[
 {
  "id": "H1",
  "statement": "AI reduces search and comparison effort, improving some consumer choices and lowering some merchant acquisition costs.",
  "verdict": "Partly supported for effort; not shown for choice quality or acquisition cost",
  "mechanism": "Assistants compress search and comparison into a conversation and hand a decided shopper to the merchant.",
  "support": "Research and comparison are the most common uses in large surveys; AI-referred visits convert better in vendor panels; organic referrals carry no documented per-order fee.",
  "support_claim_ids": [
   "CON-C210",
   "CON-C212",
   "MKT-C012",
   "ECO-C064"
  ],
  "against": "Theory and one field study find consumer value can fall once sellers respond; lab agents are swayed by position and nudges; paid units are moving inside the answer, which can raise acquisition cost again.",
  "contrary_claim_ids": [
   "CON-C001",
   "CON-C229",
   "DIS-C216",
   "DIS-C055"
  ],
  "alternative": "Higher conversion reflects shoppers who had already decided, not better choices.",
  "scope": "US-first consumer goods; surveys and vendor panels; no welfare measurement.",
  "would_change_if": "A field study measured choice quality or consumer surplus for real shoppers using assistants against a control group."
 },
 {
  "id": "H2",
  "statement": "Most current routes assist discovery or prepare checkout; delegated execution is narrower than the market language implies.",
  "verdict": "Supported in the documents reviewed (no live route observed)",
  "mechanism": "Specifications default to a human approving one exact checkout; delegation needs mandates that are new, optional or in preview.",
  "support": "Every live assistant route reviewed assists; approval-based checkout is documented in Copilot, Google and Perplexity; OpenAI shifted ChatGPT's general route to merchant checkout in March 2026, with residual in-chat documentation unresolved; documented delegation is narrow (Amazon Auto Buy for Prime members); surveys of willingness to let an agent buy ask about different tasks and levels of control and report no single level; no comparable public total isolating completed consumer retail purchases under prior delegation was found in the primary sources reviewed.",
  "support_claim_ids": [
   "RTA-C050",
   "RTA-C214",
   "RTA-C301",
   "RTA-C315",
   "RTA-C324",
   "PAY-C050",
   "PRO-C043",
   "CON-C214",
   "MKT-C043"
  ],
  "against": "The protocols already allow more: AP2 defines purchases within limits signed in advance and UCP excepts AP2 mandates from manual finalisation; Amazon's Auto Buy is delegation inside a marketplace; Visa reports controlled agent-initiated transactions; Alibaba claims hundreds of millions of orders through its assistant (subsidised); Copilot and Google moved checkout into the assistant during 2026.",
  "contrary_claim_ids": [
   "RTA-C051",
   "RTA-C337",
   "RTB-C223",
   "RTA-C301"
  ],
  "alternative": "Delegation is real but concentrated inside closed platforms, where it is not reported as 'agentic'.",
  "scope": "Holds for open, cross-merchant assistants in the US; weaker inside closed marketplaces and in China.",
  "would_change_if": "An operator disclosed volumes of orders placed under standing mandates across unaffiliated merchants."
 },
 {
  "id": "H3",
  "statement": "Agent-led selection increases the value of accurate product, availability, delivery and policy data, but does not eliminate brand, taste, trust or creative demand creation.",
  "verdict": "Supported as a mechanism; brand effect inside agents not measured",
  "mechanism": "Agents read structured data and refuse or fail on stale data; operators sell brand persuasion inside the conversation.",
  "support": "Reported failures of in-chat checkout included stale stock and delivery data; Google refuses a purchase above the confirmed total; protocols standardise product data; sponsored brand prompts and model-written ad copy are sold inside assistants; LLMs lean to global brands.",
  "support_claim_ids": [
   "DIS-C019",
   "ECO-C045",
   "RTA-C214",
   "DIS-C001",
   "DIS-C035",
   "DIS-C031",
   "DIS-C217"
  ],
  "against": "No study measures whether brand memory changes what shoppers ask agents, or brand concentration in live shortlists.",
  "contrary_claim_ids": [
   "DIS-C221",
   "DIS-C230"
  ],
  "alternative": "Operators sell brand units because advertisers buy them, not because they move agent choice.",
  "scope": "Documentary and lab evidence.",
  "would_change_if": "A matched-query study of live assistants measured how brand demand and data quality change shortlist presence and purchases."
