Writing as Learning in the Age of AI: Emig’s Insight

Writing as learning means that putting an idea into words can change the idea itself. Janet Emig made that case in her 1977 essay Writing as a Mode of Learning. Her question feels newly urgent when a generative AI tool can produce a coherent paragraph before a student has worked through what they want to say. If the paragraph arrives first, which opportunities to learn through writing might disappear?

Emig did not write about AI. Her essay offers a way to examine what happens during composition, and it can help us make better decisions about where AI assistance belongs. It cannot establish that AI writing tools harm learning; that question needs evidence from contemporary classrooms.

Why the process matters

We often judge writing by the finished product: Is the argument clear? Is the evidence well organized? Does the prose flow? Those questions matter, but they can hide the work a writer did to reach that point. A student may discover a gap in their reasoning while trying to explain a source, or recognize during revision that two claims do not fit together. A polished final paragraph tells us little about whether that thinking happened.

That is why AI use in a writing assignment cannot be understood only by asking whether the final text is acceptable. In my discussion of teaching writing in the AI era, I have returned to the value of process. Emig helps explain why the process deserves our attention in the first place.

Writing as learning: Emig’s argument

In College Composition and Communication, Emig (1977) develops a theoretical argument about writing as a distinctive way to learn. She compares writing with reading, listening, and speaking, especially speaking.

She recognizes that classroom talk can support learning and help someone prepare to write. Her point is that writing places different demands on the learner: it is usually slower, must supply more of its own context, and leaves a visible record that the writer can revisit.

For Emig, that record provides immediate feedback. A writer can reread a sentence, notice that a connection is missing, and change it. Notes and drafts also preserve the development of an idea over time. Writing asks the learner to select, connect, analyze, and synthesize while moving at a pace they can adjust. Emig draws on learning theorists to argue that these features make writing especially valuable as a mode of learning.

This is an essay of conceptual reasoning, supported by examples and earlier scholarship. It is not an experiment comparing students who write with students who do not. Some of its discussion of brain hemispheres also reflects the neuroscience available in the 1970s and should not be carried into current teaching advice as an established explanation.

The more durable contribution is Emig’s account of how a writer can inspect and reshape thought through the developing text. She ends by calling for research to describe and substantiate what learners lose when they have fewer opportunities to write.

What this could mean for AI and writing instruction

Emig’s argument directs us to a question that a rule about permitted AI use may miss: Which part of the writing task is meant to help the student learn? If students are learning to form an interpretation of a text, writing an initial explanation themselves may expose uncertainties they need to resolve.

Having AI generate that explanation at the outset could remove the moment when those uncertainties become visible. That is an inference from Emig’s argument, not a finding her essay tested.

One possible sequence is to let students make brief notes, write a provisional claim, and explain how their evidence supports it before consulting AI. They could then ask the tool to raise a counterargument or identify a place where a reader may need more explanation.

The student would decide whether the response reveals a real weakness and revise accordingly. This keeps the first attempt at meaning-making with the learner while allowing AI to enter a later feedback cycle.

I would also be careful about making unaided drafting a universal rule. Language editing, translation, dictation, and other supports can help students express ideas they have developed.

The pedagogical question is more precise: does a particular form of assistance help the learner do the thinking at issue, or does it complete that thinking for them? This matters especially when we infer competence from the finished text. A fluent paragraph may conceal either substantial student reasoning or very little of it.

Students can make some of that process visible without surrendering every draft or prompt. Ask them to keep an early claim and a later version, then explain one change in their reasoning.

Or ask them to identify an AI suggestion they rejected and why. These are suggested teaching adaptations; Emig did not study them. They would also need to be manageable within a course and accessible to students with different ways of composing.

The essay leaves me with a useful distinction. Writing can communicate what a student already knows, and it can help the student come to know it. When AI enters a writing task, we need to know which of those purposes the task serves.

For related work on how students evaluate assistance during composition, see my review of feedback literacy and AI-enhanced writing. This question also belongs in the wider Teaching & Pedagogy conversation.