Documentation

Basic search — plain keywords or a short phrase across six academic sources at once. Best for a quick first pass.

AI-assisted search — describe what you need in a full sentence, in any language; SciScrape AI builds the structured search plan for you and shows it alongside the results.

Structured search — full manual control over required terms, bonus terms, hard exclusions and seed DOIs. Best for a systematic review you need to document.

Saved theme — re-run a protocol you already built, without rebuilding it.

Narrow a search to journal articles, reviews, meta-analyses or preprints. The filter is applied before the result limit, so you get the number of results you asked for from the types you chose.

Types come from what the sources report and are normalized without AI. A paper whose sources report no recognised type is shown as Unknown and is not returned when a type filter is on; each paper's detail shows which source reported which type.

Re-ranks results by relevance: AI reads the title and abstract of your top results and scores them for relevance to your actual intent, blended 30% into the existing keyword/citation ranking.

Worth it for a nuanced or ambiguous query. Skip it for a broad, well-defined topic where the base ranking is already good — it costs extra credits either way.

Expand outward from papers you already trust: what they cite (backward) and what cites them (forward), plus a third 'similar' column of Semantic Scholar recommendations that fit neither direction.

A result reached through more than one of your seed papers, or from both directions, is the strongest kind of lead — it means multiple parts of the literature independently point at it.

Classifies each forward citation (a paper that cites your seed) as supporting, contrasting, or merely mentioning it, based on the actual sentences around the citation — not just the fact that a citation exists.

Useful for two things specifically: finding papers that dispute a result you're relying on, and filtering out the ~80% of citations that are just background mentions.

A project is your own literature library — upload PDFs, import results from a search run, or type in unpublished data (a measurement, a protocol note) — and ask questions across all of it with citations back to the source.

An imported paper starts as metadata only (title/abstract); fetch its full open-access text, or upload a PDF yourself, before asking questions that need more than the abstract.

Upload a manuscript into a project, then invite reviewers by email — they don't need an existing account to accept. Single- and double-blind modes control who sees whose identity.

A reviewer leaves a structured review (summary, major/minor issues, a recommendation, a confidence level); the author can respond point by point, then publish a revised version.

The same Review module works for a student/supervisor relationship: upload a draft, invite your supervisor the same way you'd invite a peer reviewer.

A dedicated thesis mode (supervisor-specific vocabulary, no blinding, review attachments) is planned — until then, set blind mode to 'none' when inviting a supervisor.