Introduction

The appropriation of other writers’ texts in e.g. collages precedes any digital means of writing and can be traced back to antiquity. Data collection, the ease of sharing texts online and the development of natural language generation techniques have further broadened the repertoire of possibilities for using other people’s writing in creative works. Through two case examples – AI Imagines: Classic Literature (2023) and Things of Internet (2018) – this article examines what forms of writing based on other people’s writing are made possible by text generation and platformization. In Things of Internet, Kokko rescues platform-based digital content via screenshots into a collage; in AI Imagines: Classic Literature, a vernacular work published on Amazon written by ChatGPT, canonical literature is condensed by an LLM.

LLMs, which have along with other generative programs become colloquially synonymous with AI, are often framed as representing a continuum of writing systems and writing aids. From the shift from oral tradition to written word, the invention and spread of printed works to the use of typewriters and digital text processing, writing technologies can influence both texts themselves and their writing process and thereby ways of thinking (see Kittler, 1999: 200). LLMs are presented as the latest tool for writing that should be assessed in relation to these earlier breakthroughs. For instance, incorporating LLMs in writing has been compared to adoption of written language in general and critiques of LLM use as offloading cognition are compared to Socrates’ argument that switching to writing weakens memory (Gunkel, 2024; Ippolito, 2026). Furthermore, natural language generation technologies are placed in continuity with the printing press in the sense that both can bring about lasting social change in literary production (Henrickson, 2021: 48–49). Hence, large language models are often referred to as tools that will change the way we write by heads of AI companies, researchers and authors alike. Therefore, the question is: how exactly will these tools change writing? In this article, we narrow down this question by examining the use of large language models in relation to the tradition of experimental literature and the use of texts of others as part of a literary work.

In the contexts of electronic literature, the exploration of different technologies in writing is also linked to resisting the idea of authorship as self expression, instead highlighting different modes of distributed authorship, where texts are reused, repurposed, and recontextualized in literary practice (Biggs & Travlou, 2015). This is a relevant consideration in our analysis as well, as both examined works are based on recontextualizing existing texts into new literary works. We examine how the technique of digital collage in Things of Internet compares to the use of generative AI in AI Imagines in relation to how they present earlier texts and the voices of the authors of those texts. The two works are not treated simply as parallel case studies but as contrasting illustrations of this broader question. Things of Internet represents an extreme case of human-authored conceptual practice based on digital collage and appropriation, while AI Imagines: Classic Literature exemplifies writing produced through large language models under conditions of platformized mass production. Setting these works side by side allows the analysis to examine how different computational tools operate as instruments of experimental writing.

Before the launch of LLMs to mass audiences, computer-generated literary texts was a relatively unknown niche often associated with experimental methods of writing, such as the usage of cut-up technique¹ or using computer-programs as means for exploring potential literature in the style of OuLiPo (Funkhouser, 2007: 33–34; Joensuu, 2012: 126–130). Against this background, the rapid increase of computer-generated literary texts has led to questions on the relationship of LLMs and experimental writing (Bertram & Montfort, 2024: 147) – is writing with LLMs, especially when using other author’s texts, inherently experimental? We explore this question in our textual analysis of AI Imagines: Classic Literature, which is a LLM-generated collection of plot summaries of canonical works, as well as short texts, in which the events of the works are set in the present day and in the future, called ‘reimaginations’.

The second work, Things of Internet (2018) by a Finnish renowned poet and conceptualist, Karri Kokko, is a conceptual work that uses screenshots to capture the visual and narrative structures of social media platforms, following the ideas of unoriginality (Perloff, 1986) and uncreativity (Goldsmith, 2011). By preserving ephemeral digital content in a book format, the work comments the influence of algorithms on user interactions and highlights the transitory aesthetics of online media. Positioned within Kokko’s broader oeuvre, the analysis underscores screenshots as a methodological and artistic tool for observing the sociocultural dynamics of digital platforms. On the other hand, its composition raises severe ethical questions especially regarding privacy and authority.

