How PaperCurrent chooses papers
Each paper is checked against your research profile. Here's what goes into the score, how we check the evidence, and where the limits are.
Five things we look at
PaperCurrent uses AI to assess how a paper relates to your topics, methods, and hypotheses, how it might affect your projects, and what you could do with it. Topic, method, and hypothesis scores use the same criteria for everyone. Project impact and next steps depend on what you've written in your profile.
| What we assess | The question behind it |
|---|---|
| Topic | Does it address your research area? |
| Methods | Does it use study designs, measurements, or analyses relevant to your work? |
| Hypotheses | Does it bear on a claim you're testing? |
| Project impact | Could it change how you approach a current project? |
| Next steps | Is there something useful you could do with it? |
Challenges are checked twice
A “possible challenge” means a paper may question one of your hypotheses. A second AI check looks for findings or direct arguments that support the flag. If the evidence is too weak, the paper moves to Other matches with an explanation.
The flag is a reason to take a closer look. You decide whether the finding changes your view.
Checking the quotes
Each assessment includes a quote from the title or abstract. An automated check confirms that the quoted wording appears in the source; assessments with unsupported quotes are rejected. When an abstract is missing, the result says so and confidence is limited.
Keeping a record
We keep the details behind each assessment so it can be traced back to the information used:
Research profile, scoring criteria, and AI instructions Model, paper details, and date Original assessment and your feedback
Saving your profile creates a new version for future checks. Earlier assessments keep the profile version used at the time.
How we plan to measure quality
Before the open beta, we plan to measure how many suggested papers are useful and how many relevant papers we miss. Researchers will judge papers against their own profiles without seeing PaperCurrent's scores. A second reviewer will check whether the quoted evidence supports each explanation. We'll publish the results, our targets, and what we'll change if we fall short.
How your feedback is used
When you mark a paper relevant or not relevant, we update the share of results you've found useful. Up to ten rated papers can also serve as examples in future assessments for that monitor. Your feedback does not retrain the model or edit your profile. If you see a profile suggestion, you can accept or dismiss it and make any changes yourself.
Sources and limitations
| Source | Role |
|---|---|
| OpenAlex | Finding papers and citation counts |
| Crossref | Checking DOIs and publication details |
| Europe PMC | Finding biomedical and psychology papers and abstracts |
| Unpaywall | Open-access links |
Your profile is private to your account. We send the parts needed for assessment to our AI provider, Anthropic in the United States, along with paper details and examples from your feedback. These are excluded from model training. Our app and database run in Germany. See the privacy page for the full details.
Limitations: Assessments use titles and abstracts only. PaperCurrent can miss relevant papers or misjudge their importance, so it cannot replace a systematic literature search. Use the scores to help choose what to read, then check the original papers.