Introduction

2Nowhere are these concerns more pressing than in education, where decisions about the integration of genAI into daily learning and writing practices will shape not only how knowledge is acquired and disseminated but also what is considered essential expertise in the first place. As genAI becomes more deeply embedded in all kinds of learning settings, the question of AI literacy takes center stage in education. Yet, much of this discourse remains focused on functional and technical aspects on one side and critical skills on the other, rather than on the deeper communicative and interpretive abilities required to engage with genAI systems deliberately.

3The shift towards genAI makes the ability to communicate with, about, and through genAI more crucial than ever, placing rhetorical literacy and communicative competence at the center of AI literacy: to become competent users, individuals must learn, first, to communicate with genAI in dialogical and co-productive ways; second, to communicate about genAI with rhetorical precision, especially because the technology’s simultaneous abstraction and concreteness invite metaphorization and misunderstanding; and third, to communicate through genAI by understanding how it generates text, in order to shape its outputs intentionally – whether by reinforcing, redirecting, or resisting its generative logics through prompt design, model customization, or linguistic constraint setting.

4Such communicative demands arise from the nature of genAI itself: a semiotic system that does not “think” or “understand” but instead imitates human communication in rhetorically plausible ways, even as its generative mechanisms differ radically from human modes of composing texts. Its success depends on its ability to break down text tokens into numbers, calculate probabilities, and translate the numbers back into text tokens. In doing so, genAI mimics persuasive, structured, and contextually appropriate language, which makes rhetorical awareness essential for navigating its outputs. The argument for the necessity of a specific Rhetorical AI Literacy (RAIL) hinges on a central claim developed in this paper: genAI and rhetoric constitute a double articulation in the sense described by Deleuze and Guattari (2005). They are inherently intertwined, functioning simultaneously as technology in the modern sense and as technè in the rhetorical sense, referring to systematic methods and skill-based practices of production. GenAI does not simply provide information; it engages humans in a certain form of rhetorical production, shaping discourse in ways that demand critical engagement.

5To sketch out such an engagement, this paper identifies a critical – and specifically rhetorical – gap in current AI literacy frameworks: they largely neglect how genAI functions as a rhetorical system capable of simulating human discourse, structuring persuasive arguments, and shaping meaning. Addressing this, the paper develops a flexible, production-oriented framework that operationalizes the concepts of technè, co-activity, promptology, and responsibility. This framework is designed to help users collaboratively navigate genAI’s persuasive affordances and exercise ethical accountability. In doing so, it enables the integration of genAI into both knowledge production and professional practice – recognizing that such processes depend not only on what AI allows us to do, but also on how we draw ethical boundaries and frame our technological understanding.

1. A gap between functional and critical frameworks

6AI literacy research has examined how the pervasive integration of AI into educational and everyday contexts affects the development of students’ competencies. Long and Magerko (2020: 2) define AI literacy as “a set of competencies that enables individuals to critically evaluate AI technologies, communicate and collaborate effectively with AI, and use AI as a tool.” Building on this, various frameworks delineate key dimensions of AI literacy, from technical understanding and application to ethical awareness (Almatrafi et al. 2024). In parallel, measurement scales for assessing AI literacy have been refined, adapting to different contexts, such as higher education (Lintner et al. 2024). These literacy approaches and scales often appear rather functional or techno-centric, focusing on users’ ability to understand and apply AI systems. There have, however, also been more ‘holistic’ models developed that try to conceptualize AI literacy as a comprehensive combination of technical knowledge, application-based skills, and reflective engagement (Knoth et al. 2024).

7By contrast, a strand of critical AI literacy emerging from the humanities interrogates the sociocultural implications of genAI. Scholars in this vein highlight the biases and power dynamics inherent in AI systems and question the purported inevitability of integrating genAI into writing and information education (e.g. Goodlad and Stone 2024). They draw attention to potential harms for learners, such as amplified biases, misinformation, or environmental costs, that stem from uncritical adoption of genAI. This critical perspective urges skepticism toward AI’s role and the motives of tech companies, emphasizing ethical and social reflection over technical proficiency.

