A literature review or results paper usually starts with finding and reading relevant papers first, before writing up your manuscript. However, the newest wave of AI tools, like Sourcely, allows you to write up the results first and have an AI identify references relevant to the draft to flesh out the discussion and introduction around the findings. This has many advantages,can greatly speed up your literature review, allowing you to focus primarily on your research and results, and allows for devision of labour when working on papers in a team. In other words, AI can help you find better questions to the answers that you have found, and more impactful questions can result in better venues for your publication. Let’s look at this seemingly counterintuitive approach in detail with Sourcely.
Learn which tools can speed up your research: This free 21-day email series introduces you to a range of academic tools and how they work together to speed up the time needed for a literature review. Most of these tools have been around for less than a few years, some use AI. Knowing them gives you an almost unfair productivity advantage for your research.
This post builds on the deep-dive tutorial on Sourcely, showcasing newer features like deep search and focusing on the reverse literature review process. Here is a brief overview of the involved steps.
- Start with a piece of writing, which can be a fairly rough first draft. Even ideas or questions are sufficient.
- Upload this to Sourcely and let it search relevant references based on the text.
- Working through the references Sourcely has found, you can craft more publishable paragraphs.
- Next, upload the finished writing to Sourcely to find more nuanced and relevant references, further improving the manuscript or providing new questions and ideas for your research.
This forms a loop that can be executed multiple times, allowing for an iterative improvement process of your manuscript. Here is a schematic:
Let’s look at exactly how Sourcely works and what you can expect in this reverse literature review process.
What does Sourcely do?
Sourcely takes an academic text of any length and provides references (i.e., academic papers or books) that match this text and support or contrast single statements in it. It is similar to a semantic search engine like Consensus or SciSpace, but instead of asking a question, you provide text that requires detailed references to support and contrast the whole text or parts of it. One of the most useful applications is when you write up the results of your paper and need discussion material (e.g. studies that have found similar or related things). Sourcely’s deepsearch, in particular, prioritises full-text PDFs and can show you individual paragraphs in these PDFs that support or contradict claims made in your query text.
Importantly, Sourcely provides a quick overview of the importance of different sources through badges and a relevant score.
What exactly is a reverse literature review?
A reverse literature review refers to the process of finding sources for a text that you have generated. This is the opposite of first seeing the sources to create the text. That would be a classical literature review. Importantly, the text that you start your reverse literature review with does not have to be very polished or thought out, as AI will still find relevant sources. That is what makes this approach so powerful: you start from a much more hypothetical stage and build your understanding from there, rather than engaging in a complex synthesis of many papers.
Use case #1: Which results are impactful?
As a computational ecologist, I often evaluate data objectively, looking for patterns and interesting findings before asking a specific question. This leads to me having far more results than I can publish, often raising the question of which results are really relevant. This is where Sourcely can help. By uploading your results to Sourcely, you can find papers that have identified similar, related, or contrasting results.
This example on the left is my very first draft that I copied to Sourcely to find ~20 papers with related results.
This method allowed me to find a paper that followed a very similar approach to what I did, which I thought was unique, as I had never seen it before in the broader literature. Of course, reading this paper makes it much easier to write my discussion and introduction, as there is now a template I can use to structure them. Additionally, using a subset of these twenty papers with a reference search tool like Litmaps allows me to dig deeper and find even more highly related sources.
Searching for individual paragraphs with Sourcely’s deep search
If an academic paper is open access, Sourcely will preferentially search its full text and highlight single sentences inside this full text relevant to your query, allowing you to verify the relevance of this paper. Here is an example:
The example above allowed me to identify a theoretical methods paper. suggesting a framework around my method, which I thought was unique. This is invaluable for the discussion of my paper.
Use case #2: Identifying Brilliant Research Ideas
As part of a creative research journey, it is advisable to write down speculations, thoughts, and ideas, regardless of whether they seem feasible at the moment. I have a lengthy collection of these so-called thought notes that help me generate new research ideas and be more creative. Sourcely’s reverse literature review approach allows you to check which parts of these ideas or hypotheses have already been worked on. Reading these papers, you can gauge if they are feasible or relevant ideas. Here is how it looks:
After uploading a lengthy text to Sourcely, the system will highlight single sentences that lead to interesting results in the literature, called “citation-worthy text highlights”. With one click, you can identify relevant papers to each of these highlights. Reading and analysing these papers allows you to refine your ideas or hypotheses, which you can re-upload to Sourcely for a second round of testing.
