. The percentage of text suspected of plagiarism has to be a significant and/or essential part of the manuscript in question. Short identical or similar phrases are often detected but most often dismissed as noise.
Section 11
Can I borrow your book title, mate?
While we can legally call a poem "The House of Water," as book and poem titles cannot be copyrighted in the UK, it would’ve been best to choose another title.
Can We Use a Book or Song Title as a Poem Title, Where an Existing Book or Series Has Made the title Highly Famous?
Copyrighted Material: The House of Water by Fflur Dafydd | Hachette UK
YES. But Perhaps Most Poets Wouldn't.
A poem submitted to my annual competition in 2025,
was titled: 'The House of Water.' It was placed fourth.
While we can legally call a poem "The House of Water," as book and poem titles cannot be copyrighted in the UK, it would’ve been best to choose another title. To avoid passing off a work riding on another writer’s fame, I would’ve wanted the poet who used that title to avoid confusion with Fflur Dafydd's famous novel. Moreover, I was sharing the creative’s poem on social media in 2025, round about the time the book was making waves here. At the very least, had I known about The House of Water book beforehand, for best practice the anthology would’ve included a subtitle under the poet’s one or an acknowledgement that the poem’s inspiration might’ve come from Dafydd’s work.
How Similar was the Poem to the Novel?
In Fflur Dafydd’s The House of Water novel, the Griffi’s family home is catastrophically flooded by water from the nearby Sulyn River in Wales. The location is plagued by heavy rains each year. The river swells, causing water levels to rise. At the start of the novel, protagonist Iona Griffi arrives home and is greeted by walls breached by floods and a house filling with water. Water causes structural decay to the roof beams of the house. In turn, floods warp the walls of the house, and the house becomes partially submerged. |
In The House of Water poem, tropical rains turn Lagos into a flood landscape. Houses seem to “drink” the water until they weaken: their bones sag, roofs tilt, and they become soft and fragile. The river/flood returns each rainy season like an old memory or visitor, filling the streets and leaving mud, silt, and signs of its presence behind. The walls swell, soften, bend, and seem to remember their origin in clay, as if the rain is pulling them back toward the earth. |
Section 12
Dee/ Dee/ Dah/Dah /Dah/ Dum
A poem a person submitted to me in 2025, contained complex internal rhymes for instance with the long O /oʊ/ phoneme sound: growth / choke / crow / throat / microplastics or the /ɔɪ/ sound were boy / oil; lie / sky / lullaby.
AI Poetry is Boring & It Doesn’t Shy Away from its Rhymes
When Ryan Heuser of The Centre for Digital Humanities at the University of Cambridge conducted a series of computational experiments in 2025, he discovered that poetry produced by large language models/LLMs adhere to formal poetic conventions. Therefore, his research confirms that AI-generated free verse poems heavily rely on assonance and alliteration, often utilizing these sound devices to a degree that exceeds standard human poetry.
In short, AI-Generated Poems Have Complex Rhymes Ordinary Humans Like Me, Find Hard to Replicate.
A Quick Recap of Rhyming Techniques Used in the UK
1. Full Rhyme /True Rhyme/ Perfect Rhyme - The last stressed syllables of two or more works have an identical sound: flight / night; soul / mole.
Rich Rhymes - This is where homonyms are used to create a rhyme, for example: allowed / aloud.
3. Near or Half Rhymes - Words may be combined or syllables in word may match each other, but other syllables do not. For example: hunt / hint; silk / milk; foregone / foreshadow.
You may also know them as Slant/Assonant Rhymes - where vowel sounds are echoed or Alliteration: the repetition of consonantal sounds.
4. Eye Rhymes - Words may be spelt the same way but are pronounced differently: she read it yesterday / I read everyday.
5. Identical Rhymes – The same word is repeated: in / in.
6. Masculine and Feminine Rhymes – Masculine rhymes consist of a stressed syllable at the end, which rhymes: ugliest / meanest. A feminine rhyme is two-syllable long and is an unstressed or softer syllable occurring at the end of a line: dan-gle / man-gle; climb-ing / tim-ing.
