Beyond the Metaphor Trap: How AI Compels Teachers to Reimagine Voice, Authenticity, and Writing Instruction

The Metaphorical Framework of Writing Instruction

As writing teachers confront the challenges of artificial intelligence in their classrooms, the metaphors we've long used to teach writing take on new significance. These metaphors have shaped how we conceptualize and teach writing, but they may now be hindering our ability to adapt to technological change.

Lakoff and Johnson (1980) argued that metaphors are not simply linguistic devices but fundamental to human thought. Metaphors enable individuals to understand abstract concepts by mapping them onto familiar, concrete experiences. For example, metaphors like "time is money" shape how people conceptualize time as a resource to be spent or wasted. This theory of metaphor suggests that metaphors arise from embodied experiences. For instance, physical orientation (e.g., "happy is up; sad is down") reflects how humans associate emotions with bodily locations. Metaphors often form coherent frameworks of thought that influence broader conceptual frameworks. For example, the metaphor "arguments are war" leads to expressions like "defending a position" or "attacking an argument," structuring how debates are perceived and conducted¹.

Having taught writing from elementary school to advanced education, I understand how important metaphors in writing instruction. They are used to anchor teaching practices and to scaffold learners’ comprehension by making abstract concepts more accessible. For example, teachers use metaphors like "writing is gardening" to evoke learners’ understanding that texts can be nurtured or cultivated, that texts need to be weeded and fertilized to produce flowers and vegetables. Recent studies explore how metaphors frame perceptions of AI (e.g., AI as a “stochastic parrot” or “co-pilot”), shaping ethical considerations and expectations for its use in education and beyond².

But metaphors can also lead to confusion and misunderstanding. A writing teacher who tells students “writing is like pouring water in a bottle” provides a kernel of pragmatism in that water is substance, a bottle is a container, and writers do put substance into containers. By extension, readers drink texts to satisfy their thirst, and to keep their body healthy. This metaphor of "writing is like pouring water in a bottle" illustrates common problems with metaphorical writing instruction. While it appears to offer a concrete way to think about writing, examining the metaphor reveals serious limitations and potential misconceptions.

First, it suggests writing is a simple, one-directional transfer of pre-existing content into a fixed container. This misrepresents how writing actually works. Writers don't simply pour fully-formed ideas into predetermined forms. Instead, the act of writing itself often helps develop and clarify thinking. The form or genre writers choose influences what they can say and how they say it.

Second, the extension of the metaphor to readers "drinking texts" becomes particularly problematic. It implies passive consumption of content rather than active engagement with ideas. Readers don't absorb writing like water. They interpret, question, and construct meaning based on their own knowledge and experience. The suggestion that texts function like water to "keep their body healthy" oversimplifies both the purpose and process of reading.

More fundamentally, this metaphor fails to acknowledge the social and contextual nature of writing. Unlike pouring water, writing involves complex choices about audience, purpose, and situation. Writers must consider not just what they want to say, but how their choices will affect different readers in different contexts. Water doesn't change based on who drinks it or why, but writing must adapt to its rhetorical situation. This analysis clarifies why writing teachers must be cautious about using metaphors in writing instruction, even when they seem to offer helpful concrete comparisons. The risk is that students will take the metaphor too literally and develop misconceptions about how writing actually works.

I’ve been seeing an intensification of the quest for the grail of "authentic voice" as a core writing instructional learning outcome now that AI has become a problem. For me it reveals a paradox that has longed perplexed me in the common sense understanding of writing I’ve observed in advanced degree teachers enrolled in my graduate course on composition theory. Middle school teachers are often passionate about helping students "find their voice," describing romantic moments when young writers suddenly take wing and soar with self-expression. Brave high school teachers push back, risking their reputation, arguing that mastery of academic conventions, grammar, structure, and critical thinking, matters far more than a soaring voice, especially for students who need these skills to access future opportunities.

Meanwhile, elementary teachers, often practicing without specialized training in writing instruction, take in these perspectives and decide that nobody knows what writing means as a subject to teach—or that, like traveling through the looking glass, writing means whatever I want it to mean. They focus pragmatically on basic skills like reading comprehension and summary writing, viewing voice as a luxury their struggling students cannot afford.

