Removing hidden data before you send a file

The short version

• A file can carry more than what you see on the page: a photo its location, camera and time; a PDF its author and the program that made it; a Word file its author, company, editing time, comments and tracked changes.
• FileBench has three tools that remove such data (Remove image metadata, Edit PDF metadata and Clean Word metadata) and one that reports what a file is (Check file type & info). Each removes a well-defined set of things and leaves others alone. On 15 October 2026 this guide ran the PDF and Word tools on real files, read the EXIF of real photos, and noted what stayed.
• The PDF tool blanks the document's information fields but not the XMP packet most publishing programs add, and the Word tool clears the document properties but not comments or tracked changes. Those are the limits to know before you rely on either.
• Three of the four tools run entirely in your browser. Remove image metadata has one exception, for pictures your browser cannot open, and its page says so.

Three places hidden data lives

No single tool deals with all three kinds.

• Metadata blocks written by the camera or the program that saved the file: the EXIF block of a photo, the information dictionary and XMP packet of a PDF, the document properties of a Word, Excel or PowerPoint file. This is what metadata tools remove.
• Content that happens to be revealing: comments, tracked changes, the author name stored on a PDF sticky note, hidden text, an earlier draft left on page 2. No metadata tool, FileBench's included, touches any of it.
• The file name. “Offer-letter-draft-3-do-not-send.docx” says a lot before anyone opens it, and every FileBench result keeps the original name with a suffix added.

Photos: what a phone writes, and what re-encoding removes

A camera stores its notes in an EXIF block at the start of a JPEG, before the first pixel. This guide read the EXIF of a test photo from FileBench's own fixture set, a 3024 × 4032 JPEG generated for exactly this purpose, so its location is made up. It carries 36 entries in three directories: 14 in the main one, including the make, model, software, date and time, an artist and copyright line and a description; 14 in the camera one, including the exposure time, aperture, ISO, focal length, lens, the original and digitised timestamps and a user comment; and 8 GPS entries that decode to 28° 36' 54" N, 77° 12' 30" E at an altitude of 216 m. All of that fits in 806 bytes. A real photo in the same set, taken on an iPad and published by NASA, carries 35 entries, among them the device model, the Lightroom version that edited it, the time with its UTC offset and the exact lens; it has no GPS block.

Remove image metadata does not edit that block. It decodes the picture, draws it onto a canvas (an in-memory drawing surface) and asks the browser to encode a new file from the pixels. The encoder is handed pixels and nothing else, so the new file carries no EXIF, no XMP, no colour profile and no embedded thumbnail from the original (a statement about the code, not a measurement: a canvas cannot pass them on). A PNG comes back as a PNG; everything else (JPEG, WebP, GIF, BMP, or a TIFF or HEIC your browser can open) comes back as a JPEG at quality 94 of 100, named with -clean added. The orientation stored in the EXIF is applied to the pixels when the picture is decoded, so a photo stays the right way up once the tag is gone; the result says so.

Re-encoding has a cost. A JPEG is decompressed and compressed again, so its pixels change slightly and its size changes. Transparency in a PNG is kept, but a transparent WebP or GIF becomes a JPEG on a white background, and an animated picture keeps its first frame only, and the tool says so.

There is one case in which this tool sends a picture to FileBench's server: a file your browser cannot decode itself, most often a TIFF. The browser tries first; only when that fails, and only if the file's first bytes show a format the server can read, is it sent to /api/decode, turned into a PNG and sent straight back, after which the stripping happens in your browser as usual. An iPhone HEIC photo your browser cannot open is refused with a message, not sent, because the server cannot read it either. If the photo must not leave your device, give the tool a JPEG, PNG or WebP, and check the Network tab as described at the end.

PDFs: the information dictionary is only one of two places

A PDF describes itself in two places. The information dictionary holds the eight standard fields (title, author, subject, keywords, creator, producer, creation date, modification date) and any extra keys the producing program adds. The XMP packet, an XML block that Acrobat, InDesign and many other programs write as well, repeats those fields and adds more, such as document identifiers and a history of edits.

Edit PDF metadata works on the information dictionary. Set values writes the title, author, subject and keywords you type, leaves the creator and producer as they were, and stamps the modification date with the moment you ran it. Remove standard properties sets the eight standard fields to empty text (the entries stay, with nothing in them), and if you also tick Also clear creation and modification dates, both dates become 1 January 1970 (UTC), a placeholder date rather than no date at all. Measured on a 1,395-byte PDF created for this guide with a title, author, subject, two keywords, creator, producer and both dates: after Remove standard properties with the dates cleared, all eight fields read as empty, and the author's name, the producer's name and the keyword “confidential” no longer appear anywhere in the file's bytes. The result was 956 bytes.

