Personalisation Is Not Optimisation (III)
There is a scene that repeats itself in writing courses that incorporate generative AI, and it has become invisible through sheer repetition: the student opens a conversational interface, types a prompt, receives an output, submits it. The sequence is so seamless that it appears natural. The problem is that, within that seamlessness, precisely what a writing course should make visible disappears.
What disappears? The process.
Not the technical process — that runs without friction — but the cognitive one. The process that Paula Carlino has described over two decades as the epistemic function of writing: writing not to communicate what one already knows, but to discover what one did not yet know one was thinking. What happens when someone attempts to put a vague idea into words and, by externalising the thought, transforms it. That transformation is not produced by the output of a prompt. It is produced by the effort of having written one.
In her keynote address at the V International Congress on Linguistic Research (2025), Carlino frames this with an uncomfortable clarity: learning a complex discursive genre — a thesis proposal, a case analysis, an argumentative essay — is not a matter of learning formatting rules. It is a matter of adopting the intellectual and cognitive practices of a specific disciplinary community. And if we delegate that learning to the machine, we save ourselves the trouble but forfeit the transformation.
I have written elsewhere in this space that a well-designed prompt can function as a writing protocol: it compels the writer to specify role, audience, task, output structure, and quality criteria. I stand by that claim. But I now want to push it a step further, into territory that strikes me as both urgent and underexplored: the prompt as an object of knowledge in its own right. Not as technique. As a discursive form that deserves to be studied, taught, and discussed with the same rigour we bring to teaching students how to write a summary, a report, or a critical review.
The prompt has kinship
Those of us who teach writing know that discursive genres do not emerge from nothing. They share lineage, they mutate, they migrate from one sphere of activity to another. Email inherited features from the business letter. The tweet compressed the aphorism and the maxim. The scientific abstract stabilised in two hundred and fifty words a set of rhetorical operations that once took pages to resolve.
Where does the prompt come from?
Viewed through the lens of linguistics rather than engineering, the prompt shares traits with at least four textual families: the didactic assignment (which orients an intellectual action), the writing protocol — such as those proposed by Fernando Vásquez Rodríguez — (which sequences operations of planning, drafting, and revision), the creative brief (which delimits an expected output under constraints), and, if one will permit an analogy that works better than it might seem, the recipe (which describes conditions for a process to transform raw materials into a product, knowing full well that the outcome depends on whoever executes it).
This idea does not originate here. In Prompts in Education: The Power of the Question in the Age of Generative Artificial Intelligence I argued that the construction of an utterance and the design of a question are not rhetorical artifices but epistemic acts capable of shaping how we think. Shifting from “What do you know?” to “How will you use that?” is not a cosmetic variation: it opens an entirely different pedagogical ecosystem. And in If You Don’t Like Writing, It’s Because… I argued that writing — and questioning — is a mode of thinking, not of compliance. What I add now is that the prompt inherits both functions: it is question and writing at once. A question written so that a non-human agent may execute it. And it is precisely in that dual condition that its interest as an object of study resides.
In every genre I have mentioned, we are dealing with texts whose function is not to be the result but to organise the conditions for a result to occur. The recipe is not the dish. The score is not the music. The prompt is not the text.
But there is a fundamental difference: in the recipe and the score, the executor is a human being who interprets, adjusts, improvises, and makes productive errors. In the prompt, the executor is a statistical language model that predicts probable token sequences. This radically changes the nature of the writing that precedes the text: it demands an explicitness that, paradoxically, resembles disciplined thought more closely than it does everyday speech.
And here something emerges that deserves attention: to write a good prompt, one must know, with considerable precision, what one wants to think. Which is already, in itself, an act of thought.
Claiming that the prompt is an emerging discursive genre is easy. Demonstrating it requires a different kind of work: testing that intuition with the tools that text linguistics and genre theory provide. That is what I will attempt in the next instalment of this series — The Prompt as an Emerging Discursive Genre — where I will examine whether the prompt meets the criteria that Bakhtin, Swales, and Miller attribute to genres, and what consequences the answer holds for those of us who teach writing. For now, let it suffice to pose the question and its most immediate pedagogical implications.
What Carlino compels us not to forget
Carlino’s argument in the keynote that motivates these lines has a structure worth unpacking, because it is subtler than it first appears.
She does not say: “AI is bad for writing.” She says something more precise and harder to assimilate: the risk lies not in the tool but in the uncritical delegation of the formative process. If a master’s student uses AI to draft a thesis proposal without having traversed the cognitive difficulties that genre imposes — moving from a practical problem to a theoretical one, learning to read as a researcher, formulating questions that a discipline recognises as legitimate — what is lost is not a text. What is lost is the learning that only occurs when the text resists, when the sentence will not come, when the structure of the argument collapses and must be rebuilt.
That is exactly what I wrote to my students in the reading and writing course with generative AI: if the proper place for the prompt is perhaps the point of arrival after a deep academic reflection, does it make sense to position it always at the beginning, before we have inhabited the text?
The question is not rhetorical. It has concrete pedagogical consequences.
If the prompt always comes first, it functions as closure: it resolves before the problem has been properly formulated. If it comes after — after reading, after attempting, after partially failing, after rethinking — it functions as an amplifier: it strengthens what the student has already begun to build. Carlino proposes that AI should serve as a tool of control after the student has developed their own thinking. I add: the formative prompt is not the one written instead of writing, but the one written after having attempted to write.
