How to check whether your research has already been done

The most expensive mistake in research is committing a year to an idea and discovering, at review, that it was already done, sometimes decades ago, under a name you never searched. Here is how to check properly, before you commit.

Why a normal literature search isn't enough

Around 2.5 million new studies were indexed in 2024 alone. No one can track a field by hand anymore. Worse, keyword search only finds work phrased the way you phrased it, and scientific terminology drifts. Three researchers can each run a thorough search, all miss the same paper, and only find it after publication because the earlier authors called it something else. That is terminology drift, and it is the single biggest blind spot in a literature review.

A method that actually catches prior work

Work through these five steps for any idea you are serious about:

1. State the idea as a variable × method. “Multi-generational adaptation (variable) of corals to chronic acidification (method/context).” Precision here is what makes the rest work.
2. Expand every term. List synonyms, abbreviations, and older names for each concept. The same snow fieldwork appears as “snow survey” in older papers and “snow sampling” in recent ones. This is the step most people skip.
3. Search across all of them, not just your phrasing, and follow citations backward from the closest hits.
4. Judge actual overlap, not similarity. A paper can look close and do something different, or look distant and do exactly your idea. Read the nearest few and ask: does this do what I propose?
5. Check the shape of the field. Who publishes here, and is the topic emerging, saturated, or did a preprint just pre-empt you? Timing changes the decision as much as overlap does.

Doing it in seconds instead of days

This is exactly what unkwn42 automates for climate research. You enter the idea; it expands the terminology through a concept layer built from 400,000+ papers, retrieves the nearest prior work (flagging the ones that only match after a renamed term), and compares your specific claim against each candidate: whether a paper already asserts it (Claimed), how the nearest work differs from yours (Adjacent, stated as a delta), or that the pieces exist separately and no paper joins them (Open), plus the key players and the publication trend. Every claim cites a real retrieved paper.

What “not found” honestly means

No tool can prove a negative across all of science. When nothing matches, it means not found in the indexed corpus: a strong signal, not a guarantee. A good check tells you where it looked and shows its receipts, so the judgement stays yours. That honesty is the point: the tool surfaces the evidence; you make the call. And a match is not a rejection, it is a starting point, what to match if you are replicating or what to differentiate against if you are extending.