8Both functional and critical frameworks, however, largely overlook the communicative dimension of AI. Missing is a focus on how humans interpret, evaluate, and produce communication with and through AI, especially when it comes to genAI – even though the technology increasingly behaves like an agent with goals of its own rather than a simple instrument (Pinski and Benlian 2023). A central problem in defining literacy skills and areas of competence for using genAI is that these technologies require a fundamentally new understanding of technology. Users now face a key challenge: determining the extent of co-activity between humans and genAI and understanding its hybrid status between tool and agent (Hirschauer 2016, Steinhoff 2025). Engaging with genAI, then, requires more than technical know-how or ethical principles – it demands rhetorical literacy.

9In practice, this means users must learn to critically interpret AI-generated content and communicate effectively through or alongside genAI systems. A communicative approach thus bridges the gap between technical functionality and critical reflection: it treats genAI not merely as a tool to be mastered or a threat to be mitigated, but as a communicative co-agent – not a fully autonomous speaker, yet also far from a neutral instrument – whose outputs must be interpreted, shaped, and managed within situated discourse practices.

2. Rhetorical literacy as design issue

11Since genAI at its base level functions as a communicative tool, a further extension is required: toward a conception of AI literacy grounded in rhetorical technè. In this framework, RAIL emerges as a key competency for genAI users. It shifts the focus from worrying about AI in theory to developing communication skills that help users actively make sense of and co-create content with genAI. Designing genAI systems is never just a technical endeavor; it is an act of rhetorical invention, shaping user perceptions and interactions. Interfaces must address both functional demands and ethical considerations – particularly the potential for users to overestimate or misunderstand genAI capabilities. Design theorists and rhetorical scholars have noted that interfaces and technologies function as rhetorical spaces that shape user behavior (Buchanan 1992, Selber 2004). Consequently, framing rhetorical literacy as a design issue acknowledges the integration of genAI as a wicked problem – a complex challenge with no single solution. In the case of genAI, how the system presents information and interacts with users influences their understanding and trust.

The design of machine technology, then, can itself be defined as the art (techne) [sic] of deception. The ethical question is then not if designers should deceive, it seems, but rather what kind of traps, cheats, deceits, and fooling we need and want – to deceive ourselves and have ourselves deceived. (Coeckelbergh 2018: 83)

12From this perspective, design and technology fundamentally involve deception, which is structurally integral to modern media. Instead of deeming all such deception unethical, Natale (2021) distinguishes banal and strong deception in AI, and Umbrello and Natale (2024) build on this to argue that designers can harness certain mild, inevitable illusions for social good. Treating genAI outputs as rhetorical texts, a critical evaluation of content and intent means guarding against manipulation and misinformation. The design of the human-machine interface ultimately determines how beneficial it is for users, as well as how much the technology or its developers may mislead them. The extent to which human-machine interfaces are shaped for specific contexts and user groups further highlights why this is a particularly pressing rhetorical issue for genAI – and, conversely, for generative-algorithmic production within rhetoric.

3. GenAI as rhetorical system

14Yet rhetoric has, in many ways, always functioned like a “machine” for text production. From antiquity onward, rhetorical training has relied on formulaic approaches that draw on established knowledge and topoi, exemplified by doxa or imitatio. Over the centuries, scholars such as Ramon Llull or Erasmus of Rotterdam envisioned ways to automate these processes, long before the modern notion of the solitary creative author took hold. By the mid-twentieth century, experiments with recombinative writing further illustrated rhetoric’s mechanical potential (e.g., Lutz 1959). With the advent of digital rhetoric, this tradition has shifted from recombination to connection. As Johnson-Eilola (1999) anticipated, the old “production paradigm” of authorship is now on the brink of being replaced by writing as a “fragmentary connection,” in which new meaning arises through relational linkages. Bajohr (2020) views such a connectionist paradigm as fully realized in genAI, where massive neural networks trained on vast datasets generate text in ways that humans can no longer fully comprehend.

