AI-Assisted Academic Writing

A Practical Approach

When discussing AI’s role in scientific writing with colleagues, I often hear variations of “I’ve tried AI for writing, but it’s not good enough for proper scientific text.” Although this view holds some truth, it often arises from misunderstandings of AI’s role in academic work and its proper application. Many researchers approach AI as a complete writing solution rather than recognizing it as a sophisticated tool that requires proper integration into the writing process.

Setting the Foundation

In this post, I will share my approach to AI-assisted academic writing, focusing on the fundamental principles. I plan to follow up with more detailed insights in future blog posts. My writing process results from many months of experimenting with different tools and workflows, refined through practical application across various academic writing projects.

Before diving in, two critical points need emphasis:

First, my focus here is primarily on academic rather than creative writing. The goal is straightforward: to transform ideas from my mind into clear, readable text. This is utilitarian writing aimed at effective communication, not literary achievement. Academic writing communicates complex ideas, methodologies, and findings within scientific discourse. While creative writing has its place in academia, scientific writing prioritizes clarity and precision over style. And this makes it well-suited for AI assistance.

Second, while AI provides valuable writing tools, its output often contains recognizable “AI-isms” - patterns and phrases that hint at its algorithmic origins. I work to minimize these AI-isms through careful editing, but I never attempt to mask AI’s involvement in my academic writing, as doing so would be unethical. Instead, in my academic papers, I always disclose which AI tools I have used and how. And for blog posts like this one (yes, AI assisted in its creation), AI serves simply as a tool for enhancing readability.

Understanding AI’s Role in Academic Writing

Before delving (this is an AI-ism, by the way) into the practical steps, I have found it important to understand what AI can and cannot do in academic writing. In my experience, AI excels at structuring information, suggesting alternative phrasings, and helping maintain consistency across long documents. However, it cannot generate original research insights, verify factual accuracy, or replace human expertise in the field.

For successful integration of AI into the academic writing process, the key therefore is leveraging AI’s strengths while actively managing its limitations.

A Process-Based Approach

My AI-assisted writing is a multi-step process involving my human intervention and usually at least two different Large Language Models (LLMs) that can cross-check each other. This approach, recently also termed “EchoWriting” for its ability to echo an author’s voice, goes beyond simple prompt engineering to encompass a comprehensive writing process where human expertise and AI capabilities work together to enhance both efficiency and quality.

Step 1: Creating a Style Guide

I always begin by generating a personalized style guide using an LLM to analyze existing writing samples. I typically use ChatGPT for this analysis, as it seems to excel at content evaluation. While some writers explicitly ask the LLM to include specific instructions about tone and structure, I have found that better LLMs inherently understand style guide requirements. I tend to get the best results using my own previous writing or field-typical texts as examples for generating the style guide.

In my process, the main purpose of using a style guide is to reflect my academic voice while maintaining field- or purpose-specific conventions. I also often edit the generated style guide as needed to incorporate examples of preferred terminology, sentence structures, and paragraph organization. This guide becomes my reference point throughout the writing process, ensuring consistency across different sections and papers.

Step 2: Drafting

I begin by getting my ideas down in rough form. At this stage, I do not worry about style or grammar; I focus purely on capturing my thoughts in writing. While I prefer traditional typing, voice recording with AI transcription can be an excellent alternative for writers who think better while speaking. This stage works best when it is completely uninhibited, allowing me to explore ideas freely without concern for polish or presentation.

Throughout this initial phase, I concentrate on articulating my main arguments and supporting evidence, along with major methodological details. I include critical insights and observations while noting connections between different concepts. The key in this phase is to capture ideas while they are fresh, without letting concerns about structure or completeness slow down the process.

Step 3: Iterative Refinement

After drafting, I begin multiple rounds of AI-assisted refinement. I present my draft and style guide to an LLM (I prefer Claude 3.5 Sonnet for its natural writing style), then manually review and correct its output. I repeat this process several times, typically needing about five iterations to reach a final version.

These iterations can be structured to improve different aspects of the text systematically. One effective approach is to progress from overall structure and flow to clarity and precision of language, followed by technical accuracy and terminology. Later iterations might focus on style guide compliance, with a final pass for polish and refinement. While I take a more organic approach to these revisions, following such a structured progression can help ensure comprehensive coverage of all important aspects of academic writing.

For these revisions, I sometimes work directly within LLM interfaces, though several specialized tools exist. Novelcrafter and Raptor Write are helpful tools for creative writing but also work for non-fiction texts. For academic work, I can recommend SciSpace, though there are alternatives. These tools can often also provide specialized features for academic writing, such as citation management and field-specific terminology suggestions.

Step 4: Final Editing

I usually conclude my process with professional editing tools. While not specifically designed for academic writing, tools like AutoCrit and ProWritingAid can adapt very well to non-fiction or academic styles. I use both, but when recommending just one, I suggest ProWritingAid for academic work. These tools help me identify various issues, from consistency and readability problems to grammar and punctuation errors. They also suggest improvements in sentence structure variation and word choice.

Managing AI Integration

While AI tools can significantly improve the writing process, their effective use requires careful attention and oversight. I maintain quality control by regularly verifying that AI suggestions align with my intended meaning and standards. I never assume AI-generated content is automatically correct or appropriate. And perhaps most importantly, I always verify technical terms, citations, and field-specific content, as AI can make confident-sounding but incorrect statements about specialized topics.

A Personal Process

It is important to note that my approach reflects a personal reality: I often struggle with the initial blank page but thrive in the editing process. This is a pattern that has shaped how I integrate AI into my writing workflow. Yet writing is deeply personal, and what works for one academic might be counterproductive for another. Some of my colleagues excel at rapid drafting but find editing tedious, while others prefer methodical writing with minimal revision needed.

When developing this workflow, I did not just focus on the technical aspects of AI integration. Instead, I considered the entirety of my writing process - my tendencies, strengths, and challenges. I thought about how I work best, what frustrates me most, and where AI tools could genuinely help rather than add complexity. By aligning the process with these personal patterns, I have created an approach that feels natural rather than forced.

In the end, AI remains a sophisticated assistant in my writing process, not its driving force. The ideas, insights, and scholarly contributions still emerge from years of research and academic experience. AI helps me articulate these thoughts more clearly and efficiently, acting as an intelligent editorial companion rather than a replacement for academic thinking and expertise.