Scite follows the evidence behind a scientific citation

Scite helps researchers investigate what happens after a paper is published. It connects papers with the passages in later literature that cite them, and classifies those citation statements as supporting, contrasting or mentioning. That gives a reader a more useful starting point than a citation count alone. Its search, Assistant and integrations build on this evidence layer, helping users find relevant research and then examine how a particular claim has been treated.

01 / ProductWhat Scite provides

Scite is part of Research Solutions and combines literature discovery, Smart Citation reports, an AI research Assistant and tools for monitoring collections of papers. Its features page² describes citation context, reference checking and collections that can be assembled from searches or imported research libraries. These capabilities support different questions: finding a paper, understanding its reception and checking references in a manuscript are related tasks, but they are not interchangeable.

The central distinction is between a paper and a citation statement. One paper can be cited for its dataset, a method, a background observation or its main finding. A later author may use its method while disputing one result. Looking at the passage and section where the citation occurs helps prevent those relationships from collapsing into a single positive or negative judgement.

Scite's coverage page⁶ separates indexed article records from full-text sources and extracted citation statements. As checked in September 2026, it reports more than 317 million article records and 41 million full-text sources. Those figures describe different layers of the collection. An indexed record does not mean every part of that paper is available for inspection through the product.

02 / AudienceWho benefits most

The strongest fit is a researcher who already has a question precise enough to follow through the literature. Examples include checking whether an influential experiment has been independently tested, finding how later studies adapted a method, or understanding why two reviews interpret the same evidence differently. A useful starting point can be a DOI, a known paper or a narrowly defined claim.

Scite also suits research support teams that need to monitor a collection over time. New citations can reveal follow-up work, changed interpretations or a reason to revisit a statement in an existing report. The benefit is a more directed reading queue: the researcher can focus on passages that may matter to the claim under review instead of opening every paper that happens to cite it.

For a first survey of an unfamiliar field, compare the Elicit blueprint, particularly where structured study extraction is central. For public-web research that includes company documentation and current events, see the Perplexity blueprint. Scite's distinguishing question is how scholarly sources use evidence, which is narrower and often more useful than asking it to cover every kind of research material.

03 / WorkflowA workflow for checking a technical claim

The following is a proposed workflow for a research team assessing a claim about an NLP benchmark. Suppose a draft says that a particular training method improves generalisation across several datasets. Start by identifying the original paper and writing the claim in a precise form: which model family, which datasets, which evaluation split and which outcome measure are involved?

Find the relevant citation passages

Open the paper's citation report and inspect a range of supporting, contrasting and mentioning statements. Avoid starting with a tally. A supporting statement about computational efficiency does not directly support the separate claim about generalisation. A contrasting passage may concern a different experimental setting. The reading task is to identify which passages actually bear on the sentence the team plans to publish.

The Citation Statement Search guide³ explains how results connect the citing article, cited article, surrounding passage and section. This allows searches for a method or dataset where the useful detail appears inside a paper rather than in its title or abstract. Read enough surrounding text to understand why the reference was included.

Create a small claim matrix. Each row should contain the citing paper, the precise claim discussed, experimental conditions, citation passage and the team's interpretation. Add an explicit field for whether the work uses the same underlying dataset or experiment as another row. Five articles discussing one experiment should not become five independent replications in the team's summary.

Read the strongest challenges and confirmations

Open the original papers behind the most relevant passages. Compare their experimental setup with the source of the initial claim. For the benchmark example, a change in evaluation split, data contamination checks or model size can explain an apparent disagreement. Record that difference directly instead of describing the literature as simply divided.

Use ordinary search filters to expand the candidate set around an author, date range or topic. The search guide⁴ describes combining metadata filters to narrow results. Run a broader query as well as the narrow one so that a naming change does not exclude a relevant follow-up study. Save the search terms with the matrix.

The Assistant can help organise a follow-up question, but the final synthesis should point back to the claim matrix. Ask for differences in conditions and unresolved issues, then inspect the sources behind each important answer. A useful synthesis might conclude that the method performs well under one evaluation design while the evidence for a broader statement remains incomplete.