3.1. Double articulation of rhetoric and genAI

17And there is another level to the connection between both rhetoric and genAI: the systems themselves exert a persuasive influence. Within the production pipelines of genAI, rhetorical structures emerge – whether in the form of political preferences (Batzner et al. 2024), biases (Alford 2024), or other communicative byproducts, such as metacognitive knowledge. As co-producers of communication, genAI systems are not mere tools; they rely on rhetorical principles and strategies to produce rhetorical forms. Their operation is imitative in at least three ways: they generate semantically plausible but often only superficially coherent texts; they produce a surplus of potential meanings and diffuse semantic possibilities; and they operate across language and data, constructing hypothetical relations rather than asserting stable truths about the world. As rhetoric, also genAI becomes fully actualized only in situated practice, where persuasive form meets contingent meaning and effect (Gottschling, in press). Drawing on Deleuze and Guattari (2005), we might say that rhetoric and genAI operate in a double articulation, tightly bound both analytically and productively, always including and expressing each other.

18From the vantage point of double articulation, the rhetoricity of paradigmatic usage situations of genAI becomes more apparent: genAI systems are marketed as personal assistants for everyday tasks at work or at home, but also increasingly as co-tutors or academic aids. Psychosocial chatbot applications are on the rise, raising serious concerns about their mental health impacts on young users. Other genAI tools function like advanced search engines, enhancing and disrupting a well-established technology; and future “reasoning” features may yield high-end – and high-cost – genAI agents for academic and professional scenarios. Despite the variations in usage, each scenario features a dialogic structure: genAI acts as a co-active interlocutor that can affirm, correct, or broaden users’ perspectives and texts. Rhetorically, it provides text for further re-contextualization – even if only to update users’ own cognitive frameworks.

19This, naturally, has consequences for all learning processes. Integrated across all classical officia oratoris – from inventio to actio –, genAI could fundamentally reshape how learning to write or even to critically assess is taught. Whether used deliberately as an educational tool or informally by students, it generates new forms of co-dependence between learners and technology. As such, the role of rhetorical training becomes increasingly important. The rhetorical structure of genAI systems highlights that, in developing RAIL, rhetoric should not be viewed solely as a domain-specific skill for professional communicators, e.g., in science communication or political deliberation. Rather, it serves as a core element of the pragmatist stance – mediating between the extremes of techno-functional and ethical-critical literacy and emphasizing practical, behavior-oriented use. While the pragmatist stance underscores rhetoric as central to understanding and using genAI in practice, a critical rhetorical view clarifies the broader social and ethical stakes of the double articulation. Vallor (2024) notes that genAI’s persuasive power can mirror patterns of exclusion and exploitation, while large-scale training data embed social and cultural biases that privilege dominant perspectives. At the same time, genAI’s tendency toward “common” statements evokes the rhetorical concept of doxa, and its imitative processes resemble the use of imitatio auctorum – though without human input and feedback, they risk becoming empty replication (Gottschling, in press). Balancing these rhetorical forces of reinforcement and reinvention is therefore vital to RAIL. Consequently, a key objective is to operationalize rhetorical agency as a production of probable statements for the technological production path of genAI, both theoretically and conceptually, and as part of personal communicative competence.

4. A question of technè

20Shifting to a praxeological level implies that anyone who chooses not to use genAI can find numerous, well-founded reasons to abstain. Nonetheless, scientific inquiry into the practices governing the interplay between humans and genAI must continue. After all, in-depth studies of these artificial “semiosis machines” could reveal, first, how rhetorical structures and apparatuses make texts possible and bring them into being, and second, how we humans engage with text, what assumptions we make, and where the “human element” in text production truly lies. Third, such systems must be examined not only in terms of their origins or cultural embedding but also evaluated in real-world contexts for their practical utility.

