Before you begin

VerifyScience uses Claude AI (by Anthropic) to assess research papers. To run verifications, you need two things:

A Google account

VerifyScience signs you in using Google OAuth. No separate password is created.

An Anthropic Claude API key

A key from console.anthropic.com. This is separate from any Claude subscription. API usage is billed by Anthropic directly at approximately $0.25 per million tokens โ€” each verification uses around 4,000โ€“10,000 tokens.

Why your own API key? VerifyScience does not hold a shared Claude key. Using your own key means your usage is under your control, billed directly to you by Anthropic, and subject to your own Anthropic account limits. Your key is encrypted with AES-256-GCM before storage โ€” VerifyScience never logs or returns raw key values.

1 ยท Sign in

  1. Go to verifyscience.co.uk

    The sign-in button is in the top-right corner of the page.

  2. Click "Sign in with Google"

    A Google authentication window opens. Select the Google account you want to use.

  3. Your account is created automatically

    First-time users are placed on the Free tier (3 verifications per month). The navigation bar now shows your verification count and account menu.

Sessions last 30 days. You will not need to sign in again unless you explicitly log out or clear your browser cookies.

2 ยท Add your Claude API key

This step is required before you can run your first verification. You only need to do it once.

  1. Get an Anthropic API key

    Go to console.anthropic.com, sign up or log in, navigate to API Keys, and create a new key. It begins with sk-ant-. Copy it โ€” you will not be able to see it again.

  2. Add a payment method to Anthropic

    Anthropic requires a payment method before API calls will succeed. Go to Billing in the Anthropic console and add a card. You will only be charged for actual usage.

  3. Paste your key into VerifyScience

    In VerifyScience, click Account in the navigation bar, then Claude API Key. Paste your key and click Save. The key is encrypted immediately โ€” it will not be visible again after saving.

Once your key is saved, the verification interface unlocks and you can submit your first paper.

3 ยท Submit a paper

Three input methods are available. Choose whichever suits the paper you are assessing.

๐Ÿ”—
URL (recommended)

Paste the full article page URL. Works reliably with PLOS, PubMed Central, arXiv, and bioRxiv. Use the full article URL โ€” not a short doi.org link, which is too brief to process.

๐Ÿ“‹
Text paste

Paste the abstract, methods, and results sections directly. Use this for paywalled papers (NEJM, Nature, Lancet) where the full page cannot be fetched.

๐Ÿ“„
File upload

Upload a PDF, DOCX, or other supported format. The text is extracted and analysed. Maximum file size: 10 MB.

  1. Choose your input method

    Select URL, Text, or File using the tabs above the input area.

  2. Paste the URL, text, or upload the file

    For URL submissions, use the full article page URL. Minimum 50 characters.

  3. Click "Verify Paper"

    A loading indicator appears. Your verification is queued and processed in the background. You do not need to keep the page open โ€” you will receive an email when the analysis is complete.

  4. Wait for results

    Typical times: Free / Starter โ€” 30โ€“90 seconds. Professional โ€” 2โ€“4 minutes. Enterprise โ€” 3โ€“5 minutes. Professional and Enterprise analyses evaluate more criteria in greater depth, which takes longer.

Paywalled papers: URLs from journals such as NEJM, Nature, and The Lancet return a login page rather than paper content. For these, copy and paste the abstract and methods section as text instead.

4 ยท Read your results

Overall confidence score

The confidence score (0โ€“100%) is a holistic assessment of the paper's credibility โ€” constrained to within ยฑ20 points of the per-criterion scores below it. It reflects methodology, publication context, post-publication record, and the weight of subsequent evidence.

85 โ€“ 100Strong methodological rigour โ€” well-documented methods, appropriate analysis, candid limitations. 65 โ€“ 84Adequate with minor concerns โ€” broadly sound, one or two addressable weaknesses. 50 โ€“ 64Moderate โ€” meaningful methodological concerns identified. Warrants careful reading before reliance. Below 50Significant concerns โ€” substantive issues affecting the reliability of findings. Exercise caution.

Per-criterion analysis

Each criterion receives an individual score and a detailed explanation. The number of criteria assessed depends on your subscription tier: 4 (Free), 8 (Starter), 12 (Professional), 16 (Enterprise). Higher tiers assess more dimensions and provide longer explanations.

Flags

Flags are specific concerns identified within a criterion โ€” for example, "no formal power analysis reported" or "high self-citation rate." They are not disqualifying by default; they are signals that warrant attention when interpreting the paper. A paper with flags may still be methodologically sound overall.

