Scite (Smart Citations)

Scite is a commercial citation-analysis tool built around a feature called Smart Citations: rather than reporting only how many times a paper has been cited, it uses a trained deep-learning model to read each citing sentence and classify the citing paper’s stance toward the specific claim it references — supporting, contrasting, or mentioning — and shows that excerpted sentence alongside the classification. The methodology was published in a peer-reviewed paper, Nicholson et al., “scite: A smart citation index that displays the context of citations and classifies their intent using deep learning,” in Quantitative Science Studies (MIT Press, vol. 2, issue 3, 2021).

What Smart Citations classify, and why that distinction matters

A raw citation count, or a journal/author-level metric derived from one (see CASRAI’s Citation Impact entry), treats every citation as equivalent: a paper cited 200 times looks stronger than one cited 20 times, regardless of why those citing papers referenced it. That conflation is a known weakness of count-based metrics — a claim can accumulate citations from papers that later contradicted it, papers that merely name-checked it in passing, and papers that genuinely built on and confirmed it, all counted identically.

Scite’s three-way classification is aimed directly at that gap. For a specific cited claim, a researcher can see, at a glance, whether the body of citing literature has tended to support it, push back against it, or simply mention it without taking a position — and can read the actual sentence making that determination rather than trusting a single aggregate number. This is most useful for exactly the kind of due-diligence question citation counts alone cannot answer: has this finding held up under later scrutiny, or has it since been contradicted or qualified by follow-up research?

How this differs from citation-count metrics

Traditional citation-impact metrics — raw citation counts, h-index, journal-level indicators indexed in Web of Science or Scopus — are volume measures: they answer “how much was this cited?” Smart Citations answers a different question: “what did the citing literature actually say about this specific claim?” The two are complementary rather than substitutes — a high citation count with a large share of contrasting citations is a materially different signal than the same count with mostly supporting citations, and that difference is invisible to a count alone. See CASRAI’s broader Citation Impact entry for the standard count-based metrics and their known limitations.

How this differs from citation-mapping tools

Scite is also a distinct category from citation-mapping/network-visualization tools such as Connected Papers or ResearchRabbit, covered in CASRAI’s AI-powered research assistant tools guide. Those tools build a visual graph of how papers relate to one another (via shared references, co-citation, or bibliographic coupling) to help a researcher discover adjacent literature starting from a seed paper. Scite does not primarily build a discovery graph; it evaluates the substance of individual citing statements for a given work, which is a citation-quality/context task rather than a literature-discovery task. A researcher might reasonably use both: a mapping tool to find related papers, and Smart Citations to assess how a specific finding has actually been treated by the literature that cites it.

Access and scope

Scite is a commercial product (institutional and individual subscription tiers, per its own site); it is not a free, open service in the way Semantic Scholar or OpenAlex are. As with any vendor-supplied classification, the supporting/contrasting/mentioning label for any individual citation is a model output, not a human peer-review judgment — useful as a fast triage signal and a pointer to the relevant sentence, but worth reading the underlying source sentence yourself before relying on the classification for a consequential decision (e.g. evidence synthesis, systematic review, or assessing whether a cited finding remains supported).

Related terms

• Citation Impact — the standard count-based metrics Smart Citations is meant to complement, not replace.
• CiteScore, Journal Citation Reports — journal-level citation metrics, a different unit of analysis from a per-citation classification.
• AI-powered research assistant tools — the broader category guide, which also covers citation-mapping tools such as Connected Papers and ResearchRabbit.