scite

I tested scite for evidence-based health writing where the question was not which papers exist on a topic but which papers support or contradict each other, with Elicit S172 and Consensus S172 already in the workflow for literature review batches and plain-English Q&A respectively. The thesis I wanted to evaluate: does the Smart Citations classification actually surface contradictions that raw citation counts miss?

What worked best was the Smart Citations dataset itself. 1.2 billion citations across 35 million papers, each classified as supporting, contradicting, or just mentioning the cited claim, is genuinely unique - no other research tool I have used does this. For a paper on adolescent SSRI safety, scite immediately surfaced 18 supporting citations and 6 contradicting ones from later replication studies, which is the kind of context that takes hours to assemble manually from Google Scholar. The Assistant feature layered a GPT-style chat over the full-text corpus and gave inline source links for every claim, with contradiction flags visible in the response. Institutional access via Shibboleth let me link my university journal subscriptions for full-text reads on paywalled papers.

What broke or surprised: the free tier is severely limited - 3 Assistant queries per day means a single morning research session hits the cap before lunch. Assistant response speed lagged Elicit S172 noticeably on long-form literature review tasks where I wanted summarization across 30-plus papers. The paper corpus is smaller than Google Scholar - on one test query Elicit surfaced 2 papers scite did not have indexed. The 20 USD per month Pro pricing is niche when free Google Scholar plus free Consensus tier exists, but the contradiction signal is what justifies the spend for evidence-based work specifically.

Verdict in cohort: scite for research where citation contradiction detection is the core need (evidence-based medicine, dissertation gap analysis, fact-checking scientific claims). Elicit S172 for fast literature review batches across many papers. Consensus S172 for plain-English Q&A and meta-conclusions across studies. The 3 tools are complementary not competitive for a serious researcher - run scite for the contradictions, Elicit for the synthesis, Consensus for the executive summary.