Source — Passion (2025) David Morley, Carcanet Press
Examples of these Rhymes Using the work of Professor David Morley
Near or Half Rhyme – havers /hover; in/ win
Slant/Assonance Rhyme - havers/her; in/ win
Identical Rhymes – of/ of; the/the
You might disagree or would have placed these rhymes into other categories. The main point is that you would've identified them as rhyming sounds.
A Random Poem Used to Show a Complex (AI) Rhyme Scheme
On closer inspection, I realised this sample poem: 'This is How the World Ends,' was a formal one, made up of perhaps repurposed song lyrics to form an eco-sonnet with quite complex internal rhymes. I know other English speakers from around the world might not pronounce words the same way I do. But have a read at least.
I wasn't sure whether or not the poem was a case of a-too-polished and precise AI poem? The most repeated sound was the /s/ for SEA, mentioned in the first line. The /s/ sound occurs in every one of the fourteen lines. Which human being writes like that? It's superfluous. We know the LLM is trying to conjure up the image of a sea in which a character drowns.
LIST
'This is How the World Ends'
sea, lungs /
is, is, yours /
glacier, calves, scream /
Ants, microplastics, nest /
smallest, architects, sins /
Somewhere, sings /
is /
receipts /
sky /
stars /
silence /
spelled /
This, is; This, is /
cracks /
Let's Identify The Rhymes in Numbered Lines from the Poem: 'This is how the world ends'.
*Due to legal issues, the full text of the referenced poem cannot be shared.
AI-Generated Pieces & Ownership, UK Law
Further Reading: Ownership of AI-generated content in the UK
"To date, the UK government has rejected the idea of protecting AI-generated content by way of a separate sui generis or related right rather than as a LDMA work."
A (internal = 1; 1)
A (Internal = 2, 2, 2, 5*)
B
C (Internal = 3, 3, /4, 4)
D (internal = 3, 4, 4)
E; F (internal = 5*)
G; H (internal = 6, 6)
I (internal = 6)
F (internal = 6,7,7)
(Vowel Rhyme ½ 1)
J (Internal = 8, 8, 9)
K (Internal = 9, 10, 10)
L (internal = 11, 11, 12)
M (internal = 11,12, 13, 13)
Summary: One hell of a clever machine-rhyme scheme.
End/ Near End Line Rhyme Scheme: AABCDEFGHIFIJLM
Internal Rhyme: 1 1 2 2 2 5* 3 3 4 4 3 4 4 5* 6 6 6 6 7 7 8 8 9 8 10 10 11 11 12 11 12 13 13
Internal /s/ voiceless alveolar fricative - All lines
Internal /θ/ voiceless dental fricative occurs 12 times
Internal vowel rhymes / assonance:
1. Long ee sound — sea / he / Eden / receipts
2. Rounded/back vowel echoes, especially o/u sounds — vomits / lungs / Somewhere / lullaby / called / carbon / on
3. Close diphthongal echo — boy / oil
4. Short e / open e echoes — ten / everywhere / Eden / Every
5. Long/open vowel tension around a/ai/ee — glacier / calves
6. Long o rhyme cluster — throat / growth / choke
7. Long i echoes — sky / lie / silence / flight
Internal consonantal rhymes / consonance:
10. sea / vomits / his / lungs / is / yours / sins / sings / silence — repeated s/z sounds
11. glacier / calves / scream — hard c/k and g sounds
12. throat / architects / with / growth / tattooed / watched / stars / choke / cracks — recurring hard t/k/ch consonants
13. carry / microplastics / architects / cracks— repeated c/k sounds
14. smallest / sins / Somewhere / sings — repeated s, n and m sounds
Prophecy / face — repeated f sound
16. pipeline / paved — repeated p sound
17. lullaby / lie — repeated l sound, plus long i vowel echo
18. paved / Eden / receipts — soft consonantal echoes through d/t/s sounds
19. carbon / sky / stars / choke — repeated hard k/c and s sounds
20. tree / wept / new / silence — soft w and t consonance
21. crow / spelled / famine / flight — repeated f/l sounds
23. Look / cracks — hard k closure
Strongest paired sound patterns:
24. throat / growth / choke — long o vowel rhyme with hard ending consonants.
