Written by: Michał Sobieraj, Operations Manager, Digital Colliers
In April 2025, three websites published 215,128 'best software' comparison pages between them. The pages barely qualified as content. They existed to be cited by AI answer engines, and Perplexity cited them. The investigation by trellner.com found a pattern most brands still do not monitor: manufactured sources designed to steer LLM-driven discovery. Three sites flood the retrieval pool with comparison pages, the engine cites them as authoritative, and buyers see footnotes that look like research but were engineered to rank.
Most eCommerce brands cannot tell you which AI engines mention them, what claims those engines cite, or when a competitor appears in an answer where they do not. The monitoring gap is not technical. It is organizational. More than 80% of AI projects fail, and teams that scope GenAI pilots tend to aim for transformation instead of the unglamorous pipeline work a brand actually needs.
The manufactured-source problem
When someone asks Perplexity or ChatGPT Search which platform does X best, the engine cites whatever pages its retrieval system surfaces. If those pages are SEO spam or competitor-funded comparison sites, the citation looks authoritative anyway. The reader sees a footnote link. The engine presents it as evidence. Your brand might not appear, or it appears with a claim you never made.
The trellner.com investigation documented 215,128 pages from three sites alone. The scale matters because it shows how cheap it is to flood the citation pool. A competitor or spam operator can publish thousands of comparison pages in a weekend. Each page targets a different query. The engines retrieve them because they match the semantic intent of the prompt, and the average buyer cannot tell the difference between a researched comparison and a manufactured one.
You are competing for citations in a retrieval race where the rules changed six months ago and most brands have not noticed yet.
What a monitoring pipeline looks like
You do not need a research moonshot. You need scheduled prompts.
Pick 10-20 questions a buyer in your category would ask: 'best X for Y', 'which platform handles Z', 'alternatives to [competitor name]'. Run those prompts against Perplexity, ChatGPT Search, Gemini, and Claude every week. Capture the full response text and the cited source URLs.
Log everything to a database. Build a weekly diff report: which brands appeared in each answer, which claims were cited, which sources the engine linked. Alert when a competitor's site appears for a prompt where you appeared last week. Alert when a claim about your product changes. Alert when a manufactured comparison page starts ranking for your category.
The operators shipping this in 2026 are running it off a single cron job and a Postgres table. The scope is small because the value comes from consistency, not sophistication. You need the record of what each engine said, week over week, so you can spot the drift before it costs you deals.
If your team runs SQL by hand for the audit, that is fine. The goal is not an autonomous GenAI product. The goal is a weekly email that shows you which prompts your brand owns, which ones you lost, and which competitors are cited in your place.
The window is closing
Customer acquisition cost across DTC brands has risen roughly 40% since 2023. UK eCommerce grew 3% in 2024 vs 2023. The math is tight. If 5% of your inbound research now starts with an AI engine instead of Google, and those engines cite a competitor for the prompts you should own, you lose pipeline you cannot afford to lose.
The brands that build citation monitoring in the next six months will know which claims need correction, which sources to feed the engines, and where competitors are winning prompts they thought they owned. The brands that wait will keep losing deals to citation patterns they never saw.
88% of AI proof-of-concepts never reach widescale deployment because teams scope for transformation. Citation monitoring is not a transformation project. It is a data pipeline that tells you what the engines say about you, every week, before it costs you deals. Build it before your competitors notice you are not in the answers.