The two works are born out of decidedly different contexts. With its author marked as ChatGPT, AI Imagines represents a phenomenon of hastily put together, LLM-generated self-published works on Amazon (Hongisto, 2025). The increase of LLM-generated works of this kind has been linked to the Amazon Kindle compensation system that rewards quantity (Dietz, 2023; Cabezas-Clavijo et al., 2024), as there is an ecosystem of content creators that teach how to earn money from LLM-generated books with as little effort as possible (Knibbs, 2024; Grady, 2024). Thus, the goal of the work differs from aims typically associated with electronic literature: AI Imagines seems to be first and foremost a product, not a thought-out literary work. Kokko’s work, on the other hand, clearly aligns with experimental and conceptual literature, which by its very nature seeks to expand the boundaries of literary practice and provoke fundamental questions about the nature of literature itself (Bray et al., 2012: 1–2). In the Finnish literary context of the 21st century, experimental tendencies have often manifested as conceptual and procedural approaches. Conceptual literature, closely related to conceptual art, foregrounds the ideas and frameworks behind the work; in some cases, the concept even outweighs the aesthetic value of the finished text (Helle & Piippo, 2021: 409–410). Contrasting these two works allows for consideration of their placement in the field of electronic and experimental literature, but also highlights the tension between two radically different modes of production and intention: one driven by algorithmic automation and platform economics, the other by artistic inquiry and conceptual rigor. By examining Kokko’s work alongside AI Imagines, we can explore how literary meaning is negotiated when the motivations behind the text are at odds. A closer comparative reading of these works further reveals how the methods by which texts are produced shape the kinds of meanings they generate and the interpretive frameworks they invite. These works offer an intriguing basis for comparison, as both represent literary forms that remain extremely marginal in terms of readership and visibility, while at the same time being relatively easy to execute or reproduce in terms of their mode of production, and also easily available for the public. In a certain sense, they also share a quality of near-unreadability and randomness.

The comparison is conducted from the following perspectives: 1) techniques of text generation and composition; 2) the ‘artistic quality’ of the final products, including their readability, poetics, and literary merit. The role of AI in these comparative examples serves as a backdrop for this analysis.

Experimental and Electronic Texts

The connections between experimental or uncreative writing and electronic literature are multifaceted and reciprocal. On one hand, electronic literature draws from experimental traditions such as OuLiPo, American postmodernism, and historical literary avant-gardes, in which remixing and editing other writers’ texts has played a significant part. This influence can be seen in electronic works for example in various forms of distributed authorship: collective writing and coding projects, and the practice of sharing projects for others to refine or otherwise work upon (Biggs & Travlou, 2015). Contemporary experimentalism in literature, on the other hand, draws from digital forms of literature, such as hypertext fiction, text generators, and games, and the technological possibilities they offer and utilize (see e.g. Joensuu, 2012: 46–47, 231; Eskelinen, 2012: 538–539). Most prominently in 21^st century experimental prose, new forms include conceptualism, source text writing, the use of found material, and collaborative writing, which also produce works functioning as poetry (Eskelinen, 2012: 546). One of the main connection points in these traditions is the appropriation and use of pre-existing (textual) materials, namely collage.

AI Imagines: Classic Literature (case-study)

AI Imagines is generated with ChatGPT and the program is even listed as an author alongside Robert Doyle (of whom there is no information available on the Amazon platform). The work consists of plot summaries and present day and future reimaginations of thirty-two canonical novels, such as Picture of Dorian Gray and The Three Musketeers.² Published in July 2023, the work is generated either with the free model of ChatGPT (GPT 3.5) or the paid-for version (GPT 4).³ AI Imagines is on sale both as a print edition with 239 pages and Amazon Kindle edition. The analysis is based on the ebook version of the work.

AI Imagines provides an illustrative example of a ChatGPT generated work that is explicitly based on previous works. We analyze the intertextual relations in AI Imagines through Gérard Genettes’s concepts of hypo- and hypertexts. This refers to the relationship between two texts, where an earlier text (hypotext) is transformed into a new text (hypertext), as in the case of parodies and pastiches (Genette, 1982: 5–7). Two text types characterize the hypertextual relationship between the classics and AI Imagines especially well: summary and literary spinoffs. Both are described indirectly on the work’s Amazon page:

Through the power of machine learning and natural language processing, these intelligent systems have absorbed the essence of renowned authors and their iconic works, breathing new life into familiar stories while adding a touch of AI-inspired ingenuity.

First, AI Imagines can be conceived of as a continuation of the production of abridged editions and summarizations of classics for mass audiences. Similarly to collections of book synopses such as Thesaurus of Book Digests (1949) or reader guides from CliffNotes, the work provides an overview of the plot of each classic as well as some elaboration on the themes and influence of the work. This is supposedly ‘absorbing the essence’ of the works as described on the Amazon page. The other key element of the AI Imagines are reimaginations, which pertain to the genre of literary spinoffs. This means that the texts are based on famous literary works, which are rewritten or appropriated in a way that creates alternative perspectives, for example by placing the story in a different location or historical time period (Spengler, 2015: 11–15). As the description states, in the case of AI Imagines this means ‘breathing new life’ into the works.

