KONSİLThe reflex after having an idea is usually the same: type the sentence of the idea into a search box. There are two possible outcomes. If something similar appears, the idea is shelved that evening. If nothing appears, the idea is declared novel, and months later, with data collection finished, someone finds a paper that asked the same question three years ago in another country. Both outcomes come from the same mistake: assuming novelty is a question you can answer with yes or no.
The better question: which part is already done?
Research ideas are not consumed whole; they are consumed component by component. Most ideas have four: in which population, what is applied, compared to what, and which outcome is measured. The literature has probably exhausted some of those components already. Novelty tends to live in the remaining combination, and very often in the outcome.
A concrete case. In the idea "the effect of AI-assisted feedback on learning", there are hundreds of papers on AI-assisted feedback and thousands on feedback in education. But if the outcome measured is reasoning skill rather than satisfaction, and the population is trainees rather than specialists, that intersection may well be empty. The idea is not unstudied; it is partially studied, and the part you defend is the remainder.
Searches run without making this distinction fail in both directions: too broad and every idea looks done, too narrow and every idea looks new.
How to search
- Split the idea into components and write synonyms for each. For "trainee": resident, junior doctor, doctoral researcher. For "feedback": formative assessment, corrective feedback. If your working language is not English, find the established English term for the concept before searching; searching a literal translation returns the wrong picture.
- Start with reviews. If a systematic review exists, it has drawn the map of the field for you, and its "future research" section usually names the empty regions. Check PROSPERO for reviews that are registered but not yet published[3]. Keeping a record of your own search steps makes writing it up much easier later; PRISMA-S offers a ready framework[1].
- Find the five closest papers and run a forward citation search. Papers citing them show where the topic stands now. Backward reference chasing gives you the past, forward citation gives you the present; you need both.
- Look for unpublished work. Publication is a late record of finished work. Ongoing studies appear in trial registries[2], preprint servers and institutional or national thesis repositories[5]. More than one registered study on your question raises the risk that the gap closes before you finish; compare their designs and outcome measures one by one to see whether they truly ask the same thing.
- Use more than one database. A single source always understates:
| Source | Good for |
|---|---|
| PubMed | Backbone of biomedical literature; precise searching with MeSH terms[4] |
| OpenAlex | All disciplines, open and free; strong topic and institution breakdowns[6] |
| Google Scholar | Wide net; also catches theses, proceedings and grey literature |
| Trial registries | Ongoing studies that are not yet published[2] |
| Repositories | Dissertations and reports that never became journal articles[5] |
Four common mistakes
- Treating absence as evidence. The most likely explanation for "nothing came up" is a poor query, not an empty field. When a search returns nothing, change the terms first and celebrate second.
- Searching the sentence of the idea. Search engines match concepts, not sentences. Search the intersection of two or three concepts instead of a long phrase.
- Searching in one language only. Work published in a national language may be invisible in English-language indexes, and the reverse is also true. A study done elsewhere does not make yours worthless, but you should design knowing it exists.
- Abandoning the idea on finding something similar. A similar study is often good news: it proves the question matters and that journals publish it. What matters is the size of the difference, not its existence.
A checklist for your own idea
Before you can say "I have checked this properly", you should have answers to six questions:
- Have I written out the four components separately, with a synonym list for each?
- Is there a systematic review from the last five years, and what gap does it propose?
- Have I found the five closest papers, and which outcome did they measure?
- Have I looked at what cites those five papers?
- Are there registered studies underway on the same question?
- Can I state my difference in one sentence? If not, the difference is not yet clear.
The last item matters most. A novelty claim has to fit into a single sentence a reviewer reads in the first paragraph: this is known, this is not, and this study adds that.
Sources and standards
- PRISMA-S — extension for reporting literature searches in systematic reviews. equator-network.org
- WHO ICTRP — shared search across clinical trial registries worldwide. who.int · also ClinicalTrials.gov
- PROSPERO — register of systematic review protocols. crd.york.ac.uk/prospero
- MeSH (Medical Subject Headings), NLM — controlled vocabulary behind PubMed. nlm.nih.gov/mesh
- OpenDOAR — directory of open access repositories, for theses and grey literature. sherpa.ac.uk/opendoar · in Europe, DART-Europe
- OpenAlex — open catalogue of scholarly works. openalex.org
Done properly, this search takes more than half a day
KONSİL was built to run exactly this method: it splits the idea into components, generates multilingual queries, searches the open literature and returns a verdict with its reasons. It does not hide the part that is already done; it writes the defensible gap as a separate statement. The sample report shows a real run across 99 papers.
Apply for the closed beta