Anonymize a manuscript for double-blind peer review — without uploading the file

Why this matters (the 34% number, and where it actually comes from)

Anonymization is a process across multiple layers — textual, metadata, institutional, figure-content, and venue. This article handles the metadata layer cleanly and gives you a manual checklist for the textual layer. The starting point is a peer-reviewed primary source that almost no other article on this topic cites correctly.

• The leaks broke down into five recurring categories: author initials embedded in the manuscript text, references to the author's own work as 'in press', references to the author's previously-published papers (without third-person framing), institutional identifiers in figures (a hospital name on a CT scanner, a department logo on a chart), and institution names in the methods section.
• Almost none of these leak categories are metadata. They are textual. But the metadata layer — Word's Author field, the PDF's document GUID and producer string, the EXIF data on every embedded figure — adds a sixth class that the Katz study did not measure (because in 2002 most manuscripts arrived on paper or as fresh-from-Word .doc files). Today, the metadata channel is the easiest to leak and the easiest to fix.
• A 2017 study (Le Goues et al., 'Effectiveness of Anonymization in Double-Blind Review', arXiv:1709.01609) followed up at three software-engineering venues and found 70-86% of reviews contained no author guess and 74-90% had no correct guess — anonymization is imperfect but largely effective. A 2022 paper (arXiv:2211.07467) showed that neural networks can attribute authorship from text content alone with about 73% accuracy. Together: even with perfect metadata hygiene the textual channel exists. This article handles the metadata channel cleanly. The textual channel is a manual review step — there is a checklist at the bottom of the page.

Workflow A — you have a .docx manuscript (the Word path)

If the file you are sending started as a .docx, run this seven-step workflow. The first three steps happen inside Microsoft Word; steps 4-6 happen inside FileHop; step 7 is the manual textual review that no tool can do for you. Total time: about seven minutes.

1. 1

   Step 1: Work on a copy

   File → Save As → MyManuscript_anon.docx. Do every following step on the anonymized copy. Keep the original — you need it for the camera-ready version after acceptance, and it preserves authorship for your records (important if your institution requires retention).

2. 2

   Step 2: Accept all tracked changes and delete all comments — in Microsoft Word

   Review tab → Accept → Accept All Changes and Stop Tracking. Then Review tab → Delete → Delete All Comments in Document. Confirm Track Changes is OFF in the Review tab. This step alone does not remove the underlying revision history from the .docx; that requires the next step. But it removes the visible-on-toggle markup that the next step alone cannot.

3. 3

   Step 3: Run Document Inspector (Microsoft's tool, in Word)

   File → Info → Check for Issues → Inspect Document. Tick 'Document Properties and Personal Information', 'Comments, Revisions, and Versions', and 'Headers, Footers, and Watermarks'. Click Inspect. For each item that returns results, click Remove All. This removes the Author, Company, and revision history from docProps/core.xml and docProps/app.xml, the residual w:ins / w:del / w:commentRangeStart / w:commentRangeEnd markers from word/document.xml, the comments from word/comments.xml, and the co-author entries from word/people.xml. This is Microsoft's tool. FileHop does not do this step.

4. 4

   Step 4: Convert to PDF locally — in the FileHop desktop app

   Open FileHop → Convert → DOCX to PDF. Drop the cleaned .docx in. The conversion runs on your machine; the file never uploads. The output is a PDF — but it still carries its own metadata layer (the PDF Info dictionary, XMP) from the conversion engine, so do not skip the next step. Note: 'Print to PDF' in Word produces similar output but the PDF Producer field then advertises the OS and Word version; running the convert step in FileHop gives a more neutral Producer string.

5. 5

   Step 5: Strip the PDF metadata — in FileHop

   FileHop → Compress PDF → enable 'Remove metadata'. Run the operation. FileHop removes the Info dictionary (Author, Title, Subject, Keywords, Creator, Producer, CreationDate, ModDate) and the XMP metadata stream from the PDF catalog. The compress operation also re-encodes embedded images at the chosen quality, which incidentally removes any per-image EXIF that survived inside the PDF — if your figures are embedded inside the PDF rather than uploaded separately, this is the step that handles them. If your figures upload separately (most journals require them as separate files), run Step 6.

   → Open the PDF Compressor
6. 6

   Step 6: Batch-strip EXIF from figure files — in FileHop

   FileHop → Images → Remove Metadata. Drop the folder of figure files in. FileHop strips EXIF/XMP/IPTC from JPEG, removes non-essential chunks from PNG, re-encodes WebP and HEIC/HEIF without metadata, and processes TIFF. Output a clean folder. Upload these to the submission system, not the originals.

7. 7

   The irreducible human step

   Step 7: Manual self-reference review (no tool can do this for you)

   This is the irreducible human step — the textual de-anonymization channel exists regardless of metadata hygiene. The 2022 'Cracking Double-Blind Review' paper (arXiv:2211.07467) demonstrated about 73% authorship attribution from text content alone, using citation patterns and writing style. Search the manuscript text for: your own initials (especially in the methods section — 'analyzed by JS' is a leak); any reference to your own work as 'in press' (rephrase to 'a related study'); any first-person reference to your prior publications ('we previously showed' → 'previous work by the authors showed'); any institutional identifiers in figure captions, scanner labels in radiographs, screen text in screenshots, or watermarks in supplementary material; any acknowledgements or grant numbers (move to the cover letter). No software can interpret these patterns for you.

Verify in 60 seconds (works with any tool, not just FileHop)

A short verification routine catches the cases where Step 5 didn't run, where the wrong file got uploaded, or where a figure copy slipped past the EXIF strip. The checks are tool-agnostic — they work with any anonymization workflow.

1. 1 Open the final PDF in Preview (Mac) or right-click → Properties → Details (Windows). Confirm the Author, Title, and Subject fields are empty. The Producer field may still show a PDF library name (lopdf, PDFKit) — that is fine; it does not identify you.
2. 2 In a terminal, run pdfinfo MyManuscript_anon.pdf (Mac/Linux with poppler-utils installed) or open the file in any PDF reader's Document Properties view. Confirm /Author, /Title, /Subject, /Keywords, /Creator are empty or missing.
3. 3 Open one of the separately-uploaded figure files. On Mac: File → Get Info → More Info → confirm no Camera, Lens, GPS, or Software fields. On Windows: right-click → Properties → Details → confirm Camera, GPS, and Author rows are empty.
4. 4 Search the manuscript text (Cmd+F / Ctrl+F) for your last name, your first name, your initials, your institution's name, your institution's acronym, any grant number, and the word 'we'. The 'we' search is the most useful — it surfaces every place where you wrote in first person about your own prior work.
5. 5 Open the PDF in a PDF reader, copy-paste a paragraph into a plain-text editor, and confirm the copied text matches what is on the page (no surprise residue from tracked changes that flattened into the page during a bad export).
6. 6 Have a co-author or a labmate cold-read the manuscript for 5 minutes and ask: 'who do you think wrote this?' If they can guess from the citation pattern or the writing tics, the textual channel is leaking and Step 7 of Workflow A needs another pass.

Frequently asked questions