How to preserve online video evidence for OSINT and investigations
Key takeaways
- The Berkeley Protocol on Digital Open Source Investigations, published by the UN Human Rights Office (OHCHR) and the UC Berkeley Human Rights Center, sets minimum standards for collecting, preserving and verifying online information.
- Preserving a video means keeping the file in its best available original form, its platform metadata, the surrounding web page and a record of who collected it, when and how.
- A cryptographic hash such as SHA-256, computed right after capture, lets anyone later prove that the file has not been altered.
- yt-dlp can save the platform's metadata alongside the video with --write-info-json, --write-description and --write-thumbnail.
- Archiving the page in the Wayback Machine with Save Page Now creates an independent, timestamped third-party copy of the context.

Preserving a video as evidence means saving four things, not one: the file in its best available form, the platform metadata, the surrounding page, and a written record of how you collected them, all fixed with a cryptographic hash. That is the logic of the Berkeley Protocol on Digital Open Source Investigations, the reference framework published by the UN Human Rights Office (OHCHR) and the UC Berkeley Human Rights Center. This guide turns it into a concrete workflow with open-source tools.
It is not legal advice. If your material may end up in court, follow your organisation’s procedures and talk to counsel early.
Why is online video so fragile as evidence?
Online video disappears and changes. Uploaders delete posts, accounts are suspended, platforms remove content for policy reasons, and the same clip circulates in re-encoded, cropped or re-captioned versions. Researchers who want to rely on a video later must show three things:
- Authenticity: the file is what it claims to be.
- Integrity: it has not changed since collection.
- Provenance: where, when and by whom it was collected.
The Berkeley Protocol groups this work into identification, collection, preservation, verification and analysis. Preservation is the step where evidence is most often lost.
What should a preserved video package contain?
| Item | Why it matters | How to get it |
|---|---|---|
| Video file (best available) | The content itself | yt-dlp, or the platform’s own download |
| Platform metadata (JSON) | Upload date, uploader, IDs, view counts at capture time | yt-dlp --write-info-json |
| Description, thumbnail, comments | Context that changes or disappears | yt-dlp options below |
| Page capture | How the post appeared to viewers | Wayback Machine, ArchiveBox, screenshot |
| Technical metadata | Codecs, duration, encoder | MediaInfo, ExifTool |
| Hash values | Proof of integrity | sha256sum or Get-FileHash |
| Collection log | Provenance and chain of custody | A simple text or spreadsheet log |
Step 1: how do I capture the video and its metadata?
yt-dlp can save the platform’s metadata next to the video in one pass:
yt-dlp --write-info-json --write-description --write-thumbnail \
--write-comments --write-subs --write-auto-subs \
--no-overwrites --no-mtime \
-o "%(extractor)s_%(id)s/%(extractor)s_%(id)s.%(ext)s" \
"URL"
--write-info-jsonsaves the complete metadata yt-dlp received (uploader, upload date, original URL, formats).--write-description,--write-thumbnail,--write-commentskeep context that the uploader can later edit or delete.--no-overwritesprotects files already collected.--no-mtimekeeps the file’s modification date as the moment you collected it, rather than a date taken from the server.- Naming by platform and ID avoids misleading titles and makes each capture unique.
Prefer the best quality without re-encoding: do not convert formats at this stage. If you need an MP4 for viewing, create a separate working copy later with FFmpeg, and keep the original untouched.
Run yt-dlp --version and record it in your log: a tool version is part of how the capture was made.
Step 2: how do I hash the files?
Compute a SHA-256 hash for every file immediately after capture.
On Linux or macOS:
cd youtube_abc123
sha256sum * > SHA256SUMS.txt
On Windows PowerShell (Get-FileHash uses SHA-256 by default):
Get-ChildItem -File | Get-FileHash -Algorithm SHA256 |
Select-Object Hash, Path | Export-Csv SHA256SUMS.csv -NoTypeInformation
Copy the hash values into your collection log too. Later, sha256sum -c SHA256SUMS.txt verifies that nothing has changed. A hash proves integrity from the moment it was computed, not before, which is why it must be done at once.
Step 3: how do I capture the page context?
The video alone does not show the caption, the account, the date displayed or the replies. Capture the page in at least one independent way:
- Wayback Machine Save Page Now. Go to
web.archive.org/save, paste the public URL and submit. When you are signed in to an archive.org account, extra options appear, such as saving the page’s outlinks and a screenshot; the Internet Archive help page describes the current behaviour. The result is a timestamped copy held by a third party, which is valuable precisely because you do not control it. - ArchiveBox, an open-source self-hosted archiver, saves pages in several formats (HTML, screenshot, PDF, WARC) on your own machine.
- Full-page screenshot with the browser’s own tool, with the system clock visible or the time noted in the log.
Think before using a public archive in sensitive cases: archiving can be visible, and some content shows people at risk. The Berkeley Protocol stresses security and the duty of care towards people depicted.
Step 4: how do I record technical metadata?
Extract technical information from the file with MediaInfo or ExifTool and save it as text:
mediainfo --Output=JSON youtube_abc123.mp4 > youtube_abc123.mediainfo.json
exiftool -json youtube_abc123.mp4 > youtube_abc123.exiftool.json
Expect platform files to contain little original camera data: platforms re-encode uploads and strip most metadata. That is normal and should be noted rather than “fixed”. Codec details are explained in our codecs and containers guides.