Perhaps the most important finding regarding the textual relations in AI Imagines is the homogeneity of the text that persists regardless of the supposedly different hypotexts in each chapter. This homogeneity is observable both in structure and in style. Apart from the famous lines from Pride and Prejudice (‘It is a truth universally acknowledged that a man in possession of a good fortune must be in want of a wife’⁴) and The Tale of Two Cities (‘It was the best of times, it was the worst of times’), there are no direct references to the works’ style or attempts to imitate them. There are no other directly identifiable specific textual sources either, and the style resembles more that of marketing materials than historical literary works. The reimaginations are especially full of phrases of business jargon such as ‘cutting-edge technology’, ‘digital ecosystem’, ‘fostering innovation’, ‘leveraging skills’, ‘showcasing agility’ and ‘empowering individuals’. The lack of specific textual references or the paucity of literary expression is no surprise when one considers the output is based on vast amounts of data, out of which works of fiction is only a small portion (Rettberg & Wigers, 2025). It does, however, reinforce the impression that no effort has been put into the writing process of the work.

In addition to style, the summarizations and reimaginations in the work resemble one another closely in terms of content and structure. The storyworlds are especially similar in the reimaginations, where the role of digital technology plays a key part. Out of the future reimaginations, most are entirely set in the ‘digital realm’ (a phrase that is used 32 times throughout the work), and six out of the future reimaginations are space adventures. The homogeneity also extends to the reimaginations’ plots. They follow a similar structure, reminiscent of the hero’s journey (Campbell, 2004): the protagonist either embarks on a journey or is swept into action, they face various challenges, encounter allies and enemies, but ultimately triumph. While the structure fits some included works well (such as Alice in Wonderland, The Wonderful Wizard of Oz or Robinson Crusoe), it is ill-suited in the case of many others. Particularly in long works with complicated plotlines and multiple protagonists, the text is abstracted to such a degree that most of the special characteristics of the original are left out. Most often, the reimaginations only focus on the protagonist, and other characters are not mentioned or presented as unidentified ‘friends’, ‘allies’ or ‘enemies’. In the reimaginations of A Tale of Two Cities, War and Peace and Les Misérables, all of which have complex plots and a considerable number of characters, the plot is abstracted to such a degree that not a single character is mentioned. The remarkable lack of specificity concerning the plot, characters and the storyworld in the ‘reimaginations’ comes through in the following extracts:

Characters found solace, connection, and moments of respite amidst the chaos and uncertainty. Love blossomed across enemy lines, testing the boundaries of loyalty and devotion, and illuminating the inherent humanity that persisted even in the darkest of times (p. 80, reimagination of War and Peace: 80).

As the story reached its climax, the characters’ collective efforts yielded significant shifts in public consciousness and policy. Laws were challenged, discriminatory practices were exposed, and new pathways for social justice were forged (reimagination of Les Misérables: 211).

Every reimagination in the collection ends on a hopeful note, most containing a moment of epiphany, where the protagonist learns some valuable lesson. These insights are also very similar in nature, and most typically have to do with self-acceptance, standing up for injustice, or learning the value of empathy, compassion or ‘human connection’.

This process of homogenization of plot is most striking in works that depict tragic events such as deaths of major characters. In the summaries, the deaths are usually omitted – including Cathy’s death in Wuthering Heights, Beth’s (one of the four sisters) death in Little Women, Bertha’s (Mr. Rochester’s first wife) death (or even existence) in Jane Eyre, the deaths of Captain Ahab and his crew in Moby Dick, and even the deaths of Dorian Gray and Anna Karenina. Instead of depicting the tragic fates of the main characters, the reimaginations conform to the formulaic structure with the epiphany. This leads to Dorian Gray ‘championing the pursuit of genuine connections in an increasingly virtual world’, Anna Karenina ‘embarking on a journey of self-discovery’, and the creature in Frankenstein championing ‘the integration of advanced technologies with the preservation of human values’.⁵

The homogeneity of the classics in AI Imagines does not seem purposeful, as nothing in the work suggests this kind of evaluative distance to its contents. On the contrary, the formulaic structure and writing style seem to be the default mode of generating stories with GPT models that were in use at the time AI Imagines came out (Beguš, 2024: 14–15). Coupled with the broader context of the proliferation of ‘AI slop’ works appearing on Amazon, this suggests that the homogenization of the stories is a mere byproduct of using minimal prompting. The formulaic structure of the book as a whole (containing summary, present-day reimagination, and future reimagination of similar length of each work) also suggests that no rigorous attempt was made to get the program to imitate the original works, and instead very similar prompts have been used to create each chapter.