4.1. Preferred pedagogies of AI and rhetoric

22Although such a baked-in character initially appears commendable, it does not change the fact that the statements the system is capable of producing are still only probable and not clear, accurate or helpful – thus neither training nor system prompts can eliminate the mere imitation of human text-production. Second, the preferred pedagogies of these systems manifest themselves not only in their pedagogical implications for users acquiring writing skills (Cummings 2024, Gupta and Shivers-McNair 2024), but also in how genAI itself acts as an instructor – that is, how it applies its foundational settings in interaction. Claims of pedagogical neutrality for AI systems are untenable (Basgier 2024). Rather, these tools rely on metacognitive illusions that function as apparent “knowledge about knowledge” and proclaim neutral transmission of information or composition strategies, even where their outputs and writing methods prove problematic due to a lack of situational and contextual awareness. Consequently, although genAI systems are inherently rhetorical, they cannot consistently enact a coherent rhetorical pedagogy. Because they are always text production machines first, effectively engaging with genAI systems chiefly demands from the user expertise in constructing and performing texts – in other words: rhetorical skills.

4.2. Rhetoric as technè

24In contrast to an ethically fundamental critique, a pragmatic perspective on genAI use acknowledges the need for ethical reflection but situates it rather within rhetorical production. From this standpoint, “opposing the use of AI ‘on principle’ is nonsensical in pragmatist terms,” since decisions to redesign or resist AI systems must be context‑specific and aligned with the values and goals of the communities they affect (Watson et al. 2024). Overall, the emphasis remains on ensuring that text produced with genAI is appropriate to each unique rhetorical situation.

25Indeed, such a rhetorical understanding that largely detaches ethics from rhetorical production can be found in antiquity – an approach that has been criticized accordingly as an “inadequate, if not unethical, way to understand and teach writing” (Pender 2011: 3). This view sees rhetoric as technè rhetorikè, the title of Aristotle’s work that established rhetoric as a dual science of producing and analyzing persuasive discourse. Triangulated between art, craft, and skill, technè as a general technical term has no direct modern English equivalent (Pender 2011: 4), yet according to Schatzberg (2018: 16), it stands “at the core of our preeminently modern term technology.” It is always linked to poiesis, the goal-directed creation of a product serving purposes beyond itself. Rhetoric as technè thus refers to the deliberate, process-oriented production of persuasive communication. Over the centuries, handbooks, method collections, and practical guides have been coined as technè. It therefore also denotes the non-theoretical, pedagogically production-oriented part of rhetorical science: “Techne [sic] was fundamentally about how to do things, ‘knowing-how rather than simply knowing-that.’ Furthermore, techne was not innate but teachable” to everyone, not only to an elite few (Schatzberg 2018, 18) – outlining a broad pedagogical corridor:

[R]hetoric was to be understood as an instrumental form of discourse, […] the point of studying it was to better understand the processes of producing it, and […] the point of better understanding the processes of producing it was to create explicit methods for teaching it (Pender 2011: 6).

26Although such an understanding of rhetoric may be too limited as a comprehensive scholarly agenda today, it is well-suited as a pedagogical impetus for developing RAIL: first, technè in rhetoric focuses on creating a specific communicative product intended to persuade in the real world; second, a rhetorical understanding of a complex tool can be merged with a technological understanding of genAI; and third, rhetorical history offers a productive, situation-sensitive concept of technè that can be adapted for today’s context. The following sections explain how these three levels come together.

4.3. Three levels of technè

27On the first level, Aristotle’s conception of technè consistently directs attention to the act of making itself. The essence of technè – and, by analogy, technology – lies in serving a purpose beyond its own existence and thus carries a kind of moral neutrality in its methodology. Because Aristotle (1962) distinguishes poiesis (production) from praxis (performance), responsibility for how a communicative artifact is used ultimately falls to the individual actor and no longer pertains to the technè of making. As Schatzberg (2018: 21) concludes, “Aristotle in effect excluded all ethical content from the process of making itself.” Only once a concrete work has been created, existing outside technè, does the need for ethical scrutiny arise: the procedure itself is neutral, but the person who produces and presents it to the world is accountable and must be able to give reasons.