There is also a dimension that cannot be treated as an appendix: the ethical one. In the chapter we published with Rubén Yáñez Reyna and Paola Mercado Lozano in Ethical and Legal Dilemmas of Artificial Intelligence in Latin America (Palgrave Macmillan, 2025), we addressed the ethical and educational dilemmas that AI poses to higher education institutions in Latin America. What we documented there has a direct bearing on the present discussion: Latin American universities are incorporating AI tools without clear ethical frameworks or pedagogical protocols that distinguish between productive use and uncritical delegation. The problem is not only cognitive — what is lost when the machine writes for us — but institutional: what kind of education are we validating when the process remains hidden behind an interface? Normalising technology and the habits it imposes on us is, as I wrote in Razón Pública, the finest sales strategy available. Let it not become our pedagogical strategy as well.
The prompt as second-order writing
Here I wish to advance a proposition that I am developing for the course and that I share openly for discussion: the prompt is a form of second-order writing. Writing whose object is not a text but the conditions of possibility of a text.
That distinction matters. When an environmental engineering student writes a prompt to have AI help draft the discussion of results in their undergraduate thesis, they are doing — if they do it well — something cognitively demanding: they must make explicit the theoretical framework from which they interpret the data, the argumentative structure they expect, the type of evidence that counts as valid in their field, the scholarly voices with which they wish to engage. All of that must be in the prompt, because the machine does not infer it. And to place it in the prompt, they must have thought it through first.
In AI Literacy: Key Steps I argued that interacting with language models is, above all, an act of writing and reasoning, and that the first step is developing awareness and command of writing itself. What the work with prompts in the reading and writing course with generative AI confirms is that this was not a statement of principle: it is an empirical finding. Students who arrive at the prompt without having engaged with disciplinary reading, without having confronted the difficulty of formulating a problem in their own words, produce generic instructions that yield generic texts. Those who arrive after having inhabited the text — after annotating, underlining, attempting a first draft that made them uncomfortable — produce prompts that function as genuine tools of revision and amplification. The difference does not lie in prompt technique. It lies in what the student brings with them when they sit down to write it.
That is to say: a well-written prompt is, against all appearances, an exercise in metacognition. It compels the writer to make visible what normally remains implicit. It compels answers to questions that direct writing allows one to evade: For whom am I writing? From what theoretical position? What kind of text am I requesting, and why? What are the criteria by which I will judge whether the result is adequate?
These are precisely the questions that a good writing course teaches students to ask themselves. The prompt does not replace them. It makes them inescapable.
But — and here I return to Carlino — only if the student already has something with which to answer them. A prompt empty of disciplinary content produces text empty of disciplinary content. The machine amplifies what we give it: if we give it vagueness, it returns vagueness with good syntax, which is the most dangerous form of vagueness because it looks like rigour.
What this means for a writing course with generative AI
What I have been proposing across the series Personalisation Is Not Optimisation converges here with what Carlino has documented over years of research and with what we are experiencing in the classroom every week: it is not enough to incorporate AI into the teaching of writing. We must rethink what we teach when we teach students to write prompts.
We do not teach “prompt engineering” — that is a marketing label, not a pedagogical category. We teach students to write an emerging discursive genre that possesses rhetorical structure, a specific addressee, conventions still in formation, and whose quality depends entirely on the depth of thought that precedes it.
And we teach something further, something that may be the most valuable lesson of all: we teach that there is a form of writing that precedes writing. That before the text lies the decision about what text deserves to exist, for what purpose, for whom, under what conditions. That this decision is an intellectual act no machine can make for us, and that learning to make it well is learning to think with rigour.
Carlino would say — and she would be right — that this learning only occurs when it is anchored in content that matters, in a discipline that challenges, in a problem that the student recognises as their own. Not in decontextualised exercises of “improve this prompt.” Outside the discipline, the prompt is verbal gymnastics without a body to inhabit it.
Within the discipline, the prompt can be what Vásquez’s writing protocols were for a previous generation of students: a device that makes the craft of writing visible. Only now the device is dynamic, interactive, and mercilessly honest: if you do not know what you want to think, the machine will show you with a clarity that unsettles.
Let us make that unsettlement productive.
References
Bakhtin, M. M. (1986). The problem of speech genres. In C. Emerson & M. Holquist (Eds.), Speech Genres and Other Late Essays (pp. 60–102). University of Texas Press.
Carlino, P. (2025). Escribir, leer y aprender en la universidad (2nd rev. ed.). Fondo de Cultura Económica.
Carlino, P. (2025). Keynote address. V International Congress on Linguistic Research. Video
Fawns, T. (2022). An entangled pedagogy: Looking beyond the pedagogy–technology dichotomy. Postdigital Science and Education, 4(3), 711–728.
Galindo-Cuesta, J. A. (2023). Actions and decisions: Ethics in education with artificial intelligence. Razón Pública. https://razonpublica.com/acciones-decisiones-etica-la-educacion-inteligencia-artificial/
Galindo-Cuesta, J. A. (2025). Prompts in education: The power of the question in the age of generative artificial intelligence. Escritura Digital. https://www.escrituradigital.net/prompts-en-educacion-el-poder-de-la-pregunta-en-la-era-de-la-inteligencia-artificial-generativa/
Galindo-Cuesta, J. A. (2026). Personalisation is not optimisation (II): From writing protocols to prompts as pedagogical design. Escritura Digital. https://www.escrituradigital.net/personalizar-no-es-optimizar-ii
Galindo-Cuesta, J. A., Yáñez Reyna, R., & Mercado Lozano, P. (2025). Ethical and educational dilemmas of AI in Latin American higher education institutions: Persistent challenges and inquiries. In D. Ramírez Plascencia & R. M. Alonzo González (Eds.), Ethical and Legal Dilemmas of Artificial Intelligence in Latin America (pp. 143–164). Palgrave Macmillan. https://doi.org/10.1007/978-3-031-86540-4_8
Vásquez Rodríguez, F. (2024). El protocolo: una ayuda didáctica para ordenar las acciones al redactar diversas formas discursivas.
