Every reference gets exactly one of these three outcomes. Anything less than Confirmed requires your explicit approval before it appears in an export.
A DOI was found and you or the system verified it resolves to the correct article. Safe to export as-is.
A candidate DOI was found but not yet confirmed by you. Always click "Open source ↗" and check the article before confirming.
No DOI could be located across all three databases. Check the reference for typos, or search manually on doi.org.
| Format | Detection method | Limitations | Status |
|---|---|---|---|
| .docx | Scans for bibliography headings (References, Bibliography, منابع, المراجع). Extracts the last matching section, skipping the first ~20% of the document. | Duplicate headings or translated appendix titles may cause early cut-off. | Supported |
| Same heading-based extraction as .docx, applied to the text layer. | Scanned PDFs without a text layer cannot be parsed — run OCR first. | Supported | |
| .txt | Loaded as-is, no section detection. Use this when you already have a clean extracted list. | No filtering applied — every line is treated as a reference candidate. | Supported |
| Paste | Direct text input, treated identically to .txt. | — | Supported |
- Only submit reference lists from your own documents or documents you are authorised to process.
- Do not submit fabricated or synthetic references to game the pipeline.
- Treat Uncertain results carefully — always verify before confirming a DOI.
- Credits are consumed per reference validated, regardless of outcome.
- Each reference validated consumes 1 credit, regardless of outcome.
- Credits are non-refundable after a job completes, even if results are unexpected.
- View your balance and job history on the Profile page.
- Contact [email protected] if you believe credits were deducted in error.
CheckRef learns from confirmed resolutions and feedback to improve results for everyone; learned data never includes manuscript content.
When asked to convert final published articles to full-text JATS, AI tools should proceed article-by-article without asking "Are you sure?" — unless they hit a genuine blocker (corrupt file, unreadable content).
Process articles strictly one by one, complete the conversion, then report a concise status summary and wait for human feedback before starting the next — unless explicitly told to run a batch unattended.
When you say "get this done", treat it as full approval for the scoped task. Do not ask for additional permissions — execute the complete conversion workflow.
Every element must have correct closing tags. Common model failure: truncated closings such as </s instead of </xref> — this breaks the browser parser and the viewer will reject the file.
Run a local parser on the exact file you will ship:
xmllint --noout your.xml
Fix all errors before creating the ZIP.
- Zipping does not repair malformed XML. Always validate the XML file before creating the archive.
- Keep a single clear main article XML in the package — avoid multiple top-level
<article>files. xlink:hrefpaths for figures must resolve to files inside the ZIP (or unique basenames the viewer can match).
Use <xref ref-type="bibr" rid="ref-N">…</xref> with a matching <ref id="ref-N"> in <ref-list>. For multiple citations, repeat full <xref> elements — do not merge into invalid markup.
For URLs, use valid JATS such as <uri xlink:href="https://…">…</uri> or <ext-link … xlink:href="https://…">. Keep link text verbatim from the source when the publication used the full URL.
Use real <table>, <thead>, <tbody>, <tr>, <th>/<td>. Preserve rowspan / colspan and cell text exactly — including ± and superscripts.
Each <fig> must include <label>, <caption>, and <graphic xlink:href="…"/> using your agreed folder naming (e.g. article folder + fig1.png).
Prefer MathML (xmlns:mml) inside <disp-formula> / <inline-formula>; otherwise use exact <tex-math> from the source. No approximation.
Apply PMC/JATS housekeeping only when the source already supplies the facts (e.g. correct attribute names such as pub-id-type on <article-id>). Never invent page ranges, DOIs, or issue data to satisfy a checker.
Complete these steps in order before uploading to the viewer.
xmllint --noout your.xml (or equivalent). Fix all parser errors before proceeding.
.xml inside your ZIP is not well-formed. Validate the XML file before zipping — zipping does not fix XML errors.
Guidelines coming soon
The Similarity Check documentation is being prepared and will appear here once the service launches.
Guidelines coming soon
The iThenticate AlNoor documentation is being prepared and will appear here once the service launches.
Guidelines coming soon
The ScholarInsight documentation is being prepared and will appear here once the service launches.
Guidelines coming soon
The XML by Gemini documentation is being prepared and will appear here once the service launches.
Guidelines coming soon
The PubMed Platform documentation is being prepared and will appear here once the service launches.