Caveats for the searchable-catalogue brief: no study isolates catalogues as a variable; vendor-sourced skip thresholds; Baymard age; AI-eligibility under-sourced

The honest summary of source quality across this brief:

• No study isolates "searchable catalogue" as a variable. The independent self-service research (HBR (Dixon et al., 2017) — 81% of all customers attempt self-service before reaching out to a live representative, Forrester (2016) — web/mobile self-service overtook the phone; FAQ-page use rose from 67% (2012) to 81% (2015) of US online adults, Gartner (Aug 2024) — only 14% of customer service and support issues are fully resolved in self-service; only 36% for "very simple" issues) measures self-service broadly. The link from "customers prefer self-service" to "therefore build a structured catalogue" is a reasoned inference, not a directly-measured finding. This is the single biggest evidentiary gap.
• The skip threshold (~200 items) is vendor-sourced. Three independent vendors (HawkSearch (vendor — concedes against interest) — for catalogs of "just a few dozen products," basic search and navigation are adequate, Prefixbox (vendor — faceted-search software, concedes against interest) — skip faceted search if catalog <200 products; implementation cost outweighs UX value, Luigi's Box (vendor) — independently corroborates the skip cutoff: "smaller catalogs with only a few hundred products may not require this level of complexity") converge — credible because it cuts against their commercial interest, but still vendor-sourced and not a validated cutoff.
• The Baymard 67-90% vs 17-33% figure is independent and primary but ~11 years old and e-commerce-specific (Baymard Institute (2015) — sites with mediocre product list usability saw 67-90% abandonment vs 17-33% for sites with even a slightly optimised toolset).
• No independent, current, generalizable CTR/conversion figure exists for structured data. The only clean magnitude is the Nestlé 82% — a single company's self-measurement, RING-FENCED (Nestlé 82% higher CTR on rich-results pages — Google's own documentation, attributed to one named company; RING-FENCED, never present as typical).
• AI-answer eligibility is under-sourced — the claim rests largely on vendor blogs (AI-citation eligibility via structured records — DIRECTIONAL; vendor-blog driven, Google's own position is "not a direct ranking factor").
• Bing's current rich-results behaviour is documented mainly via a 2018 announcement (Bing — supports JSON-LD, schema.org, Microdata, Microformats, Open Graph, RDFa via the Markup Validator (August 2018)) and third-party SEO guides.
• Most documentation is e-commerce-framed. Generalisation beyond products (to services, knowledge bases, registries) is principled but not heavily studied.

Practical posture for Candid: the mechanism (structured records → individually-indexable, faceted-queryable) is verified from primary sources. The magnitude of any uplift is not generalisable — treat all vendor CTR/AI-citation multipliers as quarantined (R6 — Quarantine vendor CTR / AI-citation multipliers; the only clean primary figure (Nestlé 82%) is single-company and ring-fenced) until independently proven.

See also the quarantine notes: QUARANTINE — "67% of customers prefer self-service over speaking with a live agent" (attributed to Zendesk), QUARANTINE — "Self-service interaction costs $0.10 vs $6-$12 for live agent" (attributed to Forrester), QUARANTINE — "Pages with structured data earn 35% higher CTR" / "2.5x higher chance of appearing in AI answers" / "40% more AI Overview appearances", QUARANTINE — "Microsoft reported a 22% reduction in support tickets" / "Spotify forum resolves 30% of issues", QUARANTINE — "60% of websites do not use facets and filters" / "only 10% of e-commerce sites use faceted sorting".