{
  "id": "N03",
  "url": "https://cartforagents.com/guides/sibling-sku-cannibalization",
  "title": "Product cannibalization in AI shopping: did your rewrite win, or just move picks between your own products?",
  "description": "When you sell several similar products (sibling SKUs), a better listing can pull AI picks from your other versions instead of from competitors. How to count it, with a real example where a rewrite cut a sibling pair from 165 picks to 58.",
  "answer": "When you improve one of several similar products, AI assistants may pick it more often partly by picking your other versions less. That's sibling SKU cannibalization. To know if the change really helped, count picks for your whole product family before and after, not just for the product you changed, and track how often your products fill the whole shortlist.",
  "take": "Most AI visibility checks track one product or one brand name, which hides this. In our own shelf test, one brand's sibling pair lost half of everything a rewritten rival gained.",
  "summary": "Rewrite one product page and it starts getting recommended more. Great, unless those picks came from your own other versions. Here's how to tell, using one real example from a newsletter and one from our own shelf test.",
  "claims": [
    {
      "text": "Every's Mike Taylor rewrote one article's title and description. Every's share of all citations for that question rose from 35% to about 60%, while his other article on the same topic fell from being cited in 65% of sessions to 19%.",
      "source_title": "Every: How to Get AI Agents to Recommend Your Articles",
      "source_url": "https://every.to/also-true-for-humans/how-to-get-ai-agents-to-recommend-your-articles",
      "source_date": "2026-10-27",
      "method": "documented",
      "limitation": "Articles cited by AI agents in Every's own simulation, not products. Reported by the author; we did not rerun it."
    },
    {
      "text": "In our simulation, the invented humidifier was picked 3 times in 288 Claude sessions with its original vague copy and 213 times in 288 sessions with the facts written as a spec line. Of those 210 extra picks, 107 came out of one brand's two sibling models (165 picks down to 58).",
      "source_title": "Cart for Agents: Product cannibalization in AI shopping, How we counted",
      "source_url": "https://cartforagents.com/guides/sibling-sku-cannibalization#how-we-counted",
      "source_date": "2026-11-04",
      "method": "controlled-simulation",
      "limitation": "Recounted from the October 27, 2026 simulation: all valid Claude sessions (Haiku, Sonnet and Opus 5.5) for each copy version, across three shopper briefs and both rounds, two picks per session. The published research page reports a smaller held-out subset, so its counts differ."
    },
    {
      "text": "Brand A's two siblings were picked together, filling the whole two-product shortlist, in 54 of 288 sessions with the vague copy and 1 of 288 with the spec line.",
      "source_title": "Cart for Agents: Product cannibalization in AI shopping, How we counted",
      "source_url": "https://cartforagents.com/guides/sibling-sku-cannibalization#how-we-counted",
      "source_date": "2026-11-04",
      "method": "controlled-simulation",
      "limitation": "Same sessions as above. The rewritten product was invented by us and the competing listings were real snippets; we never rewrote a sibling."
    },
    {
      "text": "Across the four listings from that brand on the shelf of seven real listings, picks fell from 241 to 83, so 158 of the invented product's 210 extra picks came from one brand.",
      "source_title": "Cart for Agents: Product cannibalization in AI shopping, How we counted",
      "source_url": "https://cartforagents.com/guides/sibling-sku-cannibalization#how-we-counted",
      "source_date": "2026-11-04",
      "method": "controlled-simulation",
      "limitation": "One shelf, one shopper need (a quiet humidifier), supplied snippets. Shows where picks moved, not why."
    }
  ],
  "sources": [
    {
      "title": "Every: How to Get AI Agents to Recommend Your Articles",
      "url": "https://every.to/also-true-for-humans/how-to-get-ai-agents-to-recommend-your-articles",
      "accessed": "2026-11-04"
    },
    {
      "title": "Cart for Agents: Vague vs specific product facts simulation",
      "url": "https://cartforagents.com/research/vague-specific-product-facts-simulation",
      "accessed": "2026-11-04"
    }
  ],
  "limitations": [
    "We have not rewritten one of a merchant's own sibling products and measured the result. Our shelf example shows a rival taking picks from a sibling pair, which is the same arithmetic from the other side.",
    "Everything in our example comes from a simulation with supplied listing snippets. Real assistants search, read pages and personalize.",
    "Every's example is about articles cited by AI agents, not products bought through them.",
    "Picks are not sales. We did not measure orders, revenue or margin."
  ],
  "related": [
    "https://cartforagents.com/research/agent-shopping-research",
    "https://cartforagents.com/guides/product-evidence-graph",
    "https://cartforagents.com/guides/product-variant-structured-data",
    "https://cartforagents.com/guides/ai-visibility-testing",
    "https://cartforagents.com/guides/recommendation-receipts",
    "https://cartforagents.com/guides/clipped-product-descriptions",
    "https://cartforagents.com/research/vague-specific-product-facts-simulation"
  ],
  "published": "2026-11-04",
  "updated": "2026-11-04",
  "checked": "2026-11-04",
  "kind": "guide"
}