When Teaching Metaphors Fail

Writing instruction is riddled with metaphors. Writing as finding and planting a "seed" metaphor suggests that ideas naturally grow and flourish if we just plant them correctly and nurture them. Teachers might tell their students to "plant the seed of their main idea" and "let it grow" through brainstorming and drafting. While this organic metaphor acknowledges writing takes time to develop, it misleads by implying writing is a jack-and-the-beanstalk affair, a search for a magic bean, that it happens naturally through nurture rather than through deliberate strategic choices. Students might wait passively for ideas to "bloom" instead of actively shaping their writing for specific purposes and audiences. When there are no blooms, what conclusion do these writers draw? “I’m not a good writer.”

The "muscle" metaphor presents writing as something that strengthens through repetitive exercise, like building physical strength. Peter Elbow³ used the muscle analogy effectively in his discussion of biceps as the creative muscle and triceps as the judgmental muscle which, when applied at the same time, creates writer’s block. But when the metaphor gets extended to include our brains, it loses steam. The human brain is not a muscle. In the age of AI and the burgeoning understanding of the complexity of the brain, any suggestion that simple exercises makes the brain bulge like a hypertrophied calf muscle collides with understanding how the brain actually works (for an opposing view, see Scott Young⁴).

"The more you write, the stronger your writing becomes" is a common refrain emerging from the muscle metaphor. Teachers arrive at this position through extending the muscle metaphor and privilege “free writing time,” often as long as twenty minutes of class time devoted to journal writing that no one ever reads. Researchers following the insight of George Hillocks⁵ have debunked this prescription as a dangerous half-truth, though it makes good sense according to the muscle metaphor. While practice matters, this metaphor suggests improvement comes automatically through repetition rather than through developing analytical skills and environmental awareness. Students might—and often do—produce large volumes of writing without learning to analyze writing situations or make strategic choices.

The "conversation" metaphor frames writing as simply joining an ongoing chat. Teachers encourage students to "add their voice to the conversation" or "respond to what others are saying." This metaphor can help emphasizes writing's social nature, but it can also oversimplify the complex demands of written communication. Students might focus on stating opinions without learning to craft carefully structured arguments that work within specific contexts. A steady diet of “free journaling” could amplify this type of “adding my voice to the conversation.”

Each of these metaphors, like "authentic voice," offers a partial truth about writing but becomes problematic when treated as an essential part of a complete model for instruction. When we teach through these metaphors rather than teaching writing as a social and cultural task that must be undertaken in the real world, we risk leaving crucial writing experience to chance. Students need functional strategies for analyzing writing situations and making effective choices, not rich metaphors about how writing works. To learn them they need teachers who grasp writing as a tool-based craft like conducting a science experiment is tool-based or inferring the causes of a historical event from primary source documents is tool-based. In this context I’m not convinced that “tool” is a metaphor.

AI compels teachers to be clear about what we mean and how what we mean plays out for the writer and for the reader. The middle school teacher who celebrates a student's emerging voice may really be celebrating writing that mirrors the teacher's own values and style. The high school teacher's insistence on academic conventions may reflect their own journey to master professional discourse. The elementary teacher's focus on basic skills might stem from their own uncertainty about what voice means in writing development.

This confusion about voice in writing instruction points to a larger issue: In trying to help students "find their voice," we may actually be trying to create writers who validate our own beliefs about what good writing should be. The Pygmalion effect in English classrooms, where the teacher works to create a perfect reflection of the ideal writer they envision, may in the end bring about a different result. Rather than teaching concrete writing skills, we risk turning writing instruction into a project of personal transformation based on our own idealized notions of writerly authenticity.

AI's Challenge to Traditional Writing Instruction

The introduction of AI makes these metaphorical approaches even more problematic. AI can "plant seeds," "exercise writing muscles," and "join conversations," but what matters to humans is the strategic deployment of writing tools to achieve specific purposes with particular audiences. For example, I prompted Claude to write a response to this post’s thesis that authentic voice has become a problem for writing instruction. I prompted Clause to generate a response in a “voice” you might “hear” in a grade school teacher’s lounge:

CLAUDE: *Sigh* So here we are, all gathered in the lounge again, talking about helping our precious little writers "find their voice." *Making air quotes* Isn't that just adorable? Twenty years I've been teaching kids to "discover their authentic voices" through writing, and now - well, now my sixth graders are using ChatGPT to write five different versions of their book reports, each one with a different "voice."