28In the context of genAI, communicative artifacts are co-produced by both technology and humans, making each party jointly responsible for the poiesis. While one might from Aristotle’s point of view expect a “true technician” (Schatzberg 2018: 18) to explain the rationale behind the production, genAI’s reasoning is neither precise nor traceable to any single source. Instead, it emerges from a complex model pipeline that encodes human co-intentionalities – manifested as biases in training data, developers’ design decisions, and reinforcement‑learning feedback. A heightened ethical-critical awareness of these co-intentionalities is therefore advisable for those who produce content under the rubric of technè. However, as part of technè, the way in which statements are produced still falls within the realm of the pragmatist “maker’s knowledge” (Floridi 2018): technè negotiates and subsumes practical knowledge about both the technological foundations of these systems and their ethical implications.

30A third level of technè combines understanding and judgment. It can be drawn from a historical framework that grants primacy to rhetorical performance, serving as a link between technè as poiesis and concrete, situated praxis. The approach stems from Isocrates, famously criticized by Plato for lacking a strict philosophical focus. While Isocrates’ theory and published output lacks the written record of a technè that we find with Aristotle, his implicit method points to a “culture of imitative performance” (Haskins 2006) that resonates with the reflections on imitatio in Quintilian (Inst. or. X, 20) and the generative-imitation principles shared by rhetoric and genAI. In Isocrates, this model is enhanced by a competence-based perspective on rhetorical learning:

For ability […] is found in those who are well endowed by nature and have been schooled by practical experience. Formal training makes such men more skilful [sic] and more resourceful in discovering the possibilities of a subject […] (Isocrates 1929: 173).

31Now, genAI, by effortlessly supplying commonplace ideas, styles, and audience appeals, seems to undercut Isocrates’ emphasis on talent and experiential learning as capabilities that resist formalization in any one competence model. Particularly given how genAI can deceptively appear to demonstrate competence to untrained users, competence-oriented learning for genAI might appear to prioritize formal training in rhetorical-technical skills. These systems present themselves as skilled writers and endlessly patient tutors, implying they can bridge the divide between novice learners and cultivated rhetorical expertise. In doing so, their imitative competence replicates the interplay of innate talent and practical experience – simulating the outward markers of expertise without engaging the situated practice that underpins genuine skill.

32Paradoxically, this apparent simplification for learners – requiring neither talent nor human teachers – underscores the importance of an enriched practical experience. The user’s ability to critically evaluate, select, and deploy rhetorical strategies becomes even more crucial, as we can learn from Isocrates, as he “knew how important practice and detail were for the development of flexibility, for they form the basis from which one is able to take advantage of the kairos” (Papillion 1995: 152). Through distinguishing between oratory as performance and grammar as a technical foundation, Isocrates (171) highlights situational awareness as “the qualities of fitness for the occasion”. Regarding genAI, this relation shows that formal competence cannot be reduced solely to ethical criteria or technological solutionism; rather, practical experience must be gained across multiple dimensions of rhetorical aptum, such as accuracy, style, audience awareness, and genre specificity, led by instructors who provide exempla and real-world insights.

4.4. From technè to educational scenario

33From this, the key question arises as to where, exactly, genAI is used and which specifically rhetorical competencies are needed to be taught for context-appropriate application. The concept of technè developed here points back to the double articulation of genAI and rhetoric, even if some usage scenarios ‘deceptively’ conceal that connection. In other words, rhetorical-communicative techniques remain active even where they are not immediately recognized or reflected upon – such as in seemingly purely technical applications, interfaces, or decision-making processes. Accordingly, rhetoric must not only intervene analytically and critically but also act pragmatically and practically, to expose and make manageable the hidden rhetorical efficacy of such systems for users.

