Search retrieval architecture has moved past the document index. The classical pipeline that ranked pages by term frequency, link weight, and anchor text still runs, but it now sits at the bottom of a stack into which Google has inserted three additional layers: a term layer (BM25 and query rewriting), a concept layer (Hummingbird, RankBrain, BERT, MUM), and an entity layer (Knowledge Graph and reconciliation). Each new layer was added on top of the previous one. None replaced its predecessor. Rendered surfaces, however, increasingly draw from the top of the stack. Indonesian UMKM digital marketing tactics, training programmes, and certification schemes remain organised around the bottom layer. Empirical findings. Drawing on a twelve-year longitudinal SERP-composition record (30 commercial queries x 6 biennial snapshots, 2014-2026), an audit of 82 Indonesian commercial websites, and a single-operator natural experiment across an eighteen-month window, the paper reports three findings. First, the share of mobile commercial-query SERP viewport occupied by classical organic results fell from approximately 78% in 2014 to approximately 28% in 2026, with most of the difference now occupied by entity-anchored surfaces (Knowledge Panel, Featured Snippet, People Also Ask, AI Overviews). Second, 73% of the audited Indonesian UMKM operate keyword-only digital strategies, 23% operate hybrid strategies that include some on-page structured data without external identifier work, and 4% operate strategies anchored to a reconciled entity record. Third, the single-operator natural experiment shows a roughly three-and-a-half-fold rise in click-through rate for entity-anchored queries over twelve months following deliberate identifier deployment, while click-through rate for the same operators keyword-anchored queries remained flat. Contribution. A four-layer model of contemporary search retrieval (document layer / term layer / concept layer / entity layer) presented as an operator allocation framework. Three KPI substitutions for operators whose measurement still tracks the document layer alone. A thirty-minute SERP-composition self-audit any operator can run. A reading of why Indonesian digital-marketing curricula remain organised around the bottom of the stack, addressed to the Indonesian Information Systems academia and to the BNSP and Komdigi certification authorities. Methodology. Practitioner-researcher position. Cross-sectional audit of 82 commercial websites across listed firms, mid-size SMEs, small operators, and individual practitioners (Bogor and Jabodetabek, 2020-2026). Twelve-year SERP composition record from operator-controlled queries across three sectors. Single-operator longitudinal natural experiment on a personal-brand domain (October 2024 to April 2026). All limits of the causal claim are acknowledged in Section 6.4. Cross-references. Companion to four published working papers in the author Digital Marketing series: DM-01 The Schema Markup Adoption Gap, DM-02 Wikidata as Digital Business Infrastructure, DM-03 SameAs or Invisible, DM-04 The Knowledge Panel as a Trust Signal, and DM-07 Zero-Click Search and the Invisible Tax on Small Business. Author. Ibrahim Anwar (Hibranwar). ORCID 0009-0006-0425-4923. Wikidata Q138856145. Head of Data Analysis and Information Systems, Komite Ekonomi Kreatif Kabupaten Bogor (Kabekraf). Deputy for Creative Economy, KADIN Kota Bogor. Correspondence: anwar@hibranwar.com.
Ibrahim Anwar (2026) studied this question.