A real document shows the limit. NIST Special Publication 800-63-3, a 76-page, 1.6 MB PDF made with Acrobat PDFMaker from Word, was run through Remove standard properties with the dates cleared. The eight standard fields came out blank. Two things did not change. Nine extra keys in the information dictionary stayed exactly as they were, among them Company (“Information Technology Laboratory, NIST”), a DOI and a publication date. And the XMP packet, 6,517 bytes, was untouched: it still names the three authors and the title, the program (“Acrobat PDFMaker 17 for Word”), the producer, the creation and modification dates and the document identifier. A CDC immunisation schedule made with InDesign behaved the same way. Any reader or indexer that reads the XMP packet can therefore still show the author of a PDF that this tool has “cleaned”.

So, at the time of writing, Edit PDF metadata clears the standard fields of the information dictionary and nothing else. FileBench has no tool that shows or removes the XMP packet or the extra keys, and the text, pictures, annotations and the author names stored on comments are not touched. For a PDF that came out of a publishing program, treat the tool as a first step, not a guarantee.

Word, Excel and PowerPoint: three parts rewritten, the rest untouched

A .docx, .xlsx or .pptx file is a ZIP archive, and its properties live in three small XML parts inside it. docProps/core.xml holds the title, subject, author, keywords, description, who last modified it, the revision number, the created and modified dates, and a category and status. docProps/app.xml holds the application and its version, the template, the total editing time in minutes, the company and manager, and page and word counts. docProps/custom.xml, when present, holds custom properties: client codes, document-management identifiers and the like.

Clean Word metadata replaces the first part with a minimal one whose title, author and last-modified-by are empty and whose revision number is 1, replaces the second with one holding an empty company and manager, and deletes the third. Measured on a .docx created for this guide with nine core properties and a custom property: afterwards the core part held four entries, all empty except the revision number; the subject, keywords, description, created and modified dates were gone, not blanked, because the whole part was replaced; and the custom property had been removed with its part. On a Word 2007 test document from the Apache POI project, the application part had listed the template, 101 minutes of editing time, the Word version and the company “Mera”; afterwards it listed an empty company and manager, and the file's three custom properties were gone. Excel and PowerPoint files go through the same three parts.

The test document carried one comment and one tracked change (the salary “48,000” deleted, “52,000” inserted), each with an author name and a date. After the tool ran, the comment, both changes, both names and all three dates were still in the file, and so were the three comments from 2008 in the POI document. The tool rewrites the three property parts and nothing else, so the text, headers and footers, hidden text, and pictures together with any EXIF they carry are exactly as they were. Before you send a document, accept or reject every tracked change and delete the comments in Word, Excel or PowerPoint itself; then run the tool for the properties.

What Check file type & info shows, and what it does not

Check file type & info judges a file by its first bytes rather than its name, and reports the detected type, the size, whether the extension matches the content, and the category. For a PDF it adds the page count, whether it is password protected or has printing and copying restricted, whether it has form fields, whether it is a scan without a text layer, and its title; for a picture it adds the dimensions. It does not read EXIF, so it will not show a photo's GPS block before or after cleaning, and it does not read Word properties. Use it to confirm what a file really is and whether a PDF still carries a title; to confirm that a photo's location is gone, open the result's properties in your operating system or in an EXIF viewer.

The file name

The three cleaning tools keep the original name and add a suffix: Q3-board-pack-internal.pdf becomes Q3-board-pack-internal-clean.pdf. Rename the file before you send it if the name itself says something. FileBench's own records of a run never include the name; the privacy page lists every field they do carry.

Where each tool runs

Each tool page carries a processing statement generated from the same data the tools run on. They say:

• Edit PDF metadata, Clean Word metadata and Check file type & info run entirely in your browser. The file is never uploaded, so there is nothing for FileBench to keep.
• Remove image metadata runs in your browser, with one exception: images your browser cannot open itself, such as most TIFF files, are sent to the server for that step and deleted as soon as the result is returned. Everything else stays on your device.

Check it yourself

1. Open one of the tools, then open your browser's developer tools (F12, or Cmd+Option+I on a Mac) and choose the Network tab. Clear the list.
2. Add your file and run the tool.
3. Look at what appears: the page's own code, /api/health, the small /api/jobs and /api/events requests that record the run and count usage (sizes and counts, never the file or its name), and on helixmedia.biz requests to Umami (cookieless analytics) and to Google's advertising servers when a page loads Google AdSense; the privacy page names the domains. For these four tools, no request is anywhere near the size of your file, with the one exception above: a picture your browser could not open goes to /api/decode, and that request is about the size of the picture.
4. Then check the result, not just the request list: open the cleaned file in Check file type & info to confirm its type and, for a PDF, that the title is gone; open its properties in the program you normally use; and for a photo, look at its details in your operating system.