How Copycat Listings Leak Into AI Answers About Your Brand

"Ask it who sells them," said Tomás.

His co-founder typed it in. Where can I buy Arbo field knives?

The answer came back in four lines, pleasant and confident, and it named three places. Their own shop was second. First was a marketplace shop called ArboOutdoor Official, which Tomás had never heard of, and which had used the word Official in its name, which was the part that made him put his coffee down.

They'd been trading for 6 years. The copy had been trading for 5 months.

An answer is assembled, not retrieved

The useful mental model here isn't "the AI has an opinion about your brand". It's closer to a researcher with 20 minutes and a browser.

Asked where to buy a thing, an assistant goes and finds pages that look like they sell that thing, reads them, and summarises what it found. It reads your site. It reads marketplace listings. It reads price-comparison pages, directory entries, Reddit threads, and the occasional blog post from 2023. Then it writes a short paragraph that reconciles all of it.

Which means the answer is only as good as the pages it found, and the pages it found include every listing that claims to sell your product. Your competitors don't have to beat you. They only have to be in the pile.

A copy gets into the pile cheaply. It has your photographs, so it looks right. It has your product name in the title, so it matches the query. It's on a marketplace with far more authority than your own domain, so it gets crawled hard and indexed fast. From the assistant's side of the glass, there's nothing to distinguish it except price, and on price it wins.

The three ways a copy ends up in the answer

Directly, as a retrieved listing. The marketplace listing is a well-indexed page that mentions your brand and offers a product. Nothing exotic is happening. The copy is simply in the index and the assistant read it.

Second-hand, through an aggregator. Price-comparison sites, deal sites and "best X" roundups pull marketplace feeds automatically and publish pages that list sellers side by side. Your product appears with four sellers, one of which isn't you. The aggregator page then gets cited as a source, and now your copy has a reference from a site that looks neutral.

Through conversation. Someone posts in a forum asking whether the cheap ones are any good. Three people answer. Two of them link the copy. That thread outranks most brand pages for exactly the query a cautious buyer types, and it reads as authentic, because it is.

The third one is the hardest to touch and the easiest to underestimate. You can take a listing down. You can't take down a conversation about it, and you shouldn't try.

Why removal alone doesn't fix the answer

This is the bit that surprises people who've already done the legal work.

Say Tomás gets ArboOutdoor Official removed. Good. The listing 404s. What has the assistant learned?

Nothing, for a while. Retrieval happens against whatever is in the index at the time of the question, and index refresh is not instant. More importantly, the aggregator pages that mentioned the copy are still there. The forum thread is still there. The pattern the assistant picked up, that Arbo knives come from several sellers and the cheap ones exist, survives the listing that taught it.

Takedowns remove the sale. They don't rewrite the record. For the answer to change, the record has to change, and the only part of the record you control is your own.

Making the official store unmissable to a machine

Here's the unglamorous half of the job, and the half brands skip because it has no satisfying click.

Say which stores are yours, in one place, in public. A page on your own domain that lists every official store and authorised retailer, by name, with links. Not a PDF. Not a line in your FAQ. A page with a stable URL that you keep current. This is the single most useful artefact in the whole exercise, because it is the thing an assistant can read and quote when a buyer asks whether a shop is legitimate.

Make the product names consistent everywhere. If your site says "Arbo Field Knife No. 2", your marketplace listing says "Arbo Field Knife 2", and your packaging says "Field Knife II", you've given the machine three products. A copycat using one exact name consistently looks more coherent than you do. Pick the string and use it as written.

Put the structured data in. Product with brand, sku, and an offer carrying the seller, the price and the currency. Organization on your site with sameAs pointing at your real profiles. This isn't decoration. It's the difference between an assistant inferring who you are and being told. Entity hygiene goes through the exact properties and the order to do them in.

Answer the comparison question yourself. Buyers ask whether the cheap ones are fakes. If you've never published anything on that, the only available answers come from strangers. A page that states what an authentic one costs, what the giveaways are, and where to buy it is content an assistant will happily cite, because it's specific and it's from the horse's mouth.

Keep the reviews flowing on your own surfaces. Volume of genuine, recent, attributed review text on pages you control is one of the few signals that's both hard to fake and easy for a retrieval system to read.

None of that is quick. All of it compounds, which is the opposite of a takedown, and it's why I'd argue the entity work is the better investment if you can only do one. Takedowns are maintenance. This is construction.

You still need the takedowns

The reason not to choose is that the two halves feed each other.

Every copy you remove is a page that stops being retrievable. Every copy you don't remove keeps generating aggregator entries and forum mentions, which keep generating citations. Leave them alone for a year and you're not fighting one listing, you're fighting the sediment it left.

So the cadence is both. Find and remove the listings on a schedule, and build the official-store record in parallel. The removal stops the bleeding, the record changes what the machine believes.

Finding them on a schedule is the mechanical part. Brand Protection Tools, which Verity AI runs, searches on a weekly, 3-day or daily cadence depending on the plan, scores each listing on photo similarity, title match and price distance, groups them by seller, and drafts the notice for the brand owner to review and file. Etsy and Alibaba.com are what it searches today. Everything else on the coverage list, including eBay and Amazon, is next or planned rather than live.

The allowlist matters more than it sounds. Put your own shops and your authorised sellers in it and the report stops being 80% your own footprint. That same list, published on your site, is the artefact in the section above. Write it once, use it twice.

What Tomás did

He filed on the shop name first, not the listings, because the word Official in a shop name is a trademark problem rather than a copyright one and the platforms treat it more seriously. DMCA or trademark? is why that distinction decided which form he filled in.

Then he built the retailer page. It took an afternoon and it's 300 words long. It lists 2 official stores and 9 stockists, names the 3 countries they ship from, and says plainly that Arbo knives are never sold below 60 euros by anyone authorised.

11 weeks later, asked the same question, the assistant put their own shop first and quoted the retailer page as the reason. It also mentioned that cheaper listings exist and are not authorised, which is a sentence it got from Tomás, because he was the only one who'd written it down.

The copy is still trading, under a different name. He knows. He checks.

For the wider picture of why this became a visibility problem rather than a legal one, start with When the AI shopping assistant recommends the fake. For finding the copies in the first place, your product photos are the way in.