With the style, storyworld, plot and themes of all reimaginations bearing a close resemblance, AI Imagines does not meaningfully engage with the hypotexts. Apart from describing very basic elements of the work, such as naming the protagonist(s) and describing some of the events in a concrete way, the summarizations and especially the reimaginations feel interchangeable with one another. In some sense, the concept of AI Imagines as a collection of short texts limits the potential of the reimaginations to deeply engage with or resemble their hypotexts in a way that literary spinoffs typically do. Compared to spinoffs such as Wide Sargasso Sea⁶ that stand on their own as literary works, AI Imagines reimaginations are only a few pages long, and read more as summarizations of imaginary longer works than as literary texts.

However, the detachment of AI Imagines from its hypotexts cannot be entirely attributed to the text type of a summary. This becomes obvious when the book is contrasted with a work that is also a summary of its hypotext, but engages with that hypotext in a much more meaningful way. Ryan Stearne’s Old Sea that presents Ernest Hemingway’s Old Man and the Sea in a condensed form is an example of such a work. The 2000-word text is achieved by combining summarization and digital humanities techniques such as calculating word frequency (Bertram & Montfort, 2024: 112). The resulting text, while condensed, is not a plot summary but a mostly dialogue-based story that can be read independently that resembles the original work both in terms of Hemingway’s direct and clear style and in terms of plot and specific storyworld elements (the old man and the boy talking about Joe DiMaggio, the fish at the end of the work measuring 18 feet long, etc.). With its specific references and attention to style, Old Sea provides an example of how a computer-generated summary can ‘absorb the essence’ of an old classic in a meaningful way. In comparison, AI Imagines seems detached from the original works.

While in the abstract, AI Imagines is based on a multitude of textual sources, including the explicit hypotexts as well as the texts in the training data, the work comes across monotone instead of polyphonic. As the examples of War and Peace and Les Misérables show, the structural similarity between different works is so pervasive that it can trample specific references to the hypotext. The works are bent to fit the same formula, even if it means leaving out references to specific events or characters. Without identifiable textual sources and voices, the mere abstract knowledge of the vast training data behind AI Imagines is not enough for it to be considered as an example of distributed authorship, especially if the term is understood as an ‘activity of exchange that enables (creates) people and communities’ (Biggs and Travlou, 2015).

Things of Internet (case-study)

In contrast to AI Imagines, Things of Internet serves as an example of ‘traditional’ experimental or conceptual literature operating on hypo- and hypertexts in digital format. It forms part of the Works on Media series by Karri Kokko. Although the work primarily reflects artistic expression shaped by the author’s creative choices, it also embodies a highly conceptual nature, minimally modified from its raw material. The artist’s work process, much like any user’s engagement with digital interfaces, is driven by selective attention toward appealing screen views, further influenced by the active guidance of the platform.

In his extensive oeuvre, Kokko frequently addresses themes of media and mediatization. His earlier works, such as Varjofinlandia (2005), Das Leben der Anderen (2010), and Retweeted (2016), similarly utilize conceptualist approaches by appropriating text from internet sources, particularly social media. Each piece is composed by capturing online text through simple copy-paste methods, adhering loosely to an assembly principle: text segments appear in the order they are discovered, with minimal editing and no explicit sourcing (Piippo, 2019).

In Things of Internet, Kokko extends this approach to rescue digital content into a collage and a book format, an endeavor that aims to preserve ‘snapshots of the internet’s visual landscape at the end of the 2010s’:

It’s said that the internet never forgets, that search engines retrieve everything. But thematic, conceptual, and artistic selections made by humans are also necessary. This time, I wanted to capture glimpses of the visual appearance of social media and the broader web as it looked in the late 2010s. My chosen method was screenshots. For weeks and months, I browsed through social media and other online content, framing and saving screen-sized snippets as they appeared. […] Editing was limited to one purely technical principle: only what fits on a single screen, unaltered and complete, was included.⁷

Kokko’s Varjofinlandia is perhaps the most well-known among these works and has been relatively frequently studied and cited as a conceptual piece (Helle, 2019; Joensuu, 2012; Kuusela, 2015). The work comprises various sentences on depression and distress collected from internet sources, initially published on a blog created by Kokko for this purpose. The text was published in its original form, with only obvious typographical errors corrected. In both the book and blog, the fragments appear in reverse chronological order, with the newest entries first and the oldest last.

