Privacy

Last Updated: August 26, 2026

Welcome to ReferenceVerify, operated by Medikalai Yazılım Anonim Şirketi ("we," "our," or "us"). We are committed to protecting your personal data and your academic work. This Privacy Policy explains how we handle information when you use our website (referenceverify.com) and our services.

1. Data Minimization & Academic Privacy

Our core principle is protecting your intellectual property.

• No document storage: Text, bibliographies and .txt/.docx/.pdf files you paste or upload are processed in memory (partly in your browser, partly on our server) to verify references against open academic databases (CrossRef, OpenAlex, PubMed, Europe PMC, Semantic Scholar, DBLP, CORE, ERIC, DOAJ, arXiv, TR Dizin, İSAM, SOBIAD, TO-KAT, YÖK Thesis Center, Google Books, OpenLibrary) and, for web sources, the cited page itself (title confirmation). Uploaded files are never written to disk or archived; once the result is shown, the processed text is discarded. Legacy Word files (.doc) cannot be read in the browser, so they are converted to text on our server in memory only — nothing is stored.
• Never used for training, never sent to plagiarism databases: your text is not used to train any model and is not submitted to Turnitin or any other similarity database.
• The only exception is your own choice:if you are signed in and leave the "save this report to my account" box ticked, the report (reference list, verdicts and — in Citation Check — the sentences that were evaluated) is stored in your account so you can return to it. You can delete it at any time from My Activity. Nothing is stored for guest use.

2. AI and Language Models

The reference verification verdict itself does not use AI: each reference is algorithmically cross-checked against the records held by the open databases listed above. A large language model (LLM) is used only in the following, separately labelled steps:

• Citation Check: judging whether the sentence you wrote is supported by the text of the source you cite (the reference itself is still resolved algorithmically).
• Find Evidence: matching a claim to candidate sources.
• Literature review / gap analysis: synthesising the real records that were retrieved.
• Journal, reviewer and conference finders, topic alerts: turning a non-English topic into English search terms.
• Custom citation format: rewriting references according to a style you describe in words.
• Live assistant: answering usage questions.

For these steps we run an open-weight model (Gemma) on our own hardware. Only the short input needed for the step is processed (e.g. the sentence being evaluated and the passage selected from the source, a search topic, or reference strings) — never your whole manuscript. Every AI-assisted result is presented as decision support, with the evidence it relied on, not as a final verdict.

What the model sees, and what it never sees.Under no circumstances is an entire manuscript, thesis, dissertation or report sent to the language model. For Citation Check the model receives only the specific sentence or paragraph containing the author's claim, the bibliographic details of the cited work, and either the abstract or — where openly available — the relevant passages of the cited source's full text. For Find Evidence it receives the claim sentence and candidate records. Manuscript text submitted to Citation Check and Find Evidence is processed only on our own infrastructure; it is not sent to external commercial AI providers (such as OpenAI, Anthropic or Google), and if our local model is unavailable the step simply fails rather than falling back to an external service. Prompts, the source material shown to the model and the model's answers are not retained as training data and are not used to train or fine-tune any model; one user's content never becomes part of the model or of another user's results. The model does not make the final decision on whether a citation is valid — it supplies evidence for editorial judgment.

3. Information We Collect

When you register for an account or purchase a subscription, we collect:

• Account Information: Name, email address, and account credentials.
• Billing Information: Financial transactions and billing details are processed securely through our Merchant of Record, Paddle. We do not store your credit card details on our servers.
• Abuse-prevention Signals: To apply usage limits correctly and to detect multiple accounts held by the same person, your account carries an irreversible hash of your IP address (not the address itself), the name of the network you connect from where it can be identified (e.g. a university network), normalised forms of your email address and name, and — only where configured — a keyed hash of your password. Your password, IP address or location cannot be recovered from these records, and they are not shared with third parties. Suspension decisions are never automatic: these signals only flag an account for human review, and you can appeal through our Contact page.

3a. Word add-in (Reference Verify for Word)

The Word add-in reads the open document only when you open its pane, and writes to it only when you press “Fix in document” (replacing that entry with its corrected form); any such change can be undone with Word’s Undo (Ctrl+Z). Only the text after the “References / Bibliography” heading is processed — exactly as if you had pasted it into the box on the website; the rest of the document is not read or stored. For citation checking (a separate tab you open yourself) the citing sentences are also read. Reference entries are looked up from your computer against open scholarly records (Crossref, OpenAlex and similar). The add-in requires a signed-in account and follows the same allowances and this same policy as the website; no data is shared with Microsoft beyond what Word itself needs to run an add-in.

4. Third Parties

• Academic databases: during verification, reference titles, DOIs, ISSNs and journal names are sent to the open academic services listed in section 1. The body text of your manuscript is never sent to them.
• Payments and e-mail: Paddle (merchant of record), Google Workspace and Resend (transactional e-mail).
• We do not sell your data.

5. Contact Us

If you have any questions about this Privacy Policy, please contact us at:

• Company Name: Medikalai Yazılım Anonim Şirketi
• Address: Çalca OSB Mah. 1. Cad. No:1 İç Kapı No:3, Merkez / Kütahya
• Phone: +90 507 241 63 44 · +90 545 669 88 77
• Email: support@medikal.ai