Why prompting should not be our focus when teaching students about AI

A new academic perspective argues that teaching students to prompt AI chatbots is less important than fostering their agency and reflective dialogue with the models, citing research by Essien et al. that warns efficient prompting does not guarantee learning. The author suggests shifting focus from 'prompt engineering' to broader literacy and dialogical skills to prevent cognitive atrophy and passive AI use.

This is a sketch of a much longer piece of work I need to start: - The role of prompting is diminishing as chatbots develop because they increasingly infer and pre-empt what the user shares. It’s no longer a crucial skill for using them effectively and we need to avoid presenting it as if it is. - As Essien et al https://www.tandfonline.com/doi/full/10.1080/03075079.2026.2686986 put it, “agency in prompting does not automatically produce learning; without reflective regulation, it may become only efficient prompting”. The student might still treat uncritically accept outputs, outsource to a self-defeating degree or engage in malpractice while nonetheless prompting effectively. - It’s hard to talk about prompting without veering into the terrain of ‘prompt engineering’. This is a contingent feature of the discourse but there’s a whole theory of what models are implicit in the notion of ‘prompt engineering’ which is really unhelpful for learning: imagining prompting as a form of natural language coding through which you unlock the capabilities of the machine. - The language of prompting also lends itself to thinking about single-shot interaction which is not how you use a chatbot in reflective and thoughtful ways. Again there’s nothing necessary about this but the discourse constrains our capacity to emphasise the dialogical element as essential while still talking about prompting. - If we see the use of chatbots for learning in terms of dialogue, then we need a dialogical vocabulary for thinking about it. What makes for a better or worse conversation? What is the student bringing to the model? How are they greeting the responses of the chatbot? How does the interaction unfold through these dynamics over time? This is the terrain we need to work on and again I think the language gets in the way. - The capabilities involved in this are much more like literacy more broadly than the language of prompting can really account for. To write fluently, confidently and thoughtfully equips you to prompt effectively. To draw upon a range of references in expansive and relevant ways unlocks the capability of the model to respond to complexity. If we talk about this as ‘prompting’ we redescribe what are relatively familiar aspirations of humanistic education in an unhelpful way. This also obscures inequalities of cultural capital in who can do this and who cannot do it. I realise in writing this my problem is less with prompting and more on prioritising it and the conceptual baggage which comes with it. Instead I suggest the focus needs to be on student agency when using chatbots. Essien et al https://www.tandfonline.com/doi/full/10.1080/03075079.2026.2686986 again: When students maintain control over goals, prompts, evaluation, and final judgement, the feared risk of cognitive atrophy may be reduced Yang et al. 2024 , and AI may function less as a surrogate and more as a dialogic partner Krakowski 2025 . This aligns with warnings that passive AI use may weaken learning Linde- baum et al. 2025 .