34Following from the argumentation in this chapter, genAI appears above all as a technological embodiment of human-like cunning. Precisely here, in the rhetorical handling of this new class of communication machines, the practical experience of humans as polymechanos – in the sense of technè as a generalized awareness of skill – becomes especially significant. Hence, technè-driven RAIL involves not merely technological-analytical competence but, above all, rhetorical text competence that also reflects on how texts are performed and understood in specific situations. As rhetorical statements emerge from the co-dependent interplay of human and machine, the next chapter sketches how a concrete framework for technè-oriented rhetorical competence might look in the context of genAI and identifies the key competencies at its core.

5. Four dimensions of RAIL

35RAIL is not domain-specific – that is, it is not aimed exclusively at rhetoricians or professional communicators. It however supplements existing AI Literacy categories by actively incorporating specific rhetorical competencies and mechanisms of text production and communicative performance. Due to the double articulation and following a pragmatic stance, the technological and critical dimensions of AI competence are here subordinated to a rhetorical structure. Because the products of genAI are fundamentally rhetorical in nature, competence in writing – as a matter of technè – and competence in situationally appropriate performance of communicative acts – as a matter of aptum and iudicium– are both essential. Learners who acquire RAIL are thus equipped to communicate effectively with, about, and through genAI, while instructors are challenged to enable individuals to recognize, critically question, and skillfully employ genAI systems and their outputs. For instructors, this especially means that practical experience must be conveyed through rhetorical teaching as well as hands-on engagement with the technology.

36In what follows, four key dimensions and structures of RAIL are identified. They may serve as a foundational framework – one that can be applied through formal training and practical experience and examined reflexively in research. Some of these dimensions emphasize technè as the foundation of rhetorical competence, while others focus more on contextual iudicium or a situation-sensitive aptum approach to texts produced through genAI systems. They all share the requirement that only through an integrative, practice-oriented approach can they be both productively employed and analytically examined.

5.1. Co‑activity and accessibility: AI as a rhetorical partner

37RAIL engages with the question of what it means for a computer to assume partial or full control of text production (Steinhoff 2025). Earlier sections emphasized the interface as the locus of human-technology interaction (Umbrello and Natale 2024, Hirschauer 2016). On closer inspection, such technological co-agency comprises diverse co-intentionalities and strategies that shape AI’s imitative text production – initiated by human prompts and refined by human decisions – making genAI systems full-fledged communication partners. Authorship and meaning are no longer localized to a single individual creator but emerge through the interplay of human and machine – a co-activity that can be described as the co-production of possible worlds, shaped by intention, interpretation, and persuasive structure (Gottschling and Kramer 2025).

38Even when confined to a chat window, co-produced text rhetorically influences participants: interface design, chatbot behavior and AI-generated language shape human perceptions and judgments. Although more intuitive than search engines, genAI’s human-like “I” is only a simulation; its apparent familiarity masks a fundamentally technical process. Human writing and algorithmic generation share real‑world contexts but remain distinct activities (Steinhoff 2025, Basgier 2024). This deceptive familiarity underscores the need for learners and instructors to recognize genAI’s persuasive power and treat co-produced text as a preliminary draft – technè – that demands human refinement. Conversely, chat output can feed back into the system, since user inputs – unless excluded by policy – may train or adapt the model further. This mutual adjustment as double articulation can be harnessed to create specialized AI agents via targeted prompts and corpora, yet it also raises privacy risks when personal data reemerges in altered form.

39Ultimately, effective co‑activity requires users to remain critically aware of AI’s rhetorical agency – treating AI‑generated content as a provisional draft that demands continuous evaluation, contextual adaptation, and ethical responsibility. For educators, this means designing instruction that explicitly cultivates learners’ ability to recognize AI’s rhetorical agency, critically evaluate and adapt AI‑generated drafts, and iteratively refine them through context‑sensitive rhetorical decision‑making.