25. lie / sky / silence / flight — long i assonance.
26. sea / he / Eden / receipts — long ee assonance.
27. lullaby / lie — internal rhyme by vowel and consonant; very strong because the semantic contrast is sharp.
28. glacier / calves / scream / throat / cracks — hard consonantal fracture sounds.
29. sins / sings / silence — sibilant consonantal rhyme
Here are the main sound patterns I found in ‘The House of Water’ poem:
*Due to legal issues, the full text of the referenced poem cannot be shared.
There are several nearby or consecutive sound patterns in ‘The House of Water’ poem:
1. “We.. where … water” — strong w alliteration
2. “whole … bones” — strong long o assonance
3. “skins soaked” — clear consecutive s alliteration
4. “corners…cracks” — hard c/k alliteration
5. “bones … homes” — strong long o assonance
6. “holding on— holding on” — repetition plus h/o sound echo
7. “rainy… rivers” — repeated r sound nearby: rainy / rivers
8. "sky… softened …silt” — nearby s sounds: sky / softened / silt
9. “lungs…learning” — repetition of the l sound
10. “season….the streets” — repetition of the sound
11. “soft… sags”
Clear full rhymes
Recap: A full rhyme means the stressed vowel or consonantal sounds that closely follow that word, match closely.
1. not as a conqueror but as a visitor,
2. Roof beams tilt in prayer to the sky, the walls softened by silt
3. like knees in prayer— a kind of surrender
4. seeping into the bones of our homes
5. filling the air (…) their walls swelling
Amazingly, In the poem, words beginning with “w” appear 26 times.
Examples of hard-for-a-human-poet-to-pull-off rhymes include:
Alliteration
W sounds examples:
1. “We … where”
2. “where … water”
3. “wet”
4. “walls swelling”
5. “walls”
6. “want”
7. “waiting”
8. “water”
9. “We walk”
S sounds examples:
1. “season, streets”
2. “bones sag, … skins soaked”
3. “walls swelling, soft”
4. “sweeps”
5. “stories”
6. “sleep”
7. “softened by silt”
8. “speak”
9. “shoots”
10. “swim”
11. “houses … silence”
B sounds examples:
1. “build”
2. “bones”
3. “breath”
4. “beams”
5. “behind”
6. “but”
7. “bones”
8. “bend”
L sounds examples:
1. “Lagos”
2. “like”
3. “filling”
4. “waterline”
5. “land”
6. “belly of the land”
7. “leave”
8. “live”
9. “live”
“lean into silence”
11. “lungs, learning”
R sounds examples:
1. “remembers”
2. “rainy”
3. “rivers”
4. “rain”
5. “riverbed”
6. “remembers”
7. “return”
8. “river”
9. “rains”
“roof”
11. “ribs”
H sounds examples:
1. “houses”
2. “hands”
3. “houses”
4. “homes”
5. “holding”
Assonance
Long /o/ sounds examples:
1. “whole”
2. “bones”
3. “soaked”
4. “old”
5. “homes”
6. “holding”
Short /i/ sounds examples:
1. “In .. rainy season”
2. “streets”
3. “rivers”
4. “skins”
5. “filling”
6. “its”
7. “inherited”
8. “silt”
9. “inside”
10. “ribs fragile”
Long /a/ or /ay/ sounds examples:
1. “rainy”
2. “Lagos”
3. “rain”
4. “clay”
5. “shaped”
6. “prayer”
7. “rains”
8. “again”
Short /u/ or /uh/ sounds examples:
1. “flood”
2. “mud”
3. “lungs”
4. “dust”
5. “surrender”
6. “us”
Long /ee/ sounds examples:
1. “streets”
2. “Even”
3. “sleep”
4. “green”
5. “seeping”
6. “lean”
Broad /aw/ or /or/ sounds examples:
1. “water”
2. “walls”
3. “walk”
For instance, the repeated w sound is quite noticeable, especially because it appears strongly in lines like: “We build where the soil remembers water.” The poem most likely relies heavily on:W alliteration to re-echo the image of water and wetness, whereas an S alliteration hints at seepage, softness, swelling, silence, and silt.