Finish by revising the draft sentence to match that scope. Save the paper collection and arrange a periodic review if the claim remains important to a product or research programme. The resulting record should let another colleague see the route from the original assertion through later evidence to the narrower wording the team adopted.

04 / PricingPricing for research and integrations

The official pricing reference¹, checked on 15 September 2026, distinguishes website access from MCP usage. The table uses US dollars. Annual prices are billed for the year; the lower monthly equivalent does not mean a month-to-month subscription at that rate.

The Free plan is principally an entry point for MCP use; it does not include the paid Search and Assistant experience. Basic is therefore the more relevant starting comparison for a researcher working mainly in the website. Pro increases integration capacity and collection limits, while the API and additional data access need their own terms checked against the planned use.

Choose an MCP credit allowance from the actual tool-calling pattern and the credit charges it produces. An assistant that searches, opens evidence and asks follow-up questions can call several tools during one visible conversation. A short user message is not a reliable unit for estimating consumption, and credits should not be assumed to equal requests. Start with a representative research assignment, record billed credits and decide where collected evidence can be reused between steps.

For a team, compare individual usage with the way allowances are allocated and shared. One person doing an intensive review may consume more capacity than several occasional readers. The important purchasing question is whether the chosen arrangement supports that uneven workload without making evidence access dependent on an individual account.

05 / DistinctionsWhy citation context changes the work

Scite's most valuable contribution is helping a researcher ask a better question about a reference. Instead of asking whether a paper has many citations, the reader can ask which result later authors used, whether they tested it under comparable conditions and what caveats they reported. That can turn an otherwise unfocused reading task into a sequence of concrete checks.

The classification is an aid to finding relevant passages. Supporting does not certify a study's design, and contrasting does not establish that the original paper is wrong. Both labels concern a particular citation statement. The strength of the evidence depends on methods, comparability, independence and the substance of the passage itself.

The coverage explanation⁵ also matters in everyday use. Metadata and full-text updates follow different schedules depending on the source. A recently published citing paper may therefore appear at one layer before its citation context is available at another. Absence of a classified statement is not evidence that no relevant later work exists.

Scite's MCP integration⁷ lets compatible assistants reach the same research resources through a connected tool. This is useful when the researcher already works inside another AI interface. Authentication and available holdings still shape access. Connecting the tool changes where the researcher asks the question; it does not make every cited source freely available in full text.

06 / QuestionsWhat to check before relying on it

Evaluate coverage using a small, known literature set from your own discipline. Include a highly cited foundational paper, a recent paper and one whose later interpretation is contested. Check whether relevant citation passages appear, whether the surrounding context is sufficient and whether the original full text remains accessible through the team's normal library arrangements.

Then assess classification usefulness on the passages that matter. Record cases where a label points to a relevant challenge and cases where the sentence concerns an unrelated aspect of the paper. This is more informative than judging the interface by the number of coloured labels. The practical question is how much directed reading the product saves while preserving the context needed for an accurate interpretation.

A team should also decide how it will handle reference checks in manuscripts. A flag is a reason to investigate the reference and its use in the draft. The appropriate response may be to replace a source, narrow a claim, add context or leave the reference after inspection. Keep that editorial decision visible so the next reviewer understands what was resolved.

07 / DecisionThe decision on Scite

Choose Scite when tracing the reception and use of published evidence is a recurring part of your work. Its citation passages and search tools give researchers a way to move beyond a list of papers and into the relationship between a claim and the literature around it.

Start with a known claim and assess whether the product exposes useful follow-up evidence that your usual search process misses or makes slow to locate. If it does, choose website access or MCP capacity according to where the team actually works. Retain a study-level record of the important findings and the conditions under which they apply.

Scite is most valuable when it improves the precision of a research conclusion. A shorter claim with a clear evidence trail is often a better result than a longer generated explanation or a larger count of supporting citations. That is the outcome to use when judging whether it earns a place in the team's research workflow.