In a manner similar to Things of Internet, Das Leben der Anderen includes the source of each text segment displayed. This inclusion arose from Facebook’s then-page architecture, which structured updates in the following format: ‘[Username] x’:

John M. Bennett was tagged in an album.

  Salla Viikka is going to the circus tomorrow!

  Gary Sullivan is waiting for Segue to begin.

  Barrett Watten: Sounds like a line from W.S. Burroughs.

  Karri Kokko: Sometimes an unsent message is clearer than a sent one.

  (Das Leben der Anderen, 7)

Excluding the author’s name from each excerpt would necessitate extensive modification to produce coherent sentence structures, thus violating the principle of minimal intervention. The work is composed solely of original updates without including comments, thereby capturing the context and conversational prompts of that time and platform. The structure further emphasizes the importance of names for meaning-making within Facebook, as it transforms the text into a performance-like format. Each named user takes their turn to deliver lines to an audience composed of themselves and their peers. By formatting user-generated updates within the newsfeed, Facebook’s algorithm appears to supply a new theatrical script complete with parenthetical stage directions.

For readers, the text induces either inclusion or exclusion depending on whether they have been Facebook friends with these users and thus witnessed these lines in their original contexts. Alternatively, readers may experience it as an absurdist play, becoming an audience member within this performative piece. The title, Das Leben der Anderen, or The Lives of Others, alludes to this experience of detached observation and parallels the 2006 German film about state surveillance by East Germany’s Stasi. This comparison invokes the theme of surveillance, extending it from personal observation to Facebook’s data collection practices.

A majority of Things of Internet screenshots were taken from Facebook’s user interface, which acts as the work’s hypotext, capturing newsfeeds, individual posts, and group feeds. Notably, there is substantial use of Facebook-linked content, pointing to how platforms incorporate and even subsume other areas of the internet traditionally accessible via web browsers. Other identifiable platforms include Twitter, the Helsingin Sanomat app (the mobile application of Finland’s leading daily newspaper), and Instagram. This inter-platform capture approach highlights the transformation of internet access, reflecting the concept of Internet of Things in reverse—here, the ‘things’ of the internet are contained within interconnected platforms (Figure 1).

Figure 1: Spread view of Things of Internet. Screenshot of the work obtained with the permission of the author.

The spread view in Figure 1 captures four screenshots that would each have disappeared into the endless scroll within seconds of appearing in any user’s feed. The bottom status bar visible in each phone frame showing battery indicators, signal strength, and navigation buttons anchors each image to a specific device moment, reinforcing the ‘frozen’ quality of the material. The progress bar at the very bottom of the spread (8–9 / 482) further emphasizes discreteness: this is one bounded unit within a much larger archive. The four images create a tonal arc from left to right. The first screenshot contains a Facebook post reading ‘I was prepared to nominate Marty Feldman for president but then I remembered he died’, a darkly comic non sequitur about mortality and politics. This sits immediately adjacent to a shared article about Emily Dickinson’s Herbarium, which connects poetry, science, and death in an entirely different register. The third screenshot shows what appears to be a WordPress blog post titled ‘Tattered Sails (after Mallarmé and Broodtha…)’ – invoking canonical avant-garde literary figures – while the fourth shows geometric abstract art alongside a post by Paul Hertz about ‘Seedlings’. The juxtaposition becomes visible here as a genuine interpretive invitation: death (Feldman), botanical preservation (Dickinson’s pressed plants), fragmentation and sailing (Mallarmé’s shipwreck poetics), and germination (seedlings) form an almost accidental thematic sequence about things preserved, collected, and grown. None of these was intentional in the original newsfeed context.