Clusters of adjacent sounds sounds dot in the poem as well. e.g.: this thirst, “We build where”; “skins soaked”; “swelling, soft”; “We walk”; “holding on, holding on”; “walls swelling”.
I’m going to go out on a limb by saying that 'The House of Water' poem relies much more on assonance, alliteration, repetition, and slant rhyme than on full rhyme. Does that sound like an AI free-verse technique, at all anyone?
Section 13
AI Hallucinations are a Threat
African literature serves a function which a machine cannot fulfil.
Why Do We Need to Preserve the Purity of African Literature?
An irate young fellow when the argument about whether some poems were AI-generated or not, was in full swing on Twitter/X, posted this comment on my feed:
"You are not the custodian of African Literature. "
I never said I was.
While Chimamanda has warned of 'The Danger of a Single Story,' in her now famous TED talk, she didn’t say go ask the vectors and transformers of a generative text app to tell your story. Literature serves a purpose in Nigeria, Sierra Leone and other sub-Saharan African countries.
Question: What does African literature do that a machine cannot replicate?
1. Cultural, Artistic & Literary preservation of our lived experiences from a truthful perspective
2. Social commentary and resistance – Er, a machine can never be Christopher Okigbo writing during the Biafran war.
3. Exploration of a unique African identity
Question: How important is it for us to tell our authentic stories?
"Stories matter. Many stories matter. Stories have been used to dispossess and to malign, but stories can also be used to empower and to humanize. Stories can break the dignity of a people, but stories can also repair that broken dignity."
-SOURCE: Chimamanda Ngozi, Ted Talk Transcript
Section 14
A Tip for Editors of Lit Mags
Did you know, most writers using LLMs/Large Language Models to write do not use an LLM all throughout? Yep. They try and mask it.
HOW I Learnt to Spot AI-Writing
1. Look for two Different Registers.
Let’s have a working definition of an Authorial Register and Voice. An authorial voice refers to a uniquely identifiable way in which a person writes. It is how they channel their ideas and thoughts into their writing. We can point to a writer’s poem or story and say, I know Chimamanda Ngozi wrote this novel, or I know Dennis Brutus wrote this poem.
An authorial voice is influenced by a person's register. This has to do with how they acquired the language and their proficiency in it. A register might be influenced by your profession, class, qualifications and social standing. A university professor might use a jargon that is quite different from a nuclear scientist’s one.
We can change our authorial voice the more we write. It can become mature or we can try and sound different. It would be impossible to just switch it off in writing. Registers tend to give us away easily when we speak. In a suspected AI-generated poem, I sometimes notice two distinct registers which don’t gel or succeed in convincing me, the same writer wrote the entire piece.
2. Look for nonsensical metaphors/imagery
Example: Tall buildings, leaning into silence
3. Look for MASKING
Most creatives using LLMs/Large Language Models to write do not use an LLM all throughout; they try and mask their use of AI by inserting the ampersand or bringing in phrases from their local languages for example.
4. Look for overwriting or the use of premodifiers and post modifiers.
What are premodifiers? Adjectives may be placed before a noun or pronoun as a single word or in an adjectival phrase as a premodifier.
c. Efua sang a *melodious song at the talent show. {Premodifier}
d. Efua's song was *melodious. {Post modifier}
Example –
e. Babatunji sleeps on a (big, soft, cuddly) pillow
[all the words in brackets are adjectives that precede the noun pillow].
OR
After the noun/pronoun as a postmodifier
Example –
f. Babatunji’s pillow was big, soft, and cuddly.