You know what's really rich? Yesterday in our PD session, we spent two hours discussing how to nurture student voice, while half the kids in my class were probably at home asking AI to "write this in the voice of a passionate middle schooler." *Takes a long sip of cold coffee*

Remember when we actually thought we were helping them discover something real and meaningful? Now I'm watching these AI programs spit out more "authentic voices" in five minutes than I've supposedly helped students "discover" in my entire career. Makes you wonder what we've really been teaching all these years, doesn't it? Maybe next week's PD can be about helping students find their authentic AI prompting voice. *Stares into empty coffee cup*

But hey, at least we can still put "Voice Development Specialist" on our teaching resumes, right? That should count for something in this brave new world we're creating.

With experience, teachers can learn to identify the tells in AI prose well enough to set off metaphorical alarm bells. They need not accuse any student of plagiarism to ask them to explain their writing choices using AI in a nonthreatening setting with the goal of learning. The AI authorship of this teacher lounge monologue reveals itself through evidence of artificial construction.

This ability of AI to generate multiple voices aligns with Bakhtin's concept of heteroglossia, the idea that we all speak in multiple voices shaped by social context. What AI makes explicit through its generation of varied voices is what humans do naturally: Adapt our expression to different social contexts and purposes

The opening (“here we are, all gathered in the lounge again”) announces the setting in an artless, feckless manner as if to say “Dear reader, what I am about to write takes place in a teacher’s lounge.” The function of the information serves the needs of a reader, but why would this fictional teacher open with the equivalent of “We are gathered here today to confront the challenge of teaching our students to find their voice?” Do people talk thusly in a teacher’s lounge? Already the writing reveals itself not just as fiction, but as artificial fiction—or bad human writing.

Physical gestures read like stage directions—the *Sigh*, the *Making air quotes*, and the business with the coffee cup feel awkward rather than plausible. Remember, AI has not been in a teacher’s lounge, absolutely crucial to acknowledge once the intuition of AI in the background is aroused. In this context, when I use the words “feel” and “awkward” to describe the tools I use to discern AI output, I’m not stretching the metaphors: Intuition (“feel”) is an abstract concept, but it has a primary, shared meaning as “sense” or “intuit.” “Awkward” isn’t referring to physical activity, but is used a clinical term meaning “clumsy, sloppy, illogical.”

AI output always centers its artificial prose on the overarching topic, although experts in AI technology are refining systems to "take more time to think" and achieve more lifelike output. In this monologue, the repetitive, ironic use of the word "voice" and the careful building of sarcasm exemplify typical AI behavior—the dominant emotion of frustration around the topic of "authentic voice" rings mechanically clear.

AI generation of language develops on the screen by selecting and arranging predictable words located mathematically in vectors probabilistically affiliated with a topic term, in this case the term “voice.” As a result, AI sets the frame by announcing the setting and the topic in the first sentence. AI usually analyzes the user prompt and chooses words from the prompt to open its response, in this case “teacher’s lounge” and “voice.” I prompted AI to write a monologue set in a teacher’s lounge about the frustration among teachers who have been told to help their writers solve the problem of “finding an authentic voice.”

The chronology is expressed in typical a AI mechanical framework from "twenty years" to "yesterday's PD" to "next week's PD” which doesn’t feel completely lifelike, though it isn’t as egregious an AI tell as the opening language. Even the emotional markers—the cold coffee symbolizing disillusionment, the empty cup representing teaching's emptiness—read like computationally safe choices. It’s important to avoid the tendency to read everything in a student’s text as AI output once the idea arises in an evaluator’s mind. Confirmation bias can skew the inference.

To see clearly how AI works, try an experiment. Prompt AI to write a dialogue among adults situated around an American kitchen table, at a luncheon of retired teachers, or in a formal faculty meeting. Coffee cups show up routinely in informal gatherings in settings like a teacher’s lounge—coffee is more probable than Diet Coke—because the statistical probability of coffee consumption is high. See what happens when you prompt AI to situate the dialogue in a teacher’s lounge in England. My bet is you won’t be surprised:

*Takes a weary sip of tea* Right then, here we are in the staff room, nattering on about helping our dear pupils "find their voice" in their writing. *Gestures with half-eaten digestive* Rather precious, isn't it? Been at this teaching lark for twenty years, getting the little ones to "discover their authentic voices" through composition, and now - well, now my Year 7s are using ChatGPT to churn out five different versions of their book reports, each with its own distinct "voice."

