Dr Guy W. Bate
Across education, AI is often introduced through what it can do. It can summarise readings, generate quiz questions, draft feedback, simulate conversations, and provide forms of support that appear scalable. These are useful capabilities, particularly in institutions already stretched by workload, student numbers, and expectations of responsiveness. Yet this way of framing AI also narrows the issue. It makes AI appear primarily as a productivity tool, when its more significant effect may be on the language through which education is organised and experienced.
Education as communicative work
Education depends on language in very practical ways. A feedback comment tells a student what has been noticed in their work and what still needs development. An assessment rubric gives form to standards that might otherwise remain implicit. A policy statement signals what an institution values and what it is prepared to enforce. A student support email can reassure, redirect, or close down a concern depending on how it is written. These are not secondary acts of communication around education. They are part of how teaching, assessment, care, and authority are enacted.
Generative AI is now entering precisely these moments. It is used to draft feedback, rewrite assessment instructions, generate exemplars, produce student-facing explanations, and standardise institutional communication. Its role is therefore not limited to producing content more quickly. It is beginning to participate in the ordinary language through which educational judgement is expressed.
The idea of communicative AI is useful here because it shifts attention away from generative systems understood merely as tools for human deployment (Coeckelbergh and Gunkel, 2025). Rather, their significance lies in how they unsettle familiar boundaries between language, communication, authorship, and meaning. AI does not simply transmit a human intention. It generates language that others interpret, trust, revise, and act upon. For education, this provides a more precise starting point: AI matters not only because it produces text, but because that text can enter the communicative conditions through which educational meaning is made (Bate, 2026).
Feedback, rubrics, and support
A lecturer might ask an AI system to draft comments on a student essay. The output may be encouraging, coherent, and aligned with the language of the marking criteria. It may also be too general, or it may soften a judgement that should be made more directly. Once edited and returned to the student, the comment becomes part of the assessment relationship. The student does not receive “AI text” in the abstract. They receive feedback that appears to carry the judgement of the educator and the authority of the institution.
The same issue appears in assessment design. AI can help produce rubrics that sound clear and professional. It can quickly generate descriptors for levels of performance, align them with learning outcomes, and smooth inconsistencies in wording. But rubrics do more than describe performance. They shape what students come to understand as valuable work. If AI-generated rubric language defaults to familiar educational phrases such as “critical engagement”, “clear structure”, or “effective analysis”, without sufficient attention to the specific task, the result may look rigorous while remaining thin. The language performs assessment quality, but does not necessarily deepen it.
Student support offers another example. AI-assisted replies can help institutions respond quickly to common queries about deadlines, extensions, enrolment, or academic processes. In many cases this will be helpful. But supportive communication often depends on recognising when a routine query is not routine. A student asking about an extension may be signalling distress, confusion, caring responsibilities, or disengagement. A well-written AI-assisted response may give correct procedural information while missing the moment where human judgement and care are needed.
The authority of fluent text
These examples show why fluency is not a small matter. One of the strengths of generative AI is that it produces text that sounds appropriate to its setting. Feedback sounds like feedback. Rubrics sound like rubrics. Policy language sounds suitably institutional. This fluency lowers barriers to communication, but it also creates a risk. Text that sounds considered may not always be grounded in considered judgement.
The risk is not only hallucination or factual error. Even accurate AI-assisted text can change expectations about educational communication. If polished feedback is easy to produce, then polish may begin to stand in for attention. If assessment language can be generated quickly, then familiar phrasing may be mistaken for clarity. If support replies can be standardised, then responsiveness may be measured by speed and tone rather than by whether the student’s situation has been properly understood.
Authorship and responsibility
This also complicates authorship. Education relies on being able to locate who is speaking and on what basis. Students need to know that feedback represents a judgement someone can explain. Staff need to know whether they are communicating their own view, an institutional position, or a platform-generated formulation. Institutions need to know when their “voice” is being shaped by templates, prompts, vendors, and automated drafting systems.
Responsibility does not disappear because AI has assisted with the wording. If anything, it becomes more important to design practices that keep responsibility visible. Educators should be able to identify which parts of feedback express their own evaluative judgement. Students should be told when AI is being used in ways that materially shape communication. Systems should make it easy to review, revise, and reject generated text, rather than simply encouraging users to accept a fluent draft.
Designing for judgement
For EdTech providers, this means designing beyond output quality. A good tool should not only generate plausible educational language. It should support the human work of judgement. For example, a feedback tool might separate descriptive comments from evaluative claims, prompting the educator to confirm the latter. A rubric generator might require task-specific criteria before producing performance descriptors. A student support chatbot might identify when a query should be escalated to a human rather than continuing to produce procedurally correct replies.
For institutions, the governance question is broader than data privacy or model accuracy. Those concerns remain important, but they do not exhaust the issue. Institutions also need to ask where AI is shaping student-facing language, which communicative acts require explicit human ownership, and how students can challenge or clarify decisions expressed through AI-assisted text.
Communicative literacy
This calls for a more concrete form of AI literacy in education. Staff and students need to understand prompts and outputs, but they also need to recognise how AI changes the texture of educational communication. They need to notice when feedback is fluent but unspecific, when assessment wording is polished but generic, when support language is warm but procedurally evasive, and when institutional communication becomes consistent at the cost of accountability.
AI will—and, to my, mind must—continue to be used in education because it offers real benefits. It can help educators manage demanding workloads, assist students in developing ideas, and make some forms of communication more accessible. The challenge is to use these capabilities without allowing generated language to weaken the connection between words and judgement.
The future of AI in education will not be determined only by whether its outputs become more accurate or more sophisticated. It will also depend on whether educators and institutions remain attentive to what happens when feedback, guidance, policy, and support are increasingly co-produced with systems that are fluent but do not themselves bear educational responsibility.
References
Bate, G. W. (2026). Review of Mark Coeckelbergh and David J. Gunkel (2025). Communicative AI: A critical introduction to large language models. Postdigital Science and Education. https://doi.org/10.1007/s42438-026-00640-w
Coeckelbergh, M., & Gunkel, D. J. (2025). Communicative AI: A critical introduction to large language models. Cambridge: Polity Press.