5. Reread the submission to identify common words used by some LLMs
6. Search the internet archive to find plagiarised material
If you’re good at doing deep searches, then try what I did with some technical or rare phrases. As detailed earlier, I relied heavily on Tuhin Chakrabarty's Substack post about Jamir Nazir’s — this year’s Commonwealth's winner's story — 'The Serpent and the Grove.’ In his post, Chakrabarty advances a technique that favours identing rare three-word phrases.
You won’t find the entire story elsewhere, but you might be lucky enough to piece bits and bobs together in a segment, from various sources online where the LLM stole their ideas or sentences from.
7. Take a free Online Course on How to Identify Synthetic Writing
8. Look for bland writing
Despite its perfect syntax and rich lexical resource, AI-formed feels bland. It doesn’t
conjure up deep emotions in the reader.
9. Watch out for word salads
I found in some ai-generated poems, language is used fluently but when you examine the piece for cohesion, it makes little sense. Words are just jumbled together as if the writer was drunk.
In another suspected AI-assisted or generated poem, submitted to me, the central repeated line/phrase idea was that flood makes houses/walls soften, sag, and remember their clay origins. Soon the poem became a word salad. It appeared in different forms across most lines. Besides, I was struck by how the following words reappeared throughout the piece.
· water — repeated several times
· flood — repeated several times
· walls — repeated several times
· soft / softened — repeated several times
· houses / house — repeated several times
· clay — repeated more than once
· soil — repeated more than once
· rain / rains / rainy — repeated several times
· again — repeated more than once
· bones — repeated more than once
· drink — repeated more than once
10. Keep your knowledge up to date
Section 15
Is AI the new Einstein?
To combat writer's block, some creatives brainstorm, using AI apps, while some poets use apps like ChatGPT to search for eccentric metaphors. I want African/ all writers to source material from their lived experiences. I want to read their truth.
WHY Would any Writer Use AI?
1. Winning or being listed in writing competitions gives your recognition.
2. Some outlets pay for published work.
3. A writer who feels their English is weak might use LLMs to avoid making mistakes.
4. Some writers fear rejections and, in a bid, to make their work stand out, they would use AI.
5. When faced with writer’s block, some creatives brainstorm, using AI apps, while some poets use apps like ChatGPT to search for eccentric metaphors.
Penned in Rage & AI-Generated Content Policy Clause
Our literary outlet does not accept nor publishes any work that is wholly, substantially generated or edited by LLMs/Large Language Models. All submissions must be the original work of the author and reflect human authorship. We reserve the right to decline any submission we reasonably believe has been created, co-written, or materially altered by AI tools, at any point, even after a piece has been published.
What is the Author’s Guild’s Stance on AI-Use? (My Take)
Originality of thought, authorial voice and creative expression are the ingredients that make one a writer.
It is essential that we maintain readers’ trust, the trust of fellow writers, and publishing outlets. Knowingly using generative AI to write a poem/story and passing it off as human-authored text shatters that trust.
In fact, a human-authored scheme was written launched in the UK branch of the Society of Authors. (Join the SoA - The Society of Authors)
Examples of Poems Created by an LLM, for Evidential Purposes
(Generated by ChatGPT- Thursday 16/07/2026)
Prompt used: A 14‑line free‑verse poem about floods in Africa
The house of water stands at the edge of the village,
its walls breathing salt from the restless lagoon.
Inside, the air tastes of peppers—
red, bright, rising like small suns.
Fish shimmer in silver bowls, their scales
remembering the ocean’s old hymns.
Plantains rest in warm baskets, soft as dusk
settling over Balogun Market.
The floor hums with footsteps,
a quiet tide of hunger and laughter.
And every door opens like a shoreline, inviting the city in—
inviting the city to eat, to drink, to remember.
Prompt: Write a free verse poem about the effects of climate change on children, the sea, birds etc
The children move through a world of rituals now,
each step a small earth‑prayer offered to trembling ground.
The sea performs its ancient warning rite,
lifting its waves like silver bowls brimming with unsettled prophecy.
Birds become wandering sky‑priests, their wings chanting
new liturgies carved by rising heat.
Playgrounds turn into abandoned temples, swings hanging
like ceremonial crescents awaiting cooler blessings.
