Since starting my PhD in veterinary pathobiology in 2014, I have looked for ways to streamline my literature-review workflow. Every day, I search for papers, import them into my citation manager, read them and take notes. I can then incorporate those references and insights into manuscripts. But the conventional approach of searching for and downloading PDF files is tedious and inefficient, involving multiple mouse clicks, scattered files and a large disconnect between my notes and the source. Ten years later, with the development of digital and artificial intelligence (AI) tools, I have finally landed on a process that can streamline my academic writing. I call it ACCU -- the acquisition, collection, crystallization and utilization workflow.
For quick keyword searches, I use Google Scholar. In settings, under 'search results', I set the export format to RefMan for compatibility with my citation manager and add my university library under 'library links' so that I can retrieve the full text when it is available.
For systematic searches, I use the PubMed database's advanced search function, coupled with the EasyPubMedicine Chrome browser extension to display the journal ranking, impact factor and citation count under each hit. On the PubMed search-results page, I can click 'create RSS' to turn the results into a web feed in an RSS reader, such as Feedly, which will then alert me to any new papers that fit the search criteria. This allows me to access all newly published papers in my field -- as well as multiple journals' most recent publications -- anytime, anywhere, even on my phone.
To flesh out my research, I use the AI-powered academic search engines Perplexity and Consensus. These can provide answers to well-defined, natural-language questions, such as, 'Is vitamin D beneficial for osteoporosis?' They have a better user interface than ChatGPT, with hyperlinks to the original sources so that I can verify the results.
Alternatively, I can use ResearchRabbit, an AI-based tool that allows me to use papers in my literature collection as seeds to find related publications. I can add the newly identified papers to my reference manager with one click.
I use the free and open-source reference manager Zotero because it is easy to use and compatible with many of my other digital tools.
Zotero covers all the basic functions of many available commercial tools and, in my experience, provides better solutions. The Zotero Connectors Chrome browser extension allows me to import papers from Google Scholar searches in batches. A right-click can save academic articles (or any web page) to my Zotero library in a ready-to-cite format. My university library provides the OpenURL link resolver, which I use in Zotero's settings to automatically download full-text articles. I can double-click on Zotero entries to highlight and annotate the PDF.
Advanced users can expand Zotero with plugins. For example, I use Notero to import all the 'What Is Your Diagnosis' articles from the journal Veterinary Clinical Pathology into Notion, a productivity and note-taking app, to create a searchable database for veterinary clinical pathologists and trainees.
If I need to generate a quick bibliography -- for instance, for a slide presentation -- I use ZoteroBib. This is a free online tool for creating bibliographies without an account or software. I simply input identifying information for my articles, such as web links, DOIs or ISBN numbers, and ZoteroBib formats them into whatever citation style I need.
I order my thoughts using a virtual card-based tool called Heptabase. Each 'card' is a piece of information that can be displayed on multiple whiteboards, each associated with its own topic. By creating cards and connecting them to one another, I can visualize their relationships and synthesize complex ideas.
To make a new card, I drag my highlights from Zotero to the Heptabase whiteboard. The new card automatically has my Zotero annotations in quotation marks, plus the citation hyperlinked to Zotero and another link to the article PDF. In this way, I keep all my annotations in one place instead of scattered across different files. Furthermore, I can connect cards with lines to create associations, or create sections to group multiple cards.
Another useful tool is NotebookLM. This is a free service from Google that organizes user-uploaded content in virtual notebooks. Users can upload as many as 50 sources per notebook and discuss the uploaded materials with the chatbot as if with a tutor. For instance, I can upload veterinary-cytology notes and ask, 'What are the morphologic features of round-cell tumours?' The AI will generate answers with links to the source material for verification. A feature called Audio Overview can generate a podcast-like dialogue between two AI 'hosts' using the uploaded content. Users can even prompt the hosts to focus on a specific source or topic, or to target a particular audience. This feature allows me to listen to educational content as a podcast while working out, making it a convenient way to stay productive on the go.
I use other AI tools to compile information about the literature. SciSpace and Elicit, for instance, allow users to import Zotero collections and generate customized summary tables. Say I have a bunch of papers and want to know which ones use sequencing technology. With either tool, I can create a column called 'sequencing' with a customized prompt: "List 'Yes' if sequencing technology is used in the paper, otherwise, list 'No'". The AI will automatically 'read' through the papers in my collection and populate the column accordingly. But as always with AI, be sure to double-check its responses, because these tools can and do make mistakes.
When it's time to turn knowledge into a manuscript, Zotero has plugins that integrate with Microsoft Word, LibreOffice and Google Docs, similar to the 'cite while you write' feature in the commercial EndNote reference manager, made by Clarivate. Once the Zotero plugin is installed, clicking on 'add/edit citations' in the word processor opens a search bar so you can find and insert references from your Zotero library. I prefer working at the keyboard to using the mouse, so I created a keyboard shortcut to do this automatically in Word.
As a non-native English speaker, I use the free online tool Grammarly to fix misspellings and grammatical errors, but tools based on large language models -- such as ChatGPT, Claude and Gemini -- can also provide editorial help if they are given the proper prompt. For example, I can ask, "Please act as a scientific journal editor to refine my article. Make the text sound like it was written by a native English speaker, while keeping the editing minimal and preserving the original meaning." Whatever you do, ethics demand that you disclose your use of AI, even if your publisher doesn't specifically require it.
For researchers and graduate students, AI literacy is now an essential skill, just as Google search was in the 2000s. AI will not replace people anytime soon, but people who use AI might replace those who don't. My ACCU workflow is a good way to start embracing digital and AI tools in your processes and has greatly improved my efficiency. I hope it helps with yours, too.
But remember, sometimes the best tool is the one you are most comfortable using, or the one you're already using. If you find yourself spending more energy optimizing your workflow than actually working, you might be wasting your valuable time.