 },
 {
  "id": "H4",
  "statement": "New discovery intermediaries may change who captures fees and customer access even where the merchant remains merchant of record.",
  "verdict": "Not supported on disclosed channel commissions (some routes stated none; others not established); supported for paid placement; partly supported for customer access (base schemas permit thin data, UCP requires a confirmation email; delivery not observed)",
  "mechanism": "The intermediary controls the shortlist, the customer fields the merchant receives and the consent shown; fees move to ads.",
  "support": "Paid units inside ChatGPT, AI Mode and Amazon's assistant; ACP requires only an email and UCP no buyer field; consent options may be partly shown; Google's 2012 move to paid-only listings is a dated precedent.",
  "support_claim_ids": [
   "DIS-C020",
   "DIS-C030",
   "DIS-C036",
   "ECO-C030",
   "ECO-C033",
   "ECO-C031",
   "HIS-C032",
   "HIS-C069"
  ],
  "against": "Public per-order fees on AI checkouts where the merchant is seller are zero or unknown; the one reported 4% fee belonged to the earlier standalone Instant Checkout model, which OpenAI scaled back; current operation and pricing for residual eligible-product checkout remain unresolved; Perplexity paused ads. Microsoft and Google say merchants keep the customer data in their checkouts; no report of a merchant losing customer details in an AI checkout was found.",
  "contrary_claim_ids": [
   "ECO-C064",
   "ECO-C065",
   "DIS-C047",
   "ECO-C008",
   "RTA-C204"
  ],
  "alternative": "Operators are buying adoption now and will price later.",
  "scope": "Public terms; private agreements not seen.",
  "would_change_if": "Operators published per-order fees, or merchants reported receiving full customer details and consent by default."
 },
 {
  "id": "H5",
  "statement": "Social influence and conversational assistance may combine within existing platforms rather than become separate successor channels.",
  "verdict": "Supported in the US record; contradicted as a general rule by China",
  "mechanism": "Platforms with audiences add assistants to existing feeds and marketplaces.",
  "support": "TikTok's Shopping Assistant sits inside TikTok; Meta's reported tests run in Meta's apps; Amazon's assistant sits inside Amazon; US social checkout was repeatedly folded back while discovery stayed.",
  "support_claim_ids": [
   "RTB-C001",
   "RTB-C013",
   "RTA-C021",
   "HIS-C067"
  ],
  "against": "In China new platforms displaced incumbents and live commerce scaled as its own channel; ChatGPT is a new intermediary outside existing commerce platforms.",
  "contrary_claim_ids": [
   "HIS-C068",
   "HIS-C072",
   "RTA-C014"
  ],
  "alternative": "Combination reflects where audiences already are, and may change if a new assistant gathers its own.",
  "scope": "US-first.",
  "would_change_if": "A standalone assistant disclosed retail sales comparable to a large marketplace."
 },
 {
  "id": "H6",
  "statement": "Routine, well-specified and repeat purchases may suit delegation better than ambiguous, expressive or high-consequence choices.",
  "verdict": "Mixed: supported by where delegation exists; not by what shoppers say",
  "mechanism": "Delegation needs a task that can be specified in advance and checked.",
  "support": "The clearest documented delegation buys a known item at a set price; scheduled reorders persisted while open-ended voice and button buying shrank; one survey ranks household items first.",
  "support_claim_ids": [
   "RTA-C051",
   "HIS-C223",
   "HIS-C224",
   "CON-C223"
  ],
  "against": "47% of European respondents are uncomfortable with AI managing recurring purchases; only 15% would trust an agent with the weekly food shop; risk lowers comfort more clearly than routine raises it.",
  "contrary_claim_ids": [
   "CON-C039",
   "CON-C013"
  ],
  "alternative": "Routine purchases are delegated to subscriptions and marketplaces already, without agents.",
  "scope": "Survey and documentary; no observed category data.",
  "would_change_if": "Category-level data on delegated orders and their return rates."
