Scite
Scite is a research search engine and AI assistant for researchers, doctoral candidates, students and institutions. Answers come from 300M+ indexed papers, and each citation is classified supporting, contrasting or mentioning, with a link to the cited sentence.
What is Scite?
Scite is an AI-assisted research platform published by Scite, LLC, a Research Solutions company. Its founding building block is the Smart Citation: a deep learning model reads the sentence in which one paper cites another and classifies it as supporting, contrasting or mentioning. More than 1.6 billion citation statements have been analysed this way, so a paper is described by how it has been cited, not simply how many times.
The index covers 300M+ articles, preprints, books, patents and datasets, drawn from 1.8 billion unique citations, with 32M+ articles held in full text. It combines the whole open access corpus with 35 publisher indexing agreements listed by name in June 2026, among them Wiley, SAGE, Cambridge University Press, the American Chemical Society, BMJ, the Royal Society of Chemistry, Wolters Kluwer, Thieme, Karger, PNAS and The Royal Society. Coverage extends beyond journals to 90M+ patent families, 719K+ clinical trials and 3.5M+ funding records. The company's own framing is blunt: When other tools hit a paywall, they guess. Scite reads the paper.
Five functions sit on that index. Assistant answers questions from indexed papers. Search queries the citation statements extracted from full text. Reports give a quantitative and qualitative view of every citation a paper has received. Collections build and monitor a corpus, importing from Zotero, Mendeley, a CSV of DOIs or a set of results and raising alerts on new citations. Reference Check audits the references of a manuscript PDF and flags retractions, editorial notices and contrasting results.
What separates it from a general-purpose assistant is traceability: every claim links back to the specific sentence in the specific paper it came from, checkable in one click. The same corpus is reachable outside the web app, through an API documented on docs.scite.ai with key authentication and enterprise SSO, and through an MCP server credited with 7,454,599 uses across ChatGPT, Claude, Claude Desktop, Claude Code, Gemini CLI and Copilot Studio, with a connector verified in the Claude directory since 28 April 2026. Institutional holdings resolve automatically to the full-text PDF. Scite claims 2,000,000+ users and cites the ISMTE People's Choice Award and the ALPSP Award for Innovation in Publishing. It also disclaims accuracy: the Service is provided as is, and deciding whether to cite a paper remains the reader's call.
What it does
- Ask a question and get an answer sourced from real papers, with a link to the exact sentence it came from
- See whether a finding was supported or contradicted by later work
- Search inside the full text, not just abstracts
- Generate a citation report for any paper in the index
- Check the references of a manuscript PDF, retractions included
- Build and monitor collections of papers
- Connect an AI assistant to the corpus through MCP or the API
When to use Scite / When not to
A quick filter to help you decide if Scite is the right fit.
When to use Scite
- Researchers and doctoral candidates running a literature review, who need to see how a paper has actually been cited rather than how often, and to keep tracking the field over time through Collections and alerts
- Authors preparing a manuscript, who must audit every reference for retractions and editorial notices before submission or publication
- Clinicians, pharmacologists and evidence-synthesis specialists checking whether a published result was later supported or contradicted
- Librarians and institutions whose users need full-text access to paywalled content through the 35 publisher agreements behind the index
- Heavy users of ChatGPT, Claude, Gemini or Copilot who want their assistant grounded in verifiable literature through the MCP server or the API
When not to use Scite
- Anyone looking for a general-purpose web search engine or chatbot: the corpus is scholarly, not the open web
- Users who need a permanently free tool, since beyond the 7-day trial a paid subscription is required, including for the Claude connector
- Mobile-first users: there is no iOS or Android application
- Readers who need a non-English interface, as the site and the product are published in English only
- Under-13s, barred by the Terms, and anyone expecting legal or medical advice or bulk automated harvesting: scraping and automated access outside the API are explicitly prohibited, and the decision to cite a paper is left entirely to the user
How to use Scite
A typical end-to-end flow, from setup to results.