5.2. “Bullshit” generators and promptology: managing AI’s rhetorical illusions

40For users, it’s crucial to recognize that genAI systems offer only a starting point for communication. Warnings like “ChatGPT can make mistakes. Check important info” understate the fact that genAI will inevitably err – and that unpredictability is precisely what makes it effective at generating plausible statements. Lacking consciousness or moral judgment, these systems produce language through statistical modeling, rendering categories like “true” or “false” irrelevant. Instead, they provide statements that aren’t clearly correct or incorrect but merely plausible – what Frankfurt (2005) calls “bullshit.” Inevitably, genAI will generate misinformation, bias, or even plagiarized content without any awareness of these failing to meet human standards.

41From a technè perspective, the human user remains the essential mediator between lived reality and the AI’s static training data. Although technical improvements or integrated web searches may refine outputs, genAI lacks sensory perception and genuine emotional understanding; it operates solely on connectionist patterns. Effective prompting therefore requires users to supply precise, context-rich information.

Formula for effective prompts

Intention/Task: What exactly should the system do?

Context: What role does the prompting individual play? What role should the AI assume, and which audience is being addressed?

Format: How and through which medium should the target audience be approached? What text genre, length, or output format is desired (e.g. plain text, table, bullet points)?

Relevant Information: Which key details about the prompt situation should be made explicit (e.g., location, timing, purpose, participants, and their roles, etc.)?

42However, prompts are not stable, and identical prompts will yield different results over time. By iteratively refining prompts, users can guide AI-generated text – minimizing, managing, or creatively reshaping the “bullshit” genAI produces. Prompting strategies should align with task complexity. In practice, prompts approximate but cannot replicate real communicative exchange. Since genAI does not perceive reality but relies on a generalized model, users must remain aware of the gap between actual context and AI’s imitation. Deliberately leveraging that gap enables more effective, context‑sensitive use of genAI outputs.

5.3. Rhetorical usage practices: adopting and adapting

43GenAI systems can serve functions ranging from verifying existing knowledge to exploratory search. From a rhetorical perspective, users should be cautious about over-relying on genAI to fill knowledge gaps, given its opacity and propensity for errors. Instead, it is more effective to deploy genAI where users can spot and compensate for its shortcomings with their own expertise. In this co‑activity, the distribution of roles is essential. Ideally, genAI should neither act autonomously nor serve as the sole instructor but function as a cooperative partner – or even a learner – that can receive and offer feedback. This collaborative model combines text‑production skill (technè), the user’s critical judgment (iudicium), and situational appropriateness (aptum).

44The rhetorical practice of imitatio auctorum shows that creative variation unlocks new stylistic possibilities (Gottschling, in press). GenAI operates similarly: its creativity lies in reshaping existing content. To make these variations rhetorically productive, users’ critical judgment is paramount. Following Quintilian’s view of rhetorical education, genAI use should treat imitation as a tool for innovation while remaining mindful of the system’s limits. Domain experts especially can detect errors or weaknesses when AI exceeds its competence, producing hallucinations or incorrect results. Educators can support learners in this dimension of RAIL by designing scaffolded, co‑active exercises that position genAI as a collaborative partner in familiar situations – prompting learners to iteratively generate, imitate, and creatively vary AI‑produced text while feeling comfortable critically evaluating, correcting, and refining its output.

5.4. Human responsibility and contextual accountability

45Although genAI systems can significantly enhance communication, the responsibility for their outputs rests entirely with users. Users must ensure that any content generated is accurate, appropriate, and aligned with their objectives. Technology companies explicitly disclaim this responsibility, even though their systems inherently reflect human decisions, biases, and values.