Summary

An executive summary synthesising the paper's key strengths, main concerns, and overall credibility assessment. Professional and Enterprise summaries are substantially longer and more detailed.

Domain calibration Professional+

On Professional and Enterprise plans, a domain indicator appears below the paper title in your results: Assessment calibrated for: [domain] โ€” for example, Assessment calibrated for: Clinical trial methodology or Assessment calibrated for: Qualitative research.

This indicator tells you that VerifyScience detected your paper's research domain and adjusted the assessment criteria accordingly. A qualitative interview study is not evaluated by the same criteria as a randomised controlled trial. The specific frameworks applied are proprietary and not named โ€” this is intentional, to protect the integrity of the assessment and prevent results from being gamed. Free and Starter accounts do not see this indicator.

Important: VerifyScience results are AI-assisted assessments intended to accelerate human expert review โ€” not to replace it. Always apply your own expert judgement before making decisions based on a verification result. Full methodology disclosure is at verifyscience.co.uk/methodology.

5 ยท Compare two papers Professional+

Comparative verification analyses two papers side by side in a single operation. It is useful for comparing a preprint against its published version, evaluating two competing methodologies, or triaging which of two papers warrants deeper expert review.

Requires Professional or Enterprise plan. Comparative verification consumes 2 verifications from your monthly quota โ€” one per paper.
  1. Navigate to the Compare tab

    The Compare tab appears in the main verification interface, to the right of the single-paper submission area.

  2. Enter Paper A and Paper B

    Paste the URL or text for each paper in the respective fields. Both papers are analysed in parallel โ€” neither waits for the other to complete.

  3. Click "Compare Papers"

    Both analyses are queued simultaneously. Allow 2โ€“4 minutes for Professional tier. You will receive an email when both are complete.

  4. Read the side-by-side results

    Results appear with both confidence scores, a delta showing which paper scores higher overall, and a per-criterion comparison across all assessed dimensions.

6 ยท Scan a reference list

Reference list verification scans an entire bibliography for retraction flags, citation metrics, and methodological concerns. Navigate to the Ref Scan tab in the main navigation.

Three tiers

Tier 1 โ€” Signal Scan

Instant. Cross-references every citation against OpenAlex and Crossref for retraction status, citation count, and open access. Free for all tiers.

Tier 2 โ€” Quick Check

AI-powered abstract triage on individual references. Flags methodological concerns in seconds. Costs 1 verification credit per reference.

Tier 3 โ€” Deep Verify

Full 12-criterion professional assessment on any reference that warrants closer inspection. Costs 1 verification credit per reference.

  1. Paste your references

    Paste one reference per line โ€” URLs, DOIs, or formatted citations. The system extracts identifiers automatically.

  2. Run the signal scan

    Click "Scan References" to run the Tier 1 signal scan across all citations. Results appear in seconds.

  3. Upgrade individual references

    Select specific references and click "Quick Check" (Tier 2) or "Deep Verify" (Tier 3) for AI-powered analysis on those citations.

Selection limits by tier: Free users cannot run reference scans. Starter: up to 10 references. Professional: up to 50. Enterprise: unlimited.

7 ยท Export results

Ten export options are available from the results view โ€” one PDF report, two machine-readable formats for reference managers, six citation style formats, and a share link.

PDF Full formatted report with confidence score, per-criterion scores with colour bars, explanations, flags, and executive summary. Includes an EU AI Act disclosure. Supports Unicode โ€” Greek, mathematical notation, and CJK characters render correctly.
BibTeX Machine-readable citation entry with the VerifyScience confidence score embedded in the note field. Import directly into Zotero, Mendeley, LaTeX, or any BibTeX-compatible tool โ€” the score travels with the citation. File extension: .bib
RIS Machine-readable format for reference managers including EndNote, Mendeley, and Papers. Confidence score included in the N1 notes field. File extension: .ris
APA 7th Formatted citation in APA 7th edition style (American Psychological Association) with VerifyScience score appended. Used widely in social sciences, education, and psychology.
MLA 9th Formatted citation in MLA 9th edition style (Modern Language Association) with confidence score. Standard in humanities, literature, and language studies.
Chicago Formatted citation in Chicago Manual of Style with VerifyScience score. Used in history, arts, and social sciences.
Vancouver Formatted citation in Vancouver style with confidence score and verification date. Standard in medicine, nursing, and life sciences.
IEEE Formatted citation in IEEE style with VerifyScience score. Standard in engineering, computer science, and electronics.
ACS Formatted citation in ACS style with confidence score. Standard in chemistry and chemical engineering.
Share link Generates a public URL for the result. Recipients do not need a VerifyScience account to view it โ€” useful for sharing with collaborators or supervisors.