- Create an account from 'Start Free Trial' on the home page or the pricing page
- Sign in directly, or through MS Azure, ORCiD, Google or Facebook
- Choose your plan when you take the 7-day trial: it rolls into a paid subscription automatically unless you cancel before it ends
- Ask the Assistant a question, or run a full-text search with boolean queries and 25+ filters
- Open any paper in the index and generate its Smart Citation report
- Drop a manuscript PDF into Reference Check to audit every reference
- Build a collection from Zotero, Mendeley, a CSV of DOIs or a set of search results, then set alerts on it
- In ChatGPT, open the Scite app, authenticate with your Scite account and call @Scite in a prompt; reconnect before each session, as sign-out can be silent
- In Claude, go to Settings > Connectors, search for Scite, click Connect and sign in with your Scite account: the connector is verified and needs no configuration file
- For any other client, wire OAuth 2.0 manually against the authorize and token endpoints on api.scite.ai/mcp/oauth/ with the 'mcp' scope; to stop, go to your account, then Subscription, then Cancel Plan, at least 30 days before the following month
Pros & Cons
Pros
- Citation classification at a scale no competitor matches: you see how a paper is cited, not only how often
- Full-text access to paywalled content through 35 publisher agreements, where other tools stop at the abstract
- Traceability: every claim points back to the precise sentence in the source paper, checkable in one click
- A contractual commitment not to train on customer data, written twice, in the Terms and in the privacy policy
- Fits into tools already in use, from ChatGPT, Claude, Gemini and Copilot to Zotero and Mendeley, without changing the workflow
- Coverage beyond journals, taking in patents, clinical trials, funding, preprints and datasets, with retraction detection in Reference Check and Collections
- A low and legible entry price for an individual researcher at 20 USD per month on annual billing, backed by industry recognition (ISMTE, ALPSP) and NSF and NIH/NIDA grants
Cons
- No permanently free plan: past the 7-day trial everything is paid, and the trial converts into a paid subscription automatically if you do not cancel
- Cancellation must be made at least 30 days before the following month or that month is billed, and fees are non-refundable
- The displayed 20 and 50 USD per month prices assume an annual commitment, the month-to-month rate is not shown, and Team and Enterprise pricing is not public at all, only through sales@scite.ai
- MCP credits are capped at 250 per month on Basic, which heavy use through ChatGPT or Claude can exhaust
- No mobile application, and an English-only interface
- Light compliance documentation: no DPA published, no list of sub-processors and no security certification claimed, with SOC 2 and ISO 27001 returning zero occurrences, and the acronym GDPR appearing nowhere even though EU and UK rights are covered
- Targeted advertising and sharing with advertising networks, potentially a sale under Californian law, 10-year retention after last contact or account closure, contractual liability capped at 100 USD with mandatory New York arbitration and a class-action waiver, support routed through an address on another domain (researchsolutions.com), and robots.txt and llms.txt both returning 403 although the footer links to llms.txt
Pricing & Plans
There is no permanently free plan. Scite publishes two individual tiers, Basic at 20 USD per month and Pro at 50 USD per month, both quoted as billed yearly, so the lowest paid entry point is 20.00 USD per month on annual billing, with announced savings of 96 USD on Basic and 120 USD on Pro against the non-annual arrangement. A 7-day free trial is offered on both plans, after which the selected plan is charged. Team and Enterprise pricing is not published and is handled through sales@scite.ai. Payment is taken in US dollars, charged monthly and non-refundable, through a third-party processor; the publisher does not store card details.
- unlimited Assistant use
- full-text search
- Smart Citation Reports
- Citation Alerts
- 250 MCP credits per month
- ChatGPT and Claude integrations and collections of up to 1
- 000 papers
- everything in Basic plus API access
- 2
- 500 MCP credits per month
- collections of up to 10
- 000 papers and access to the patents
- clinical trials and funding datasets
- the tab exists on the pricing page but its content is loaded on click and is not served in the HTML
- so no amount is public
- quotes go through sales@scite.ai or the contact sales form
- A 7-day free trial is common to both individual plans
Data, GDPR & hosting
A consolidated view of how Scite handles your data.
GDPR overview
The acronym GDPR appears nowhere on the site, and Scite never claims compliance in so many words. The privacy policy, effective 26 March 2026, nonetheless carries a dedicated section, 'Additional Disclosures for Residents of the European Union (EU) and United Kingdom (UK)'. It sets out legal bases, namely contract performance, legal obligation, consent and legitimate interest, and rights of access, rectification, erasure, restriction, portability, objection and withdrawal of consent, plus the right to complain to a local supervisory authority. Data is processed in the United States; for transfers from the EEA the company states it implements the standard contractual clauses approved by the European Commission, with a copy available on request. Rights are exercised by writing to customersupport@researchsolutions.com with 'Privacy' in the subject. No Article 27 representative, no DPO, no published DPA and no named sub-processors.
Who owns the data?
Under the Terms, Scite, LLC owns the Service, its proprietary citation classifier and the usage data it generates, together defined as Company Materials. That ownership stops short of the article text itself, which belongs to third-party publishers, and of the AI Content: the company states it acquires no right, title or interest in it. Users keep ownership of the Media they upload but grant a worldwide, non-exclusive, royalty-free, irrevocable, perpetual, sublicensable and transferable licence to operate the Service, a licence that expressly excludes training, fine-tuning or retraining any model. Where FERPA applies, the institution retains ownership and control of student records.
