Scite classifies what a citation actually does to a claim

The writer's problem

A citation manager like Zotero or EndNote stores a reference; it does not tell a writer whether the paper behind that reference still holds up. An academic author or a nonfiction writer building a source list often needs to know whether a later study supported, challenged, or simply mentioned an earlier one before repeating its claim. Scite's homepage and pricing page, as retrieved on 16 September 2026, describe a product built to answer that question rather than to organize a bibliography.

What the documents show

The homepage states that its Smart Citations show how later research has supported, challenged, or discussed a paper, claim, method, journal, author, institution, or funder, a three-way classification distinct from a manager that only records that a citation exists. The same page states the index now covers 1.6B+ citations across 300M+ scholarly sources, and describes its AI assistant as one where every answer is grounded in real papers, never generated or hallucinated — a claim about the product's design made by the company itself, not an outside audit result. The pricing page lists a free Connect tier with 25 monthly credits and no Assistant or Search access; a Basic tier at $20 a month, billed yearly, adding full Assistant and Search access and Smart Citation reports for collections up to 1,000 papers; and a Pro tier at $50 a month, billed yearly, adding API access and collections up to 10,000 papers plus patent, clinical trial, and grant datasets. The homepage also states Scite is now part of Research Solutions, a corporate detail worth noting when judging how long a given pricing structure might hold.

The editorial choice

This is an editorial distinction worth keeping visible in a manuscript's method notes: a supporting or contrasting label from Scite describes a relationship the company's own classifier assigned, not a peer-reviewed consensus. An editor working with an author who cites Scite's classification to argue a finding is contested should ask the author to read the contrasting paper itself, since the tool's role is to surface the relationship, not adjudicate it.

What stays with the author

Scite's own materials describe what its classifier looks for; they do not describe how a specific contested classification was reached in an individual case. Reading the flagged paper before repeating a contested or supported label in print remains the author's responsibility, and neither page claims to replace that reading.

• Is a citation labelled supporting or contrasting here taken from Scite's classifier, or from a direct reading of the paper itself?
• Which paid tier produced the classification, and does its collection-size limit matter for this manuscript's source list?
• Does the never hallucinated claim rest on an independent test the pricing page cites, or only on the vendor's own description?

A classifier that names its categories in public documentation is easier to interrogate than one that does not; naming them is not the same as settling the argument they describe.