Most "AI research tools" are search engines with a chatbot bolted on. Scite does something different and, in my experience, more useful: it has analyzed over 1.6 billion citation statements across 300 million+ scholarly sources and labeled each one as supporting, contrasting, or just mentioning the paper it cites. That last part matters. When I find a paper claiming X, the real question is not "does this paper exist" - it is "what happened when other researchers tried the same thing." Scite answers that directly, with the actual citing sentences on screen.
Two features carry the tool:
Smart Citations. Open any paper and you see how every later citation treated it. A cancer-nutrition study with 400 citations looks a lot less solid when 60 of those citations are explicitly contrasting. I have caught disputed-adjacent claims this way before they made it into client deliverables.
Reference Check. Upload your own draft (PDF or manuscript) and Scite checks your reference list against the citation record - flagging which of your sources were disputed by later work. It is a pre-submission smoke detector. I run it on every evidence-heavy article before delivery.
The Assistant is a chat layer on top: it answers questions with inline citations and shows the citing context for each. There is also an official MCP server, which means Claude or any MCP client can pull Scite's citation data straight into your workflow - that is how I run claim-checks without leaving my editor.
Pricing, Plainly
Official pricing as of October 2026 - the plan page is dynamic, so treat these as this month's numbers:
| Plan | Price | What You Get |
|---|---|---|
| Basic | $12/mo billed yearly (~$20 monthly) | Full Assistant & Search, 250 MCP credits/mo, Smart Citation reports, 1,000-paper collections, citation alerts |
| Pro | $40/mo billed yearly (~$50 monthly) | Everything in Basic + API access, $60 API credit (limited time), 2,500 MCP credits/mo, 10,000-paper collections, patent/clinical-trial/grant datasets |
| Enterprise | Custom | Unlimited pooled users, SSO/IP/domain access, 50,000-paper collections, FAERS/MAUDE/510(k) regulatory datasets, usage reports |
There is a 7-day free trial on Pro. There is no meaningful free tier - after the trial you are paying $144/year minimum. Team seats ride on one bill; universities usually buy Enterprise, so check institutional access before you pull out a card.
The Money Angle: Sell the Truth-Check
Nobody hires you to find papers - Semantic Scholar is free. What people pay for is *certainty*: knowing a claim survives contact with the citation record. Scite is the cheapest tool on the market that delivers that, and here is how I have turned it into invoices:
- Evidence dossiers ($500-$3,000 per review). Health-tech startups, supplement brands, and law firms regularly need to know whether the studies behind their marketing claims hold up. A dossier tracing 20-30 key claims through supporting and contrasting citations takes me 4-8 hours with Reference Check and the Assistant, and prices at $1,000-$2,000 for mid-market clients. One dossier covers a year of Pro.
- Cite-checking retainers for content agencies ($200-$500/mo per client). Content teams get burned by citing studies that later work contradicted. I run their draft claims through Scite monthly and deliver a "disputed/solid/retracted" report. It is 2-3 hours of work per client once the MCP workflow is set up.
- Systematic-review support for academics ($1,500-$5,000 per project). Screening citation context is the miserable part of a review. Scite does not automate the judgment, but it cuts the time per record dramatically.
- Build on the Pro API. The API plus the $60 credit is the quiet sleeper here - a citation-check bot or due-diligence tool wired into a client's workflow is a sellable product, not just a service.
The honest framing: Scite is not a fat affiliate program. You make money by selling the output - verified claims - and the tool makes that output 5-10x faster to produce.
How I Actually Use It Week to Week
- Claim triage. Paste the 15-20 load-bearing claims from a client brief into the Assistant, pull the citing context for each, and sort them into solid/disputed/mixed. About 90 minutes for a full brief.
- Competitor teardown. When a rival product cites "17 studies," I check how many of those are actually contrasting. Sales teams love that slide.
- MCP workflow. In Claude, I ask for a claim-check and the Scite MCP server returns classified citations inline. The 2,500 Pro credits cover roughly my monthly volume; Basic's 250 credits are enough for light personal use only.
- Collection building. I keep a 1,000-paper collection per client vertical with citation alerts on. When a key source gets disputed, the client hears it from me first.
Where It Falls Short (Honest)
- The labels are machine-made. Supporting/contrasting classification misfires on methods-heavy citations - a paper can be labeled "contrasting" for disagreeing about methodology while confirming the core finding. Scite shows the actual citing sentence, which is exactly why you can overrule a bad label in seconds, but never trust a label you have not eyeballed.
- No full-text rescue. Scite finds citation context for paywalled papers but usually cannot show the paper itself. Without institutional access you will hit walls.
- Coverage skews biomed/CS. Deep in medicine and tech, noticeably thinner in the humanities and in very recent preprints.
- The trial-to-paid cliff. Seven days is enough to learn the tool, not enough to finish a real review project. Plan the trial around a small pilot.
- Collection caps. 1,000 papers on Basic sounds generous until a systematic review shows up.
Scite vs Elicit vs Consensus
I run all three and they do not overlap the way people assume. Elicit screens and extracts across a defined paper set - it is the assembly line for a 300-abstract review (my full Elicit review). Consensus is the fast "what do papers generally say" aggregator - great for a first answer, risky as the final word (review here). Scite is the auditor: it does not find answers, it checks whether the answers survive. If you buy one for client work, it is Scite; if you buy two, add Elicit. The full four-way comparison with ResearchRabbit and Litmaps is here.
Who Should Pay for It
Grad students and clinicians with institutional access should run the trial first and check whether the library already subscribes. Independent researchers, consultants, and content people who sell judgment - that is who Basic pays for immediately. Pro is for anyone shipping client deliverables weekly or touching the API. Enterprise is for universities and pharma. If you never cite research in anything you sell, you do not need this tool at all - NotebookLM covers casual source-wrangling for free.