The artistic motivation and process behind Things of Internet is articulated by Kokko:

I wanted to focus on how the internet appears visually. What is the fragmented visual experience we scroll through daily, spending countless hours engaging with it? What will be preserved of this visual presence over the years? Capturing the experience of browsing and scrolling is challenging, so I chose to document screenshots. My method was ordinary: I scrolled through my social media feeds and other daily internet content and captured images whenever I encountered a single screen that framed a subject or theme. I practiced appropriation in its purest form, capturing whatever fit the screen without further editing, framing, cutting, or pasting.⁸

The technique of screenshotting allows for a degree of objectivity by freezing digital moments that would otherwise be ephemeral. Unlike continuous scrolling, screenshots create discrete boundaries that enable closer inspection of how content is presented on social media. Furthermore, in each layout, Things of Internet displays screenshots in a dual-page spread, merging four images into each reading view. This structure evokes associations with the book’s physicality, inviting interpretive connections among images in a narrative sequence, leading to unexpected juxtapositions or humorous contrasts that would not emerge in a typical newsfeed. The most glaring example of accidental humor is the green loan advertisement for ‘Kuluton ensilaina 400€’ appearing smack in the middle of a page about ‘Tattered Sails (after Mallarmé and Broodtha…’) (see Figure 1). The algorithm has placed a predatory consumer loan ad directly beneath a post invoking Mallarmé, one of the most hermetically difficult, anti-commercial poets in the French tradition. Someone reading about Symbolist poetics is simultaneously being offered a loan. The juxtaposition is almost too perfect to be accidental.

The digital collage technique employed in Things of Internet differs from traditions that work to conceal the origins of their fragments. Here, the boundaries imposed by the screenshot and the screen itself remain legible throughout, rendering the relationship between hypo- and hypertext unusually transparent – the source material does not need to be excavated but is structurally on display. At the same time, this visibility reveals that the hypotext is not a unified entity: the feeds captured in each screenshot are themselves heterogeneous assemblages, combining user-generated content, algorithmically delivered material, repurposed images, shared links, and commercial intrusions from sources that remain only partially identifiable. Some elements, moreover, belong to the layer of interface interaction rather than content production — as in the text selection visible in the leftmost screenshot on page nine, which captures not just what was on the screen but what the user was doing with it, inscribing the gesture of collection into the collected material itself.

The methodology mirrors hyper-reading strategies, with Kokko’s choices reflecting selective attention to screen content rather than following links or conducting searches. According to James Sosnoski’s original definition, hyper-reading is ‘reader-directed, screen-based, computer-assisted reading’ (Sosnoski, 1999: 167), and it includes activities such as searching, filtering by keywords, juxtaposing two open windows, skimming, following hyperlinks, and extracting parts from longer texts. Thus, hyper-reading is a unique way of processing text. The reader’s gaze, guided by this logic, grows attuned to spotting narrative continuities and thematic resonances across pages, engaging in close reading through the lens of a digital landscape frozen in time. When examining the screenshots of Things of Internet, certain patterns can be observed: narrative structures of the material, methods of framing, and the formation of meaning in relation to the different elements and layers of the screen view, as well as the prevalence and possible interdependencies of these features.

Things of Internet can well be regarded as a conceptual or experimental work of uncreative writing. The author’s activity—taking screenshots—operates on the logic of hyper-reading, but instead of following searches and hyperlinks, this combining and picking gaze and activity focuses on one screen view at a time so that even those sections that are not usually the focus of attention or are at risk of becoming invisible, following Lori Emerson (2014), are prominently highlighted. Such sections include, for example, the paratexts produced by the platform that frame updates and other news feed content (‘shared’, ‘liked this’) and the partly unintended continuities they form. The reader’s gaze follows this internal logic of the work and quickly becomes attuned to seeking such narrative continuities in the screenshots. At the same time, the reader practices close reading by delving into the meanings framed by the screen—or the page. The context of the work in Kokko’s production brings the tradition of poetry into the reader’s activity, where it has often been typical to position an individual poem on one page. Thus, the idea of the page guides the reader to form meanings within the content of the page rather than through individual updates or other shares.

The digital agency displayed in Things of Internet involves both the interface’s framework and the users’ contributions. Although the authorship of the work appears distributed, as the originators of individual posts remain identifiable within the screenshots, this visibility does not translate into literary authorship in any conventional sense. It is Kokko who has selected, sequenced, and composed the work, and it is this editorial and curatorial agency that constitutes authorship at the level of the text as a whole. The original posters appear within the work rather than as authors of it. At the same time, it should be noted that the authorship of all the textual layers present in the work (Kokko’s included) is itself assemblic and deeply entangled with the digital and algorithmic environments through which these texts were produced and published; authorship here is not a property of an individual but an outcome of interactions between human agents and the digital tools, platforms, and infrastructures that shape what can be written, shared, and seen (see Roine & Piippo, 2021).

