Home · ← Elicit | 7/8 | Gemini →

Scite

Smart citations — supported, or contradicted?

Overview

Scite

Not "was this cited" — "did later work agree with it."

What it is

What it is

scite.ai — shows whether a paper has been supported or contradicted by later citing work, not just how many times it was cited.

Where it fits

Where it fits

D1·1Environment check
D1·2AGENTS.md
D1·3Literature tools
D1·4Referee skill
D1·5Subagents
D2·1Hypothesis
D2·2Design
D2·3Data lab
D2·4Results
D2·5Manuscript
D2·6Governance

The specific D1·3 check (“has this paper been contradicted?”) — plus D1·4/D1·5, since literature-scout accepts a Scite result the same way it accepts a Semantic Scholar one, and D2·5, since a citation backed this way can end up in the manuscript.

Access & cost

Access & cost

scite.ai, ≈ €16/mo Premium — check whether UvA library access already covers it before buying seats.

In the course

D1·3 exercise

Open each of your 3 Semantic Scholar results in Scite — has it been contradicted since publication? If a citation's been contradicted, the literature-scout rule says report both sides, not just the flattering one.

Why it's here

Why it matters for this course's case

The shared dataset has a deliberate spurious-causality trap. Scite-style thinking (has this been challenged since?) is the literature-side mirror of identification-skeptic's confound-hunting on your own results.

In the course

Optional in the hosted app, too

This app's setup form has an optional Scite API key field — if set, literature results show support/contradiction inline (app/pipeline/literature.py).

In the course

Feeds the manuscript, too

literature-scout accepts a Scite result you've pasted into the chat as valid support for a citation — the same standing as a Semantic Scholar search result. Anything Scite surfaces can end up cited in the manuscript (D2·5), and gets cross-checked again by the referee skill's citation check (D1·4).

Limits

A caution worth repeating

The hosted app's Scite integration is a best guess at their API shape, coded to fail silently rather than break the literature stage. Worth checking against Scite's actual docs before relying on it for anything beyond this course.

How it fits together

How it composes

Semantic Scholar finds it, Scite tells you if it held up, Consensus/Elicit tell you what it's saying — three different questions about the same paper.

Limits

What it doesn't replace

Doesn't verify a citation is real in the first place — that's still the Semantic-Scholar-or-nothing base rule.

Cost

Cost trade-off

The only paid-only tool in this stack with no meaningful free tier for real use — which is why the course proposal's “Recommended” cost tier includes it specifically, rather than leaving it optional-and-skippable like the rest.

Recap

Recap

Back to D1·3's decision graph on the home page.