 },
 {
  "id": "H7",
  "statement": "Lower interaction friction may raise conversion without creating incremental demand, increasing profit or improving customer welfare.",
  "verdict": "Untested: no source measures the share of AI-route orders that are new to the business; a descriptive pilot cannot, a contemporaneous holdout could",
  "mechanism": "Shoppers referred by assistants arrive late in the decision, so they convert better whatever the assistant adds.",
  "support": "Conversion gains are reported by vendors with selection effects; no source measures incremental demand, margin or repeat purchase; conversion evidence conflicts (Walmart's in-chat third versus vendor uplifts); Meta's mandatory in-app checkout did not last.",
  "support_claim_ids": [
   "MKT-C012",
   "ECO-C066",
   "ECO-C044",
   "HIS-C070"
  ],
  "against": "Vendors report large and growing conversion uplifts.",
  "contrary_claim_ids": [
   "ECO-C040",
   "ECO-C223"
  ],
  "alternative": "AI referrals bring better-informed shoppers who buy more over time.",
  "scope": "Vendor panels; no holdout tests.",
  "would_change_if": "A holdout test by route showed incremental orders and margin."
 },
 {
  "id": "H8",
  "statement": "Standards can reduce integration costs while commercial access, permissions, incentives and route-specific implementations remain fragmented.",
  "verdict": "Supported within scope",
  "mechanism": "Common schemas and transports lower plumbing costs; governance and access stay with the founders and the operators.",
  "support": "ACP and UCP overlap without referring to each other and use different discovery files and tool names; each is controlled by its founders; MCP and A2A carry messages without commerce meaning; access is decided by private terms (Shopify sole discretion, operator approval); identity schemes sit apart from spending authority.",
  "support_claim_ids": [
   "PRO-C017",
   "PRO-C035",
   "PRO-C039",
   "PRO-C003",
   "PRO-C020",
   "PRO-C053",
   "RTB-C027",
   "PAY-C072"
  ],
  "against": "Shared members in both camps (Stripe, Meta, PayPal, Adyen) and common transports could make the split invisible to merchants through payment providers.",
  "contrary_claim_ids": [
   "PRO-C024",
   "PRO-C063"
  ],
  "alternative": "Fragmentation is transitional.",
  "scope": "Specifications at pinned commits; deployment not observed.",
  "would_change_if": "An official ACP–UCP interoperability statement, or published live cross-system transaction volumes."
 },
 {
  "id": "H9",
  "statement": "Existing marketplaces, brands and trusted merchants may retain more power than a universal-agent narrative predicts.",
  "verdict": "Supported in the US; not universal",
  "mechanism": "Incumbents control catalogue ingress, checkout, payments and delivery; merchants can refuse agents.",
  "support": "Catalogues reach assistants through Shopify, Google and Amazon; Walmart replaced in-chat checkout with its own app; Amazon runs delegation on its own terms and sued an outside agent; the specifications keep the merchant as seller and after-sales owner; retailers blocked the first shopping bot in 1995.",
  "support_claim_ids": [
   "DIS-C060",
   "RTB-C018",
   "RTA-C337",
   "PAY-C035",
   "HIS-C071"
  ],
  "against": "The Ninth Circuit vacated Amazon's injunction against Perplexity's agent; in China new platforms displaced incumbents.",
  "contrary_claim_ids": [
   "RTA-C322",
   "HIS-C068"
  ],
  "alternative": "Incumbents' power reflects the early stage, not a stable outcome.",
  "scope": "US-first; litigation pending.",
  "would_change_if": "An independent assistant gained checkout access to a large marketplace's catalogue, or courts settled that marketplaces cannot exclude shoppers' agents."
 }
]