{
  "id": "N10",
  "url": "https://cartforagents.com/guides/recommendation-receipts",
  "title": "Why did AI recommend that product? Check its reasons with a recommendation receipt",
  "description": "A simple one-page record for checking an AI shopping recommendation: what the shopper needed, which listing states each fact, the alternatives, and what is still unknown. With real examples from 597 test answers.",
  "answer": "A recommendation receipt is a short record of one AI recommendation: what the shopper needed, the product picked, the listing that states each fact the AI used, the products it passed over, and what nobody stated. Fill it in whenever you test your products in an AI assistant, and check every fact against the listing it was credited to. It tells you whether the reason holds up, not why the model really chose.",
  "take": "In our own test answers, most reasons were careful, but some credited one product with its sibling's runtime. If you only read the AI's one-line summary, you can't tell the two apart.",
  "summary": "When ChatGPT or Gemini recommends a product, it gives a reason. Sometimes the reason is right, sometimes it borrows a fact from a different product. A recommendation receipt is a one-page way to check, line by line, before you act on what the AI said.",
  "claims": [
    {
      "text": "In 597 simulated picks for a detailed shopper brief (a light sleeper with a 450 sq ft bedroom, budget under USD 90), every model gave a one-sentence reason for choosing two humidifiers from the same shelf.",
      "source_title": "Cart for Agents: Vague vs specific product facts simulation",
      "source_url": "https://cartforagents.com/research/vague-specific-product-facts-simulation",
      "source_date": "2026-10-27",
      "method": "controlled-simulation",
      "limitation": "Supplied listing snippets with one invented product; models never searched, opened pages or bought anything. 432 sessions were Claude (Haiku, Sonnet and Opus 5.5), 61 GPT-6.1-Sol, 57 Gemini 3.8 Flash and 47 Gemini 3.1 Pro."
    },
    {
      "text": "Brand A's smart model, as the models saw it, stated 26 dB but no runtime; its standard sibling stated 62 hours. Gemini 3.8 Flash picked the smart model without the standard one in 28 sessions, and in 4 of them its reason gave the pair runtimes of up to 62 hours.",
      "source_title": "Cart for Agents: Why did AI recommend that product?, How we checked",
      "source_url": "https://cartforagents.com/guides/recommendation-receipts#how-we-checked",
      "source_date": "2026-11-04",
      "method": "controlled-simulation",
      "limitation": "We re-read saved reasons from the October 27, 2026 simulation. Some reasons describe two products at once, so the 62 hours may have been meant for the invented product's 40 hours as a range. Either way, the smart model's listing did not state it."
    },
    {
      "text": "Claude Opus 5.5 kept the siblings apart. In 3 sessions it picked the smart model without the standard one and wrote that the 62-hour figure belonged to the standard model.",
      "source_title": "Cart for Agents: Why did AI recommend that product?, How we checked",
      "source_url": "https://cartforagents.com/guides/recommendation-receipts#how-we-checked",
      "source_date": "2026-11-04",
      "method": "controlled-simulation",
      "limitation": "Three sessions. A model's stated reason is not proof of why it picked."
    },
    {
      "text": "In all 21 detailed-brief sessions where GPT-6.1-Sol picked two humidifiers that showed no room size, its reason said the 450 sq ft coverage was not confirmed.",
      "source_title": "Cart for Agents: Why did AI recommend that product?, How we checked",
      "source_url": "https://cartforagents.com/guides/recommendation-receipts#how-we-checked",
      "source_date": "2026-11-04",
      "method": "controlled-simulation",
      "limitation": "We matched reasons by keyword and read all 21 by hand. Small sample from one model."
    },
    {
      "text": "One humidifier listing stated \"up to 32 hours\" in its description and \"up to 36 hours\" in a feature bullet. GPT-6.1-Sol called out the mismatch in 20 of the 25 sessions where it picked that listing; Claude Haiku 5.5 did in 21 of 111, Claude Sonnet 5.5 in 3 of 73, and the two Gemini models in none of 64.",
      "source_title": "Cart for Agents: Why did AI recommend that product?, How we checked",
      "source_url": "https://cartforagents.com/guides/recommendation-receipts#how-we-checked",
      "source_date": "2026-11-04",
      "method": "controlled-simulation",
      "limitation": "Keyword match for words such as \"conflicting\" or \"inconsistent\"; a reason that noticed the mismatch in other words would be missed. Shows what reasons said, not whether the mismatch changed picks."
    }
  ],
  "sources": [
    {
      "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": [
    "Everything here comes from a simulation: we handed models the same supplied listing snippets and asked them to pick two. Real ChatGPT, Gemini or Copilot sessions search, read pages and personalize, and we didn't test those.",
    "One of the products, the Velmori Drift, is invented. The competing listings are real Shopify catalog listings from October 26, 2026, cut to short snippets.",
    "A receipt checks whether a stated reason is supported by the listings. It can't tell you what actually drove the choice. Price, ratings and brand may matter more than the reason says.",
    "Our counts use keyword matching plus hand reading of the flagged reasons. We read every sibling-runtime and room-size case quoted here, but not all 597 reasons."
  ],
  "related": [
    "https://cartforagents.com/guides/agent-commerce-optimization",
    "https://cartforagents.com/guides/product-claim-provenance",
    "https://cartforagents.com/guides/product-constraint-coverage",
    "https://cartforagents.com/guides/negative-product-attributes",
    "https://cartforagents.com/guides/ai-visibility-testing",
    "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"
}