Browsing the feed is not hyper-reading in the sense presented by Sosnoski (1999), as the user does not make active choices themselves. The boundaries between the contents of the feed maintain the illusion of structured reading, but in reality, the user processes an endless stream of various text and image materials as if at the mercy of the platform. The screenshot not only stops this stream but also allows for a closer reading of how the contents are offered to us by the platform. Screenshots capture, hence, not only the interactions with these interfaces but also, implicitly, the curated, algorithm-driven content presented to the user. A distinctive feature of Things of Internet is its function as a critical distancing mechanism opposed to its hypotexts. The form of the book even in digital form, juxtaposed with the origins of its content, creates an analytical separation from the platform-specific algorithms that organized the original digital experience. In summary, Things of Internet transcends the purely artistic, providing a unique reflection on the interfaces of social media and the framing of digital interaction. Through the use of screenshots, the work explores how platforms shape user behavior and engagement. Kokko’s work suggests that the screenshot, as a technique, captures the nuances of digital media. Screenshotting not only halts the stream of content but also supports novel methods for analyzing social media.

Discussion: Unreadability, AI Slop & Uncreative Writing

From the point of view of the use of earlier texts, the two works highlight the differences between a collage and text-generation with LLMs. In Things of Internet the shift from the structural logic of the feed to the page anticipates broader questions about how boundaries, fragmentation, and juxtaposition operate as narrative devices. These questions become even more pronounced when comparing Kokko’s collage-based approach to the automated text generation of large language models.

Other users have produced most of the visible textual material in Kokko’s work. However, not all visible material is produced solely by other users, the author, or the platform. There are also advertisements and other commercial content. In most images, the boundary between two posts or other content is also visible, which is natural considering that most of the material is collected from the news feed. Such a boundary is often taken as a given threshold text in the usual browsing of the news feed, a transparent way to structure the platform’s content. However, it is a very effective means that in regular use keeps individual posts separate from each other. The boundary does not activate in the same way in a screenshot operating on the logic of the page but fades into the background so that what is captured on the screen at once forms an independent, meaningful whole. Thus, paradoxically, this fading highlights the significance of boundaries in everyday news feed browsing.

Whereas Things of Internet represents a work where different textual sources – the users, the advertisers, the platform – are recognizable, in AI Imagines distinguishing textual sources is mostly extremely difficult. The knowledge of LLMs operating in a way where texts are generated by drawing from countless earlier texts does not translate into insights regarding intertextuality in close text analysis. The text is so generic, that even the effect of named hypotexts in each chapter to the style and structure is very limited.

Despite their differences in the use of textual material, AI Imagines and Things of Internet also have similarities, as they are both examples of marginal and unreadable digital works. Both works are marginal for different reasons. The case of Things of Internet is globally marginal not only because it belongs to a niche literary tradition but also because it is predominantly in Finnish, addressing a relatively small language community. The work was published exclusively as an Issuu file through the small press Osuuskunta Jyväs-Ainola on its Blogspot-based blog. Both the blog and the file, however, disappeared from public access in 2025 – a fate common to much digital material. Furthermore, Things of Internet appears to be an outlier work even in the author’s rather conceptual oeuvre, for it is not even listed on his Wikipedia page among the rest of his works. AI Imagines, on the other hand, is somewhat paradoxically marginal because of its ordinariness: it is a very average example of the larger phenomenon of AI slop on Kindle, with no exceptional qualities that would make it noteworthy.

Both AI Imagines and Things of Internet also share the quality of unreadability. Kokko’s work is rather unreadable, as conceptual works often are, and requires a significant effort from the reader for its poetic qualities to emerge, pushing it to the margins of literary culture, and rendering its critical impact rather minimal. It does, however, possess a certain kind of narrativity, comparable to the figural narrativity of a lyric (see Prince, 2005: 387), or it could be called a quasi-narrative (Richardson, 2019: 21). At the micro level, narrativity in Kokko’s work emerges not from a premeditated plot but from the contingent juxtapositions captured in screenshots—unexpected proximities between unrelated posts in a user’s feed, or the accidental sequencing of comments by different individuals that form an unplanned continuum. This minimal, relational narrativity relies on shifts in position, absurdities, and surprising connections, which demand from the reader an active act of narrativization, or willingness to entertain its possibility (Richardson, 2019: 21–28; Säntti, 2025: 54–55).