Reuse rights
The privacy policy, effective 26 March 2026, lists the declared purposes: delivering the Service, communicating with users, usage analytics, marketing and targeted advertising, security and legal obligations. The categories collected include identifiers, financial and commercial data, professional data, communications, generated content, authorship information, network activity, device data and inferences; signing in through MS Azure, ORCiD, Google or Facebook passes name, email address and language preference. Cookies fall into four buckets: strictly necessary, analytics through Google Analytics, functional and targeting. One exclusion is stated twice, once in the Terms and once in the policy: Customer Data, defined broadly enough to cover inputs, outputs, queries and usage data, is never used to train, fine-tune or improve any large language model or other machine learning system. Sharing identifiers, network activity and device data with advertising networks may qualify as a sale under Californian law, although no data is sold for money and no data on minors under 16 is sold, and no automated decision-making produces legal effects. Authorship Information is made public by design, described as a fundamental aspect of the service. On the user's side reuse is bounded rather than free: the Terms prohibit scraping and automated access outside the documented API.
Data retention & training
Hosting summary
Scite declares processing in the United States: as a US company, most of its operations are conducted there and personal information will be processed there in order to provide the Service. For transfers originating in the European Economic Area, the privacy policy states that the standard contractual clauses approved by the European Commission are implemented, with a copy of the mechanism available on request, and more broadly that appropriate protections are provided for cross-border transfers as required by applicable law. Nothing further is named: no hosting region, no data centre and no infrastructure provider appears in any of the documents, and security is described only as reasonable technical and organisational measures, without detail or certification. One network observation, outside anything the site declares: the public IP resolves to an Amazon CloudFront edge node in Zaventem, Belgium, which is a content delivery point of presence rather than a storage location, and not the publisher's country.
Things to keep in mind
Risks and trade-offs to weigh before adopting Scite.
- The 7-day trial converts into a paid subscription automatically: cancel before it ends or you will be billed
- Outside the trial, cancellation must be made at least 30 days before the following month, otherwise that month is due, and fees are non-refundable; the displayed prices assume an annual commitment
- MCP credits are limited to 250 per month on Basic, and heavy use through ChatGPT or Claude can exhaust them mid-project
- Targeted advertising and sharing with advertising networks may qualify as a sale under Californian law; a 'Do Not Sell or Share My Personal Information' link is announced at the bottom of the site
- Data is kept for 10 years after your last contact or after the account closes, and Authorship Information is made public by design
- Mandatory individual arbitration in New York, a class-action waiver and a 100 USD liability cap; the Service is provided as is, with no accuracy warranty, and no security certification, DPA or sub-processor list is offered
- Convenience can quietly erode judgement: a classified citation and a sourced answer are a starting point, not a verdict. Deciding whether a paper supports your claim, and whether to cite it, stays with you, and the fact that Scite indexes medical and clinical literature does not make it a clinical decision-support tool
Setup & Integrations
Technical difficulty
Low for most users. Web use requires only an account and no installation, and sign-in can go through MS Azure, ORCiD, Google or Facebook. The Claude connector takes three clicks in Settings then Connectors, is verified by Anthropic and needs no configuration file, while ChatGPT only asks you to open the Scite app and authenticate. Effort rises at the edges: a manual OAuth 2.0 integration with published endpoints is available for Copilot Studio, Power Apps or any non-standard client, and the API requires creating a key, documented in a quickstart guide alongside a dedicated AI onboarding page for coding agents.
Deployment
Integrations
Behind Scite
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What makes Scite different from a general-purpose AI assistant?
What is a Smart Citation?
What does the index cover?
Is there a free trial, and what happens when it ends?
Is there a permanently free plan?
How much does a subscription cost?
Is there a student discount?
Is there an API?
Can Scite be plugged into ChatGPT or Claude?
Is my data used to train models?
Should you pick Scite?
Scite is a deliberately narrow tool: scientific literature, not general-purpose search. Its value rests on an asset competitors do not have, citation classification at scale, with 1.6B+ citation statements labelled supporting, contrasting or mentioning, and 35 publisher agreements that put 32M+ full-text papers within reach where other assistants stop at the abstract. Answers stay traceable to the sentence they came from, which is what turns a plausible summary into something you can cite. The undertaking not to train on customer data is unusually explicit, written into both the Terms and the privacy policy rather than a marketing page.
The trade-offs are just as concrete. There is no durable free tier: seven days, then a subscription that starts automatically unless cancelled, at 20 USD a month for Basic or 50 for Pro, and only on annual billing since the month-to-month rate is not published. Anything above the individual tiers is priced privately, through sales. The compliance paperwork is thin for an institutional buyer: no DPA, no named sub-processors, no security certification, and the acronym GDPR appears nowhere on the site, even though EU and UK rights, legal bases and standard contractual clauses are all set out in a privacy policy effective 26 March 2026. Retention runs to 10 years, liability is capped at 100 USD and disputes go to individual arbitration in New York.
The service is published by Scite, LLC, backed since late 2023 by Research Solutions, a listed group. For a researcher, doctoral candidate, editor or evidence-synthesis specialist who reads and cites every day, the entry price is modest against the time saved. For an occasional user, or an institution that needs signed data-protection documentation before it can buy, the case is much weaker.
- Choosing a selection results in a full page refresh.
- Opens in a new window.