Using Generative AI for functional rather than expressive writing

I advocate distinguishing between what I call functional and expressive documents in academic writing. Functional documents, such as reports, summaries, or abstracts, tend to involve academics reporting on what they have done or explaining what they will do. These involve writing, but it rarely tends to be a joyous writing of the kind that truly engages us. It is rather an accounting for our activity that can often be connected to more expressive forms of writing, but which serves a particular purpose.

Until we find ourselves in a position where the automation of writing is possible, there’s little reason to make these distinctions. We might experience them as different, not least of all because of this different enjoyment of them. But the practicalities are fairly similar in the sense that we find ourselves writing. I suggest that the role of creativity in each of them is revealed when we consider how machine writing could improve or degrade these activities.

Using generative AI to produce functional documents not only saves time, it can improve the quality of those documents. While I might be unusual in my struggle with abstracts, I’ve long realized how difficult I find it to produce a compelling overview of a text I have written. In contrast, this is a task which GenAI tools excel at, particularly if you provide an exemplar to guide the process. The creativity resides in how we provide a compelling and effective summary which is effective for indexing the text and providing a preview for potential readers. It is not a mechanical task devoid of creativity but rather a tightly defined one in which the creative contribution serves a narrow instrumental purpose. I tend to find writing reliably energizing but these sorts of tasks can often be weirdly draining given the limited length of the text which is being produced. It tends to constitute an item on a to-do list, distinct from the writing process, with the inevitable drag on time and energy that entails.

Using GenAI for functional writing tasks like this can have a considerable effect on the feel of the working day, making it possible to move quickly through activities which otherwise would have been cumulatively draining. This doesn’t mean that we should entirely outsource the production of something like abstracts to tools like ChatGPT and Claude, but it does mean that we might comfortably rely on them in most, if not all, circumstances to produce our first drafts. My experience is that the initial summarization needs a little tweaking to be adequate for your purposes.

As with any prompting exercise, being clear about what you can expect and providing examples can simplify the process considerably. For example, I might say to Claude that “I am an academic writer who has written a journal article on subject X for journal Y who now needs to write a 150-word abstract based on the preprint text I have provided” before explaining that “I need this abstract to be compelling and succinct, matching the language of my preprint where possible, with the intention that a reader is provided with a clear sense of the content of the article, the contribution to knowledge and why it would be valuable for them to read.” Exactly what you want this to look like will vary between people and fields, so it’s essential to provide examples of what you’re looking for. If you have a perfect example of what you’re looking for in an abstract then provide this alongside the prompt.

This is another example of how prompting can be a spur to reflexivity. In order to explain what the output should look like, you need to be able to state explicitly what you think a good abstract should look like. Until I began to use GenAI to support functional writing, it had never occurred to me to explicitly state this. I would simply dive into the writing process with a vague sense of what I was trying to do, usually at the end of a writing process with an impending deadline and a corresponding tendency to settle for an imperfect abstract that could easily have been improved.

It’s now easier for me to write an abstract with the qualities I’m looking for because I’ve taken the time to state what those qualities are. If you’re not sure what exactly you are looking for, you can use GenAI to help you analyze examples to articulate your nascent intuitions. Look through recent papers in the journals in which you aim to publish, taking a minute or two to linger on the abstract for each paper that has recently been published. What do you like about it? What do you dislike about it? Share the abstract with Claude or ChatGPT and ask for an analysis of its structure, strengths, and weaknesses.

By engaging with AI in this way, we can develop a clearer understanding of our own writing preferences and standards. This reflexive process helps us become more conscious writers, even as we leverage AI to handle the more functional aspects of our academic writing. The time and mental energy saved can then be redirected toward the expressive writing that brings us more joy and intellectual satisfaction.