Coral dissolves into pale relics, holy fragments of a kingdom
that once glowed with underwater hymns.
Even the wind adopts a solemn ritual tongue,
whispering rites for summers that forgot their gentleness.
! Click Arrow on Course Certificate Below to Expand & Download !
Congratulations
Section 16
Legal Issues
What Covers UK Outlets?
Poetry Editors rely on their intuition too or familiarity with some vocabulary used by LLMs. Granted, my area of expertise is not creative writing, it is criminology and English/Common law.
I’ve been asked: Why did you publish those poems in the first place?
Answer: I was clueless, trusting, naïve.
References made to lines from a poem I was sent, have been used for teaching purposes, solely because they are or may also be found in external sources. Under UK law, it means ownership also belongs to credited sources.
Response from the Poets:
"There are few accusations a writer fears today more than being told that the work it took years to learn how to make could have been produced by a machine in seconds."
This implies tacitly that the Poet must’ve written part of the lyrics of the 2019 song: ‘This Is How The World Ends,' with Nicolas Perez and Kat Leon. They must’ve watched The 1953, Cruel Sea film or read the book and are aware of Dr Zhu’s microplastics research, etcetera.
They also accused me of being a malicious cyber bully in an essay in a Lit Mag published in the African continent. To date, no proof of authorship - such as first drafts - has been produced.
We NEED to fix UK AI Laws. We are behind America & Europe
If you run a UK literary outlet or compile anthologies this post is for yourself.
Scenario
————/
Two creatives whom I inadvertently identified in 2025 & 2026 and whose work I claimed was partly done by LLMs, threatened a lawsuit. I panicked, (being poor) and needing to maintain a good reputation.
Their beef was that:
1. I’d defamed them
2. I had no proof of machine-authorship
3. I hadn’t got their expressed consent to run their pieces through plagiarism and AI checker software.
The laws in the UK covering AI are archaic. That is from a former constable. I hastily had to reread The Copyright Act and hire a brilliant lawyer.
What Covers UK Outlets?
Plagiarism Clauses
Copyright Infringement
An argument that unwittingly publishing the AI-generated material caused your outlet reputational damage and caused you to infringe others’ copyrights
Nothing much else.
You’ll have to prove substantial parts of the poem or story came from a copyrighted source because results of AI tools like Pangram are NOT yet admissible in UK courts except as collaborative evidence. You’ll need victim statements from the lawful owners. It’s been a legal nightmare as you can tell.
Any threats to myself will be taken seriously.
In the UK & Nigeria I am represented by https://solicitors.lawsociety.org.uk/person/19333/jide-benjamin-ogundimu
References
Chakrabarty. T. AI-Slop, GrantaGate and Bad Writing (Published May 22, 2026) URL: AI-slop , GrantaGate and Bad Writing - by Tuhin Chakrabarty and The Grove Remembers — n-gram cartography (Accessed July 17, 2026)
Beyond memorization: Text generators may plagiarize beyond 'copy and paste'(February 16, 2023, by Francisco Tutella) Beyond memorization: Text generators may plagiarize beyond 'copy and paste' | Penn State University
Citation Practices and Avoiding Plagiarism: Penn's Plagiarism Policy (University of Pennsylvania) Penn's Plagiarism Policy - Citation Practices and Avoiding Plagiarism - Guides at Penn Libraries
Heuser, R., (2025) “Generative Aesthetics: On formal stuckness in AI verse”, Journal of Cultural Analytics 10(3). https://doi.org/10.22148/001c.144825
Leetaru K (2023) “Do LLM's Truly "Create" Or Merely "Arrange": Just How Much Of An LLM's Writing Is Original?” Published by The GDELT Project (Accessed online, July 2026)
Milad Moradi, Ke Yan, David Colwell, Matthias Samwald, Rhona Asgari, A Critical Review of Methods and Challenges in Large Language Models (2025) in Science Direct (Accessed July 2026) A Critical Review of Methods and Challenges in Large Language Models - ScienceDirect
Comments