Defenex
← Blog

The evidence file: what to capture before you send any takedown

Enforce · 4 October 2026 · 8 min

An open case folder holding a screenshot, a clock, a fingerprint pattern and a sealed envelope.

The most common mistake in brand enforcement happens before the notice is sent: filing it without first recording what you're complaining about.

It seems like a small thing. The listing is right there; anyone can click the link. But infringing pages don't stay put. The seller edits the title once the complaint arrives. The listing is removed, then relisted under another account. A phishing page shows harmless content to anyone who looks like an investigator. If the platform asks for more detail, or you need to escalate, or you want to show that this seller has done it four times, a link that now returns a 404 proves nothing.

An evidence file fixes that. It's a record of each finding as it was when you found it, which you can rely on later. This article covers what goes in it and how to keep it in order.

This is a practical guide, not legal advice. If a matter may go to litigation, ask counsel how evidence should be collected and preserved for your jurisdiction.

What to capture for every finding

1. The full URL, exactly

Copy the URL from the address bar, not from a search result, because search results often go through redirects. Keep any listing ID or item number separately as well; marketplaces identify listings by ID, and the URL format can change.

2. A full-page screenshot

Capture the whole page, not just the visible screen: title, images, price, seller name, description and any "ships from" details. Most browsers can take a full-page screenshot natively or with an extension. Make sure your system clock is correct and that the date and time are recorded, either visible in the capture or in the file metadata you store with it.

3. The page itself

A screenshot shows what you saw. A saved copy of the page (HTML, or a web archive file) shows what the page contained, including text you might not have noticed. Save both.

4. A third-party archive copy

Your own screenshot is your word. A copy held by an independent archive is harder to dispute. Public web archives such as the Internet Archive's Wayback Machine let you request a snapshot of a page. It won't work for every page: some sites block archiving, and pages behind a login can't be captured. When it does work, it gives you a timestamped record that you didn't make yourself.

5. The seller or operator

Record who's behind the page, as far as you can tell:

6. What makes it infringing

This is the part most often missing. Write one or two sentences saying what the page does and quoting the specific text or describing the specific image that shows it.

"Listing title reads 'Replica [Brand] Cooler Wholesale'. Main image is our product photo from our product page (link), unchanged."

Use the page's own words. "This is a counterfeit" is an assertion. A quote from the page saying "replica" or "1:1 copy", next to your trademark, is evidence. Stay factual: describe what the page shows rather than concluding that it's illegal.

7. Your rights

Record which right the finding infringes and attach the proof, once, in a place every finding can point to:

Which right you rely on decides which notice you file. See DMCA or trademark complaint?.

8. Test purchases, if you make them

For high-stakes cases, a test purchase can establish that goods are fake in a way a listing can't. If you make one, record the order confirmation, payment record, shipping label, and photos of the package and product as received. Ask counsel first if the case may go to court, because how a test purchase is made and documented can matter.

Make the file trustworthy

Evidence is only useful if someone else can trust it later. A few habits help a lot:

The log that ties it together

Each finding gets one row in a log. The captures are the evidence; the log is what lets you use it.

Field Why it matters
Finding ID Your own stable reference across relistings
URL / listing ID What you're reporting
Platform / host Which enforcement channel applies
Seller / operator ID Links repeat infringers together
Category Counterfeit, lookalike domain, phishing, impersonation…
First seen / last seen Shows how long it has been up, and whether it came back
Evidence files + hashes Points to the captures
Right relied on Trademark reg. no. or copyrighted work
Notice filed: date, channel, report ID Your record of action
Outcome and date confirmed Removed, rejected, no response, relisted

The seller ID and first seen / last seen columns turn a pile of single complaints into a repeat-infringer case. When the same seller ID appears in five rows, the next notice can say so and cite the earlier report IDs. That's the argument that gets an account closed rather than one listing removed. Relisting is the normal pattern (see why counterfeiters relist), so the log will probably be the most useful thing you build.

Check removals instead of assuming them

When a platform says a listing was removed, open the URL yourself and record the result. Sometimes "removed" means hidden in one country, or the listing has already been replaced by an identical copy. Record the date you confirmed the removal. That date is how you know the notice worked, and it's what you'll compare against if the listing comes back.

How long it takes

Done by hand, a full capture for one finding takes a few minutes once you have a routine, and longer the first few times. That's manageable for a handful of findings a week. With dozens, the capture work starts crowding out the decision work: whether a finding is real and which channel it belongs in. That's usually when people start automating capture, or hand the work to a service that keeps an evidence record as part of every finding.

For the next step, filing the notice, see reporting counterfeits on Amazon, eBay, Etsy and AliExpress.


Start with findings that already have evidence. A free Defenex scan searches the open web for counterfeits, lookalike domains, phishing and impersonation of your brand, and quotes the text from each page that supports the finding. No account; about 90 seconds.