Publishers and ecommerce brands are now planting fake product listings and content traps to catch AI crawlers scraping their sites for training data, according to Digiday. The technique—called LLM honeypotting—borrows from decades-old cybersecurity playbooks and repurposes it for the generative AI era. Brands hide synthetic listings invisible to human shoppers but irresistible to bots, then monitor where that fake data surfaces to identify which AI models ingested it.

The mechanics are straightforward. A brand publishes a product page or article snippet designed to look legitimate to a scraper but tagged in a way that makes it traceable. The page sits in the site architecture, crawlable by bots but hidden from customers through CSS or no-index directives. When an AI model later reproduces that invented detail in a chatbot response or generated listing, the brand has proof of unauthorized scraping and a legal foothold to pursue enforcement or demand licensing fees.

This works because large language models cannot easily distinguish between real and synthetic training data once it enters the corpus. A fake SKU, a fabricated product spec, or a planted review phrase becomes embedded in the model's weights. When the model generates text, it may reproduce the honeypot verbatim or in paraphrase, creating a signature the original publisher can detect. The strategy mirrors how photographers watermark images or how academic publishers seed fake citations to catch plagiarism, but it scales across product catalogs and editorial archives.

For a small physical-product brand, the play is accessible and does not require a security team. Start by creating three to five fake product listings on your Shopify or WooCommerce site. Give each a unique SKU that does not exist in your inventory system and a fabricated feature—unusual dimensions, a nonexistent colorway, a proprietary material name you coined. Set those pages to noindex in your SEO plugin so Google does not surface them to shoppers, but leave them crawlable in your robots.txt so bots can reach them. Publish the pages in a subfolder that mirrors your real catalog structure.

Next, set up a Google Alert or a Talkwalker Free Alert for the exact fake SKU or the fabricated feature phrase. Check monthly whether that language appears in AI-generated product recommendations, Amazon listings, or competitor catalogs. If it does, you have documentation that someone scraped your site and either trained a model on it or resold your data. That documentation supports a DMCA takedown request, a licensing negotiation, or a cease-and-desist letter. The cost is zero beyond thirty minutes of page creation and the discipline to monitor alerts.

For brands with a legal budget, honeypotting also creates leverage in licensing conversations. AI companies increasingly pay publishers for training access—Digiday notes deals with news organizations now routine—but scraping remains widespread among smaller model builders and data brokers. A brand that can prove its proprietary data ended up in an unlicensed model can demand retroactive payment or threaten litigation with hard evidence. The honeypot turns an invisible harm into a documented violation.

The pattern extends beyond product pages. Recipe sites plant fake ingredient lists. Review aggregators seed nonexistent user feedback. Editorial publishers insert fictitious sources or quotes in draft articles visible only to crawlers. Each trap is a tripwire that converts data theft into a provable event, shifting the burden of proof from the victim to the scraper. The tactic does not stop all scraping, but it makes unauthorized use detectable and costly, which changes the economics of AI training at scale.