46This situation prompts a critical question: to what extent can a practice‑oriented RAIL mitigate the standardizing tendencies of genAI? In other words, how can we use these tools not merely to reproduce prevailing norms but, because of rhetorical technè, to create inventive “social possibilities” (Pender 2011: 16) that challenge existing power structures within genAI (Vallor 2024)? Too often, genAI produces polished but predictable text – stereotypical, prepackaged patterns that reinforce familiar clichés and biases. Such outputs constrain creative and critical thought and risk perpetuating mainstream viewpoints and colonial frameworks.

47This makes it even more important to harness the communicative potential of genAI through a critical, rhetorically informed approach – while also fostering users’ critical thinking. The aim of RAIL should be not merely to replicate familiar patterns when dealing with genAI, but to deliberately introduce ambiguities and challenges that can unsettle the rather rigid default ‘mindset’ of a genAI system. Building on Annette Vee’s perspective, one might describe this as the goal of a future writing and communication practice:

In the current/future of AI writing, how do we avoid producing stochastic students or becoming language models ourselves? […] If a student is taught and rewarded for commonplaces and stock genres, they will reproduce their training data: the boring commonplaces no teacher relishes reading and no writer learns from reproducing. Instead, we should teach and write for perplexity […] to avoid the commonplaces that block critical thinking (2023).

48Ultimately, RAIL must clarify user responsibility and equip both instructors and learners with strategies to engage with genAI in ways that are thoughtful, challenging, and creatively generative.

Conclusion. Researching for perplexity: future research topics and pedagogical strategies

49Returning to the initial question of which problem ChatGPT – or any text-generating AI system – solves regarding literacy and rhetoric, the argument presented here suggests that genAI can be both a problem and its own solution. Through their dual rhetorical-technological structure, these systems pose ethical and communicative challenges, yet they also enable productive outcomes in communication and education contexts. By reviewing AI Literacy concepts, this paper has underscored the need for systematic research and deployment of a specific RAIL that unites rhetoric and genAI. Focusing on the concept of technè it examined how text emerges from the co-activity of humans and machines, leading to four key dimensions of rhetorical competence in genAI.

50Advancing RAIL as a research program and pedagogical practice requires a focused and multi-dimensional agenda. Several interrelated questions now emerge that must guide future inquiry. First, we need valid and transferable metrics for assessing Rhetorical AI Literacy – both as an independent competence and in relation to existing AI literacy frameworks. Second, it remains to be clarified whether promptology should be developed as a distinct research domain or situated within a broader communicative model grounded in dialogue and reciprocity. Third, the field would benefit from a refined methodology of rhetorically analyzing and categorizing outputs – capable of distinguishing hallucinations, creative inaccuracies, and metacognitive illusions – to more precisely describe both common errors and productive ambiguities. In addition, any rhetorical framework must remain responsive to shifting user practices, evolving technological capabilities, and changing production ecologies; flexibility and situatedness are not optional features but defining conditions of a viable literacy model.

51Moreover, rethinking technè in the context of genAI offers an opportunity to bring classical rhetorical theory into conversation with contemporary design questions, particularly around agency, intention, and use. At the same time, any pragmatic implementation of RAIL must be accompanied by concrete guidelines that help users employ generative systems deliberately, ethically, and in contextually appropriate ways. This also calls for sustained research into how genAI is reshaping everyday information practices – how users interpret, evaluate, and act upon machine-generated content across diverse social and professional settings. Finally, and most pressingly from a pedagogical standpoint, we must ask how AI can be integrated into teaching and learning in ways that enhance rather than erode critical thinking, responsibility, and rhetorical awareness. These are not marginal concerns but constitutive of a rhetorical approach to AI: they define the space in which a literacy of the artificial becomes a literacy of the possible.

52By emphasizing RAIL, this paper not only acknowledges the complex interplay between technology and rhetorical practice but also provides a framework for engaging with it more productively. The goal is to cultivate a communicative practice that does more than replicate existing patterns. Rather, it should spark perplexity, encourage critical reflection, and inspire genuine innovation.