Such interpretive labor—requiring heightened imaginative engagement and tolerance for indeterminacy – is characteristic of conceptual literature, where meaning is not given but constructed through effortful reading. In this sense, Kokko’s work foregrounds the tension between randomness and pattern, inviting readers to negotiate coherence in spaces where it is only provisionally available. AI Imagines is not ‘unreadable’ in a similar sense. No effort is needed from the part of the reader to form a coherent narrative from the presented text: as the analysis shows, the narrative form is so dominant it overrides the differences between the hypotexts. Instead of difficulty to narrativize the text, which is how unreadability is often defined (Bekhta, 2018; Alber, 2010), the unreadability of AI Imagines emerges from the numbing effect of repetition. It is difficult to find motivation for reading when there is nothing surprising to expect.

Conclusion

When AI Imagines and Things of Internet are examined in relation to repurposing and remixing of existing texts to form a new work, significant differences emerge. As the analysis of AI Imagines showed, it is not stylistically innovative or surprising in the same way as texts produced with experimental techniques. Indeed, getting stylistically unconventional text out of commercial large language models can be challenging. As Polina and James MacKay (2024) aptly put it when describing the result of creating mock cut-ups with ChatGPT, the tendency of LLMs seems to be to conform to, not to question, conventional forms of literature:

We constantly felt ourselves in the process being pushed to be nice, to make sense, to fit to a bland norm of human discourse. The cut-up […] is a tool for escaping dominant narratives that operate virus-like on our conscious minds, allowing us to perceive otherwise hidden psychic realities. The guard rails placed around ChatGPT operate in exactly this viral manner, influencing the writing of the estimated 100 million plus users who have signed up to the platform. […] the dominance of these master narratives in text generation is troublesome, and their effects in a world where AI generated text threatens to take over the internet are yet to be fully appreciated (MacKay & MacKay, 2024).

While unconventional writing with LLMs is difficult, it is not impossible. By right prompting, one can use large language models like ChatGPT in a way that resists following conventional patterns of language use. As MacKay and MacKay (2024) state, even the mock cut-ups can be created to avoid the ChatGPTs tendency for narrativising – although there seems to be significant resistance in the model for disconnectedness such as the use of repetition and nonsensical words. More generally, LLMs can be used in various creative and experimental ways. One can for example train them on different curated datasets and post-edit the outputs (Erslev, 2023), or explore new forms of writing with LLMs through dialogue format (Fülöp, 2024). This kind of experimentation is absent in AI Imagines.

At least at the moment, what seems to create the difference between literary experiments with LLMs and works like AI Imagines is that LLMs have to be modified or challenged by prompting in some way to get non-conventional outputs. The ever-wavering line of what can be considered experimental literature could perhaps be found somewhere in this area of creative play with the models. On the other hand, the mere conventionality and banality of most LLM-generated outputs does not exclude their experimental potential. There are plenty of examples of uncreative writing (in Goldsmith’s terms) that use or frame existing mundane texts in a way that invites the reader to read them differently and that pushes the boundaries of what can be called literature – Things of Internet being a poignant, yet necessarily not a very impactful, example.

While AI Imagines harnesses large language models as a largely autonomous writing apparatus embedded in platformized mass production, Kokko’s work foregrounds screenshotting and collage as tools of critical selection, framing, and preservation. Both rely on the reuse of others’ texts, yet they mobilize their respective tools toward markedly different ends. This comparison suggests that the experimental potential of computational writing tools does not reside in automation as such, but in how tools are visibly framed, constrained, and challenged within a literary project, where forms of unreadability may in fact be central to the emergence of a meaningful reading experience or productive practice, rather than an obstacle to it.

When considering the potential experimentality of AI Imagines, one should also take into account the broader context of the work. As what is considered experimental is often determined in relation to what is ‘mainstream’ (Bray et al., 2012), one has to consider how LLM-generated texts becoming more and more common affects approaching them as ‘experimental’. Against this background, the question is whether AI Imagines achieves something innovative that the hundreds of other LLM-generated books on Kindle do not. We would argue this is not the case. Advocating for experimentality of a book like AI Imagines, thus grouping it together with works that push the boundaries of literary conventions, would dilute the meaning of the term.

Competing Interests

Tuuli Vilja Vellamo Hongisto is a guest editor for this special collection and has been kept entirely separate from the peer review process for their article. Laura Piippo has no competing interests to declare.