This refinement loop is an excellent example of how AI can speed up scientific discovery and enhance creativity by surfacing relevant known facts (i.e. papers) without compromising your critical thinking abilities.
Use case #3: Finding New Developments in the Field
When searching for references in Source.ly, you can use filters to specify, for instance, a particular date or venue for your results. These filters give rise to a very interesting workflow. If you copy and paste an older review, for instance, Smith 2020, but set the filter to only give you results after 2020, the publication date of the review, you will get the developments of the field matching the text of the review paper, which can allow you to identify research gaps and developments that happened since the publication date.
Here too, you can make use of the citation-worthy text highlights, which sort your papers by a topic sentence and provide up to 10 results (20 if running a deep search) for each topic sentence. The resulting papers are guaranteed to be very recent. If the input paper was highly relevant to your own literature review, this workflow will help you identify the newest developments in the field and greatly improve the quality of your lit review. For best results, upload very popular and impactful review papers.
Use case #4: Improving your Manuscript
Almost every peer reviewer will suggest that you add additional citations or ask you if you are also aware of a specific paper that came out on your topic. But before you submit your paper to peer review, Sourcely can do that for you, thus greatly improving your chances of acceptance. For this use case, I uploaded my unpublished manuscript to Sourcely. You can then highlight single sentences that you might be unsure about and ask Sourcely to provide specific guidance for them. I made a video on this workflow:
This workflow is particularly useful for conclusion paragraphs, where you often don’t have references. But adding them might lend more weight to your arguments. It can also help to find references that are newer than the references that you used for specific sentences in your manuscript, thus improving credibility.
Where does AI get its information?
Abstracts and titles of every paper are available through global databases like PubMed, Semantics Scholar, or OpenALEX. Tools like Sourcely pull this information to provide you with an AI interface to search them. This makes hallucinations impossible. So-called open-access papers allow access to the full text of the PDF, and almost every AI tool will prioritise full-text access. However, the majority of scientific publications are not open access, which limits tools like Sourcely to only using their abstracts.
What is the difference between a quick search and a deep search?
The main difference is the number of papers that the AI will scan, constituting the depth of the result. Think of a deep search as running numerous quick searches and adding them together into a more meaningful, deeper result. Deep searches usually take 5-15 minutes to complete, while quick searches can run in just a few seconds. Due to the higher amount of computation required to run a deep search, these are usually limited to a certain number per month, even if you are a premium subscriber.
The best approach is to start with a quick search. If AI finds relevant results that advance your research, switch to a deep search to gather additional details.
ChatGPT vs Sourcely for Literature Review
ChatGPT is a general search engine that can answer broad questions and give you ideas. Sourcely is a semantic search engine for academic papers. Because ChatGPT’s answers are rarely backed by scientific papers (a phenomenon referred to as hallucinations), its answers can often be simply wrong, while Sourcely bases its answers on scientific literature. This effectively eliminates hallucinations in Sourcely. However, it can still make mistakes, mainly by providing papers that are tagged as relevant, but might not be relevant to your query. Use your best judgment and adjust your query to get optimal results.
Sourcely Pricing Options
Sourcely’s pricing is primarily dependent on how many deep searches you need per month. There are three plans: Pro($14 / mo), Ultra($31 / mo), Max ($64 / mo), giving you 10, 30, or 1000 monthly deep searches, respectively:
Use the discount code EFFORTLESS40 and EFFORTLESS20 to get a 40% annual or 20% monthly discount.
Summary: Is Sourcely worth it?
Sourcely is a unique tool with a wide range of applications for your literature review. For early career researchers and students, Sourcely can provide citations and lend credibility to their essays and assignments. Experienced researchers can use Sourcely to polish their manuscripts by finding more relevant or more recent references, thus improving their manuscript quality. Sourcely can also function as a regular semantic search engine, assisting throughout the entire literature review process.
This combination of features makes it a great choice for any academic who is tasked with a literature review. If you want to learn how Sourcely fits into the entire literature review process, check out the Effortless Literature Review course.
