Teaching Reading and Writing with Generative AI: Reflections, Opportunities, and Challenges

When ChatGPT appeared in November 2022, it captured global attention: it could explain linguistic concepts, converse coherently in several languages, and compose a sonnet on request. Within a year it reached 100 million users. Yet generative AI (GAI) did not begin in 2022. Writers and readers have long relied on word processors, electronic dictionaries, spell-checkers, machine translators, and digital reference tools. What has changed is the depth of assistance: GAI now participates in the very act of reading and writing, reshaping the literacy practices of entire communities and posing new challenges for education at secondary and higher levels.

A useful way to understand GAI is the metaphor of a scale model: a detailed replica of everyday verbal communication that allows us to refer to almost anything in our reality. Technically, GAI rests on large language models (LLMs), mathematical constructs containing the probabilities that words relate to one another, learned from millions of millions of authentic texts through deep learning. These models can be visualized as vector spaces in which related terms, such as king and queen, lie close together. It is essential to distinguish general artificial intelligence, which automates tasks once reserved for humans, from generative AI, the subfield that produces content such as text, images, video, and music. Key terms include the algorithm, the set of instructions a computer follows; the prompt, the input given to an AI in each query; and hallucinations, the errors or invented information AI systems can produce.

The race to digitize and automate writing began decades ago. Word processors displaced electronic typewriters in the 1980s, personal computers and early spell-checkers arrived in the 1990s, and online multilingual resources such as WordReference became popular at the turn of the century. The Spanish Royal Academy dictionary went digital in 2001, and Wikipedia and terminology databases supported technical writing. Machine translators and style correctors steadily improved, followed by assisted writing tools and programs such as LanguageTool. ChatGPT is therefore one more step in a long trajectory rather than a sudden break. Its significance lies in its conversational, general-purpose capacity to produce fluent text on demand.

Despite its impressive performance, GAI raises serious concerns. First, hallucinations undermine reliability: one study estimated ChatGPT's average accuracy across ten types of specific reasoning at about 63 percent. Second, models reproduce the biases present in their training corpora, which can yield sexist, racist, homophobic, or Anglocentric outputs. Third, users often enter personal data into prompts, raising privacy concerns. Fourth, students may invest minimal effort, which can be counterproductive for learning. Fifth, GAI can be used to produce disinformation. Beyond these risks, critical reading demands asking who is behind the technology. Many algorithms are designed primarily to generate profit, whatever the promotional rhetoric about changing the world or democratizing progress. GAI also creates a misleading impression of rational thought, although it does not truly understand or possess a reliable model of truth.

The landscape of GAI tools evolves rapidly, and many familiar applications silently incorporate AI into their interfaces. Tools were once classified by interaction mode: chatbots (ChatGPT, Gemini, Copilot), text generators (Quillbot, Jasper, DeepL), voice tools, multimedia generators (DALL-E, Midjourney), and AI-powered search engines (Perplexity, Consensus). However, multimodal convergence has made such classifications obsolete, since leading tools now handle voice and images as well. Behind these products are companies such as OpenAI, whose technology powers ChatGPT, DALL-E, Copilot, and Perplexity, and Google, whose alternative is Gemini. Smaller developers include DeepL, Midjourney, Humata, ChatPDF, Consensus, and Quillbot. Most originate in the United States, with some from Germany, and all operate globally on the web.

An ethnographic study of Spanish-speaking professionals who work with writing reveals sophisticated uses of GAI. An Argentine journalist uses Copilot to generate headlines and introductions optimized for search engines, transcribes audio with specialized tools, formats articles in her newsroom's system, and later analyzes their impact and reuse potential with analytics software. A Spanish science journalist transcribes and translates interviews with AI, asks ChatGPT to summarize key ideas and draft social media posts, and always verifies outputs against external sources. A Chilean lawyer has trained ChatGPT on national legal conventions to generate templates, organizes client accounts chronologically, and tests the tool on analyzing court rulings for contradictions or procedural errors, accepting that results contain more hallucinations in complex tasks but still offer relevant ideas. A Spanish advertising professional used image generators with carefully engineered prompts to create promotional visuals for a reading comprehension product. These cases show that experts succeed by combining AI output with critical judgment, iteration, and verification.

For teachers, GAI can help design materials, adapt texts to different proficiency levels, generate examples and feedback, and reduce time spent on routine tasks, freeing attention for personalized guidance. For learners, it can act as a tireless tutor that explains concepts, offers alternative formulations, provides feedback on drafts, and answers questions at any hour. Used well, it supports the development of metacognitive awareness, since students must evaluate, compare, and justify the outputs they accept. The educational value depends on how the tool is integrated, not on the tool alone: the central goal should be to strengthen student agency rather than substitute for it.

Legitimizing GAI in the classroom requires explicit rules and a pedagogical sequence. A workable protocol begins by teaching students the basic concepts of GAI and its limitations, so that they understand hallucinations and biases. Next, students practice prompt design, learning to specify purpose, audience, genre, and constraints, and to iterate when results are unsatisfactory. Students then compare AI-generated and human-written texts for the same task, identifying differences in accuracy, style, and originality. Verification is a core stage: learners check claims against reliable sources. Finally, reflection and documentation are required, with students explaining how they used GAI, what they accepted or rejected, and why. This process emphasizes self-regulated or adaptive learning, in which learners monitor their own progress, decide when to rely on the tool, and take responsibility for the final product. Transparent disclosure of AI use replaces prohibition as the main norm.

The widespread availability of GAI calls into question traditional notions of plagiarism and authorship. When a text is co-produced by a human and a model, the distinction between original and borrowed work becomes blurred. Some scholars speak of a post-plagiarism era, in which the key issue is not whether assistance was used but how it was used, whether the writer understood, evaluated, and took responsibility for the content. Hybrid human-AI writing may become the norm in professional and academic settings. Education must therefore move from policing textual origin toward assessing the writer's process, judgment, and ethical engagement. Teaching students to write with AI, rather than only against it, prepares them for the literacy practices of the near future.

Generative AI is transforming how people read and write, offering powerful assistance alongside serious risks of error, bias, privacy loss, and superficial learning. Demystifying the technology, understanding its history and commercial context, observing expert practices, and adopting transparent classroom protocols are essential steps. The challenge for education is not to choose between rejecting or embracing these tools, but to cultivate critical, adaptive, and responsible literacy that keeps human judgment at the center of writing and reading.