Quite rich, actually. Just yesterday in our professional development session - bloody waste of an afternoon if you ask me - we spent two hours discussing how to nurture student voice, while half my lot were likely at home asking the computer to "write this like a proper keen Year 7." *Dunks digestive rather aggressively in tepid tea*

Remember when we actually believed we were helping them discover something genuine? Now I'm watching these AI programs generate more "authentic voices" in five minutes than I've supposedly coaxed out of pupils in my entire career at St. Margaret's. Makes you wonder what we've been playing at all these years, doesn't it? Perhaps next week's training can focus on helping students find their authentic AI prompting voice. *Stares at soggy remains of biscuit*

Still, mustn't grumble. We can always put "Pupil Voice Development Specialist" on our CVs, can't we? That ought to count for something in this brave new world of education we're meant to be creating. Fancy another cuppa?

With experience, one learns that AI uses repetition mechanically to sustain its focus on the semantic nucleus of the prompt. We have a monologue, a teacher’s lounge, a heavy repetition of the word “voice,” and the underlying sense of a pendulum swinging. I didn’t know the term “digestive”; so, I asked Claude:

"A digestive biscuit, usually just called a "digestive," is a staple British snack - a round, semi-sweet whole wheat cookie with a distinctive grainy texture. The name comes from the Victorian-era belief that its main ingredient, sodium bicarbonate, aided digestion. Think of it as Britain's everyday cookie, similar to how Americans might reach for an Oreo or graham cracker.”

As I suspected, AI selected this term because it isn’t used except in British English and therefore cleanly identifies the setting using strong statistical modeling. It changed middle schoolers to grade 7, it changed coffee to tea, it changed cold coffee to tepid tea, and it changed teaching resumes to CVs. Without belaboring the point, it rendered in British English the sentiment it rendered in American English. The center of the user prompt is “voice,” and the repetition of this center remains the same whether in England or the United States.

The neat thematic closure with its "Voice Development Specialist" credential and "brave new world" reference ties everything up perfectly and is featured in both versions. A human teacher's frustrated outburst would include unfinished thoughts, meandering observations, specific references to actual students, and less obviously symbolic gestures. This artificial perfection in structure and symbolism betrays the text's non-human origins.

Perhaps AI's disruption offers an opportunity to move beyond metaphor toward more precise and practical writing instruction. The exercise I’ve demonstrated—prompting AI to write a monologue on the same topic in two different cultures—provides a rich opportunity for teaching young writers what voice actually is when we disconnect it from “authenticity” and reconnect it to “mask” or “persona.” Students can work in pairs to prompt AI and change the setting for the second prompt.

Then they can print out the monologues, exchange them, and collaborate to try to figure out what their peers used to prompt the AI, what evidence they find in the AI output that convinces them they figured out the prompt, and how AI responded effectively and ineffectively to create a mask. Recent studies in using AI to teach writing in an English-as-a-Foreign Language context has already begun to explore such uses. AI can generate alternative expressions of similar themes and ideas in various contexts to showcase changes in vocabulary, syntax, grammar, and pragmatics in new ways.⁶

The metaphors that shape writing instruction reflect deeper assumptions about how meaning works in both writing and reading. 'Close reading,' a dominant skill taught in high school, exemplifies how metaphorical thinking influences our entire approach to literacy instruction. Just as we imagine voice as something to be 'found,' we imagine meaning as something that exists 'close to' the text's surface, waiting to be uncovered through careful observation.