8 ยท Manage your quota

Your monthly verification quota is shown in the navigation bar as X / Y โ€” verifications remaining out of your monthly total. It resets on your billing anniversary date, not the first of the calendar month.

If a job fails

Your verification credit is refunded automatically if a job does not complete within 30 minutes. No action is required โ€” the credit appears in your quota without needing to contact support.

Upgrading your plan

Click Upgrade in the navigation bar. The new quota is applied immediately in full โ€” you do not wait for the next billing cycle. Upgrades are prorated automatically by Stripe.

API calls and GUI submissions share the same monthly quota. If you are using both, the combined total counts against your limit.

Chrome extension

The VerifyScience Chrome extension lets you verify papers directly from PubMed without leaving the page.

Install

  1. Install from the Chrome Web Store

    Search for "VerifyScience" in the Chrome Web Store, or use the direct link from verifyscience.co.uk. Click "Add to Chrome".

  2. Pin the extension

    Click the puzzle piece icon in Chrome's toolbar and pin VerifyScience for easy access.

Connect your account

  1. Click the VerifyScience icon

    You will see a "Connect to VerifyScience" button. Click it.

  2. Authenticate

    A new tab opens at verifyscience.co.uk. If you are already signed in, the connection completes automatically. If not, sign in with Google first. The tab closes on its own after 3 seconds.

Verify a paper

  1. Navigate to a PubMed article

    Go to any paper on pubmed.ncbi.nlm.nih.gov.

  2. Click the extension icon

    The popup shows the paper's title, DOI, and PMID extracted automatically. Your quota and subscription tier are displayed.

  3. Click "Verify This Paper"

    The extension submits the paper's abstract and metadata for analysis. Results typically appear within 60 seconds โ€” directly in the popup.

  4. View the full report

    Click "View Full Report" to open the complete results on verifyscience.co.uk, including external signals, domain calibration, and all per-criterion details.

Supported sites

PubMed (pubmed.ncbi.nlm.nih.gov). bioRxiv and Google Scholar support arriving soon.

Privacy

The extension accesses only paper metadata from PubMed pages. No browsing data is collected. Data is transmitted only when you click "Verify This Paper". Full details in our Privacy Policy.

Common questions

I saved my Claude API key but the Verify button is still greyed out.
Ensure your Anthropic account has a payment method added at console.anthropic.com. The API key itself is valid, but API calls will be rejected by Anthropic until billing is active. Try refreshing the page after adding a card.
My job shows "processing" for a long time and never completes.
Jobs time out after 30 minutes and your credit is refunded automatically. If you submitted a paywalled URL, the system may have retrieved a login page rather than the paper โ€” try resubmitting as text. If the issue persists, email [email protected] with the job ID shown in your verification history.
The confidence score is higher (or lower) than I expected from the criterion scores.
The overall confidence score is a holistic assessment, constrained to within ยฑ20 points of the mean of the criterion scores. It reflects the paper's full academic standing including its citation record, post-publication scrutiny, and field context, in addition to its methodological scores. A server-side guardrail enforces this constraint automatically.
Can I submit a DOI instead of a URL?
Yes. You can paste a DOI (e.g. 10.1136/bmj.n71) or a doi.org URL (e.g. https://doi.org/10.1136/bmj.n71) directly into the URL input. The system resolves DOIs automatically. For best results with paywalled publishers, use the full article page URL from PubMed Central or the publisher's site, or paste the abstract as text.
Does VerifyScience work with papers in languages other than English?
Yes. The analysis engine works in any language Claude supports, including Spanish, French, German, Chinese, Japanese, and Arabic. The assessment output is returned in English regardless of the paper's language.
Is my paper content stored by VerifyScience?
The submitted content is stored in the verification job record so you can retrieve your results later. It is not used to train AI models. Data is held in Supabase on EU infrastructure. You can delete your account and all associated data at any time from Account โ†’ GDPR โ†’ Delete my data.
How do I get access for my whole research group?
Institutional access with multi-user dashboards, aggregate analytics, and SSO integration is available. Visit verifyscience.co.uk/institutional or email [email protected] to discuss pilot arrangements.

Ready to verify your first paper?

Free tier includes 3 verifications per month โ€” no payment required to start.