Step 5: how do I keep a chain of custody log?
A log can be a spreadsheet. One row per capture:
| Field | Example |
|---|---|
| Capture ID | 2026-11-18-001 |
| Original URL | https://platform.example/watch/abc123 |
| Collected by | Analyst initials |
| Date and time (UTC) | 2026-11-18 14:05 |
| Method and tool version | yt-dlp (version recorded), Save Page Now |
| Files and SHA-256 | video, info.json, page capture |
| Archive URL | Wayback Machine link |
| Notes | Account suspended next day; comments disabled |
Then store the package read-only, with at least two copies on separate media. Never edit the originals: analysis, clipping and enhancement happen on working copies, each with its own hash.
How do I verify a preserved package later?
Preservation is only useful if anyone can check, months later, that nothing changed. Build verification into the routine rather than doing it once:
- Re-hash on a schedule. Run the same SHA-256 command monthly or before any hand-over and compare with the value in your log. A single changed byte produces a completely different hash.
- Verify after every copy. When files move to a new drive or to a colleague, hash them at the destination before deleting anything at the source.
- Keep the log append-only. Never edit an old line of the custody log; add a new dated line that explains the correction.
- Test that the file still plays. Open the video in a second player such as VLC and note the duration shown, which should match the
durationfield of the info JSON.
| Problem found | Likely cause | What to do |
|---|---|---|
| Hash mismatch | File re-saved, re-encoded or corrupted | Restore from another copy and log the incident |
| Info JSON missing | Captured with a tool that ignores metadata | Re-capture if the source is still online, otherwise note the gap |
| Duration differs from the page | Live stream captured partially | Record start and end times in the log |
| Page now deleted | Normal platform churn | Rely on your archived capture and the third-party Wayback copy |
A simple naming scheme also prevents mistakes: YYYYMMDD_platform_videoID_caseNN, used identically for the video, the info JSON, the screenshot and the hash file, keeps every item of a package together when sorted.
What legal and ethical limits apply?
Collecting public content for investigation is common practice, but it does not suspend other rules:
- Copyright still applies to the video; exceptions for research, reporting or legal proceedings vary by country (see is downloading videos legal).
- Data protection applies to identifiable people in the EU and many other places.
- Platform terms may restrict automated collection.
- Safety: avoid exposing victims, witnesses or yourself. The Berkeley Protocol includes a dedicated security chapter.
For long-term storage habits, the method in our channel backup guide (3-2-1 copies, periodic checks) applies equally to evidence archives.
Open-source tools in this guide
| Tool | Licence |
|---|---|
| yt-dlp | Unlicense (public domain) |
| MediaInfo | BSD 2-Clause |
| ExifTool | Same terms as Perl (Artistic / GPL) |
| ArchiveBox | MIT |
Frequently asked questions
Is a screen recording good enough as evidence?
It is a useful complement, because it shows the page as a person saw it, but it is a re-encoded copy. Investigators generally try to collect the original file served by the platform as well, with its metadata, and to hash both.
Why compute a hash right away?
A SHA-256 hash is a fingerprint of the exact bytes of the file. Computed at capture time and recorded in your log, it lets you or a third party later demonstrate that the file presented is identical to the file collected.
Does the platform keep the original upload?
Platforms usually re-encode uploads and strip much of the original camera metadata. What you can collect is the platform's version plus the platform's own metadata (upload date, uploader, IDs), which is why both should be preserved.
Should I archive the page in the Wayback Machine?
Usually yes, for public pages. Save Page Now creates an independent timestamped copy held by a third party. Think first about the safety of people shown in the content and about alerting the uploader, especially in sensitive investigations.
Is it legal to download videos for an investigation?
It depends on your jurisdiction, your role and the content. Some laws have exceptions for research, journalism or legal proceedings, while copyright and data protection still apply. Organisations doing this work set legal and ethical policies first.
Sources and open data
- Berkeley Protocol on Digital Open Source Investigations (UN digital library record) · OHCHR and UC Berkeley Human Rights Center
- Berkeley Protocol project page · UC Berkeley Human Rights Center
- Save Pages in the Wayback Machine · Internet Archive
- Get-FileHash (PowerShell) · Microsoft Learn
- yt-dlp README (filesystem and metadata options) · yt-dlp project
Related guides
- yt-dlp guide
A practical yt-dlp guide: install and update it safely, pick formats with -f and -S, name files, grab subtitles and keep a download archive.
- Internet Archive
How to download videos from the Internet Archive legally: find public-domain films, read the licence field, and bulk-download with the official ia CLI.
- FFmpeg for video files
FFmpeg commands for downloaded video: merge separate audio and video, remux TS to MP4 without quality loss, cut clips, join parts and convert only when needed.
- Is downloading legal?
When downloading a video is lawful and when it is not: copyright, fair use, private copy, platform terms and DRM, with the primary legal texts.
- Back up your channel
Back up every video you uploaded: Google Takeout for bulk export, YouTube Studio downloads, metadata, Brand Account pitfalls and a 3-2-1 storage plan.