Lately, I’ve read rumblings of discontent chipping at the edges of close reading, but the pedagogical problem of fuzzy outcomes remains. Close reading is at best a vague outcome with a checkered past, not the means to achieving an outcome. I am not alone in arguing that close reading's focus on the "four corners of the text" can limit students' ability to connect readings to broader contexts in the world outside the text or to personal experiences that could be relevant to a reader’s response to a text. Snow and Matthews (2016)⁷ highlighted that this narrow approach may reduce engagement and fail to integrate background knowledge, which is crucial for comprehension:

“Close reading is an excellent technique for probing sentence structure, nuances of word meaning, subtleties of text organization, and the structure of textual arguments. But it is not a technique for building background knowledge, or integrating background knowledge with the language of the text, which are major bottlenecks for many struggling students” (Snow, 2016).

One wonders what the spillover effects are for students’ understandings of the writing process if their understanding of reading means staying inside the boundaries of a text. When we design learning activities around metaphorical objectives like "authentic voice" or “close reading,” the common ground which literacy instruction needs to span a twelve-year curriculum depends on practitioner perspectives that have no literal anchor.

Even the detailed specifications in the Common Core standards and the ubiquitous tips and tricks offered on line don’t help specify a theory of close reading or a theory of authentic voice. Close reading means confining readers to the interior territory of a text. We use to call this approach “reading for literal comprehension” or “reading the lines.” Although researchers in literacy have theories of the crossover between these psycholinguistic processes unfolding in sociocultural circumstances, these metaphors do not inform one another much if at all.

Of course, ignoring what the text says isn’t wise, but forcing readers to stick to the lines of the text under the micromanagement of the teacher hardly qualifies as a comprehensive approach to secondary reading instruction—or any level of instruction for that matter. Answering “right there” and “think and search” questions are half the way there. Dealing with “on your own” questions is the desired outcome, the space between comprehension and composition, which gets short shrift when opportunities to work in that space are reserved for an as argumentative writing assignment, which is hardly a writing assignment but proof of success on a close reading assignment. Argumentative essays grounded in textual evidence are but a small planet in the universe of written discourse

Reimagining Voice in the Digital Age

Enter AI to thoroughly scramble established thinking about "authentic voice" in writing instruction. When students can generate multiple voices at will through AI prompting, what becomes of our romantic notion of discovering one's true voice? What is a “true voice"? AI forces us to confront an uncomfortable reality: Perhaps voice was always more about strategic choice than innate authenticity.

Consider a student using Claude to write multiple versions of the same argument, one for a college application essay, another for a public policy proposal, a third for social media activism. Each version adopts a distinct voice suited to its context and audience. Which of these represents the student's "authentic" voice? Or does authenticity now lie in the strategic ability to deploy different voices effectively? Or is “authenticity” an aesthetic notion of valuing rather than constructing?

This technological disruption may actually align writing instruction more closely with Bakhtin's concept of heteroglossia, in itself a metaphor, the idea that we all speak in multiple voices shaped by social context. Rather than pursuing the myth of a singular authentic voice, perhaps we should teach students to develop a repertoire of voices, the judgment to choose among them, and the skill to craft a text that represents them.

AI becomes not a threat to authentic expression, but a tool for expanding students' vocal range while maintaining their agency as writers. Teachers must teach students how to distinguish cut-and-paste AI output from human-composed text to protect them from false confidence, cognitive offloading, and pure plagiarism. The next level is teaching students to use AI output productively to serve their authentic communicative intentions—we do have authentic intentions when we’re not under the yoke of compliance, right?—while maintaining control of their semantic vision (Moss, 1992).

1. Lakoff, G., & Johnson, M. (1980). Metaphors we live by. University of Chicago Press.

https://web.iitd.ac.in/~sumeet/metaphor.pdf

2. Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (pp. 610-623). https://www.science.org/doi/10.1126/science.adt6140

3. Elbow, P. (1998). Writing without teachers (2nd ed.). Oxford University Press.

4. Young, S. (2016, January 31). The brain is like a bunch of muscles [Blog post]. Northwest Tutoring. Retrieved from https://www.nwtutoring.com/2016/01/31/the-brain-is-like-a-bunch-of-muscles/

5. Hillocks, G. (1986). Research on written composition: New directions for teaching. National Conference on Research in English.

6. Zhang, X., & Wang, C. (2023). Artificial intelligence in language learning: Current applications and future directions. Journal of Educational Technology & Society, 26(1), 57-74

7. Snow, C. E., & Matthews, T. J. (2016). Reading and Language in the Early Grades. Future Of Children, 26(1), 57-74.