No, ChatGPT alone doesn’t write good (scientific) articles yet. It’s far from sufficient to simply enter a prompt quickly and have the AI produce the desired output.
However, AI brings some interesting innovations to the writing process. I’d like to introduce you to my current workflow and toolkit.
Before I go into details, an important note: Legally speaking, such an AI workflow might be delicate. However, all steps only use published data, so ultimately the machines do nothing different than the human brain – just faster, more structured, and without experiencing fatigue.
Step 1: Create an Initial Literature Selection
My writing starts at elicit.com. I have an idea of what I want to write and first get an overview of current and relevant literature. I start a search query and scan the results using the AI-generated summaries. With the filter function, I narrow down the results to 4-5 relevant papers or books that I use for my next step. I export the result list as a .bib file.
Step 2: Check Literature for Relevance
Now I switch to researchrabbitapp.com. This second AI tool helps me verify my Elicit list. I can see relatively easily and quickly whether the found articles are relevant to my topic. I import the .bib file and have it show me how my selected titles fit into the literature landscape. How often were they cited? Are they current? Are the authors well-known and embedded in a network?
These are important criteria by which I can review my literature selection. Research Rabbit also suggests similar titles to me, and I can gain further inspiration from the author network.
In this second step, my literature list grows and I know which titles I should focus on. Since I can connect Research Rabbit directly to my own literature management in Zotero, articles and books that I have already read and processed can also be considered.
Step 3: Read and Annotate Literature
Now that I have found the necessary literature, it’s time to read. I need to refresh my expertise on the topic, because in the end, I want to be able to review the texts generated by the AI.
So I obtain the PDFs for all articles and read them, while also making notes and marking important passages. I do all of this directly in Zotero, my literature management tool of choice. Nice side effect: In Zotero, I can also document the bibliographic data for all articles right away. For later.
Zotero offers some initial AI functions via plug-ins, such as summarizing texts using ChatGPT.
Step 4: Evaluate Literature with AI
What Zotero only manages in rudimentary form, another AI tool does much more comprehensively and easily: ChatPDF. Here I can upload PDFs from my Zotero collection and query them. This makes it easier for me to grasp the structure of a paper, extract important methods, and summarize statements.
This is where my first thoughts and formulations that I will use later when writing are now created. These text blocks remain saved and I can use them directly in the next step.
Step 5: Now It’s Time to Write!
And with that, we’ve reached the actual writing. I also do this in a special AI tool. Jenni.ai offers me a simple editor with ChatGPT already built in. And the best part: Jenni.ai works hand in hand with ChatPDF. So I can use the results from the last step directly while writing.
With generative AI, I can of course also design structures for my text, make my texts longer or shorter on demand, have them rephrased, and even have entire passages suggested.
Jenni.ai also takes care of the correct citations. I import the current literature list from Zotero and can then insert the desired citations directly in the editor. Browser plug-ins like Deepl, Grammarly, or LanguageTool help me translate or correct texts directly in the online editor.
I can export the finished text in various formats and, if desired, further edit it in Word.
PS: A brief note on costs: all mentioned tools are available in a free version. For the workflow to function in all its details, the Pro versions are ideally needed.
This article first appeared on LinkedIn.
