Brand teams usually discover counterfeits the slow way: a customer complains about a product they did not buy from you, and an investigation follows. By then the listing has been selling for months. The reason it went unnoticed is rarely that nobody searched. It is that the searching was done from one country, in one language, on a handful of marketplaces, while the listings live somewhere else.
Counterfeit detection is a coverage problem before it is a matching problem. Here is how to build the coverage, and how to turn what you find into something a marketplace will act on.
Why counterfeits are invisible from head office
Three mechanics keep listings out of view.
Marketplace results are localised. What appears for a search term depends on the storefront you are in, and often on your inferred location within it. A listing that dominates results in one country may not surface at all in another, so a search from your office measures one market and tells you nothing about the rest.
Sellers restrict where they ship or display. A listing configured for one region can be effectively invisible elsewhere, which is convenient for a counterfeiter who wants distance between the sale and the brand’s home market.
Language fragments the search. Counterfeit listings are written for local buyers, in local language, often with deliberate misspellings of your brand and with model numbers rendered differently. An English-language search for the exact brand name finds the honest listings and misses the ones designed not to be found.
The consequence is that coverage has to be built deliberately across markets and languages, rather than assumed.
Build the coverage matrix first
Before writing any detection logic, decide what you are searching and where. The matrix has four axes.
Marketplaces, including the large global platforms plus the regional ones that matter in your categories, and the social commerce surfaces where a growing share of this activity happens.
Markets, meaning the country storefronts you care about, which should include the places you sell and the places counterfeits of your products typically originate or transit.
Search terms, which is where most programmes are too narrow. Include your brand and product names, common misspellings and transliterations, model and SKU numbers, and the descriptive phrases a buyer would use who does not know your brand name. Add local-language variants for each market rather than translating mechanically, since sellers use the phrasing buyers actually type.
Cadence, weighted by risk. New product launches and high-value lines deserve frequent sweeps; long-tail catalogue items can be checked monthly.
Be explicit about what the matrix does not cover, because a sample presented as a census creates false confidence in a legal context.
Search as a local buyer
This is the part that requires infrastructure rather than process. To see a market’s results, the request has to come from that market, which means residential exits in each country you monitor, with the language and locale set to match.
The reason a datacenter connection is not sufficient here is the same as elsewhere in brand protection monitoring: marketplaces treat traffic from hosting ranges differently, so you are liable to get a generic or degraded result set rather than the one a local shopper sees. Where a marketplace localises below country level, city-level targeting matters too, and the language, timezone and locale you present should agree with the exit country, per matching geo, timezone and locale.
Practically: one sweep configuration per market, with the country, language and search terms bound together so they cannot drift apart.
Signals that separate a counterfeit from a legitimate reseller
Not every third-party listing is a counterfeit, and treating them as equivalent produces reports that get ignored. Score listings rather than flagging them binary.
Price. A price well below your authorised floor is the strongest single signal, though a steep discount alone can also be grey-market or clearance stock.
Imagery. Reused official photography is common, and so are photos with subtle differences in packaging, logo placement, or included accessories. Image comparison against your own asset library does a lot of work here.
Listing text. Machine-translated descriptions, missing or invented model numbers, incorrect specifications, and trademark usage that no authorised partner would write.
Seller profile. Recently created accounts, a catalogue spanning unrelated premium brands, a location inconsistent with your distribution, and rapid listing churn.
Packaging and variant claims. Colours, bundles or editions you never produced are a definitive tell, and they are the easiest thing for a reviewer to confirm.
Combine these into a score and route only the high-confidence findings to human review. The goal is a queue an analyst can work through, not an alert stream nobody reads.
Capture evidence at the moment you find it
Marketplace listings are edited and removed, so a finding without a capture is a memory. Every flagged listing should be recorded with a full-page screenshot, the listing URL and marketplace item identifier, the seller name and profile, the price and currency, the shipping origin and destinations offered, the images as files rather than as links, and a timestamp with the market the check ran from.
That last field matters more than it appears: if enforcement or litigation follows, the fact that a listing was visible to buyers in a specific country on a specific date is often the point in dispute, and it is only demonstrable if you recorded the vantage point.
Keep the raw captures alongside the derived verdict, and set retention to cover the enforcement cycle without holding personal data longer than your policy allows.
From detection to enforcement
Detection only pays off if it feeds a process. Route findings into tiers: obvious counterfeits with strong evidence go straight to marketplace takedown programmes, borderline cases go to human review, and patterns involving the same seller across marketplaces go to your legal team, since a repeat operator is worth a coordinated response rather than a series of individual notices.
Track outcomes rather than volume. The numbers that justify the programme are takedown success rate, time from listing appearance to removal, and repeat-offender recurrence, not how many listings you flagged. And feed the outcomes back into the scoring, because the categories of listing that get removed successfully tell you what your evidence standard should look like.
Related enforcement work runs on the same collection layer: unauthorised sellers and MAP violations and trademark abuse detection.
Keep the collection clean
A few operational points that determine whether the programme keeps working.
Pace politely and spread sweeps across the available window, since marketplaces are among the more defended targets and an aggressive sweep gets throttled, per rate limiting and throttling. Validate that what came back is a real result page rather than a challenge or an empty set, because a silent failure in one market reads as an absence of counterfeits, which is the worst possible false negative here, per detecting blocked or fake content. And monitor coverage per market so you notice when a country stops returning results.
Stay on public listing data, respect each marketplace’s terms, and do not make test purchases as part of automated monitoring; buying a sample is a deliberate legal step your team takes knowingly, not something a crawler does.
The bottom line
Counterfeits hide behind geography and language more than behind clever wording, so detection starts with a coverage matrix across marketplaces, markets, and local-language search terms rather than with better matching. Search each market from inside it, because localised results are invisible from anywhere else. Score listings on price, imagery, text, seller profile and variant claims instead of flagging them binary, so analysts get a workable queue. Capture full evidence at the moment of discovery, including the market you searched from, since listings disappear and vantage point is often the disputed fact. Then route findings into tiers with an enforcement path, and measure takedown outcomes rather than flag counts.
The market coverage underneath is what brand protection monitoring needs, running on residential proxies with country and city targeting so each sweep sees what local buyers see, billed per GB so cadence stays a programme decision.