# Who controls local search? Methodology v1.0

Status: preregistered pilot rules; final methodology locks after ten searches. User explicitly authorised pilot-led refinements on 28 September 2026. This supersedes a literal no-pilot reading of the generic Research sequencing rules for this study only. No outcome data collected before this record.

## Scope and sample

Five query terms: plumber, restaurant, hairdresser, accountant, florist. Twenty named locations in sample-v0.1.json, all five terms at each location. No independent/best/near-me variants and no outcome-driven substitutions. These terms operationalise five sectors, not every service in those sectors. All nine English regions and all four UK nations are covered; 12 English, three Scottish, three Welsh and two Northern Irish locations. This is purposive geographic coverage, not a representative probability sample, and smaller/rural places are underrepresented. Derry is the fixed English query spelling; no variant experiment.

Pilot: all five terms in Preston and Glasgow (Q001–Q005 and Q061–Q065), selected before observation for different city sizes and nations. It tests collection/ranking/classification practicality rather than estimating UK shares. If substantive rules change, recollect these ten under the locked method; otherwise carry forward with explicit provenance. Exactly 100 distinct final queries.

## Collection controls

Use actual Google desktop All results in one browser and session. English language, UK country preference, exact place-qualified query. Candidate request parameters hl=en, gl=gb, pws=0; verify non-personalisation against the visible footer rather than treating parameters alone as proof. Existing browser is signed in, as observed on Google home before pilot. Do not describe this as incognito, clean-session or signed out. Do not change account settings or sign out the owner's session. Capture sign-in state, visible non-personalisation label, displayed location and its basis, exact URL and UTC timestamp for every query. Keep device/viewport consistent and record observed dimensions. No geolocation permission changes, residential proxies, CAPTCHA bypass or simulated city-local user claims. Explicit city queries from one UK vantage point answer a bounded question about searches FOR those places, not what residents IN each place necessarily see. Google says location and other context still affect non-personalised results: https://support.google.com/websearch/answer/12410098?hl=en .

Capture the first ten eligible standalone organic web results in displayed order. Exclude ads/sponsored, Maps/local packs, AI answers/citations, People also ask, image/video/news/shopping carousels, knowledge panels, related searches and sitelinks. A standalone editorial, Reddit, YouTube or social-profile web result counts; module cards do not. Featured snippets are excluded as special modules, with exclusion recorded. Distinct standalone same-domain results each occupy a rank; deduplicate only repeated rendering of the identical card. Record result title, exact destination URL, hostname, registrable domain, eligible rank 1–10 and Google pagination page. Use visible Next if first page has fewer than ten; maximum two pages and record shortfall. Never call this page-one share if pagination was needed. Do not substitute generic web-search tool rankings for Google observations.

Retain a DOM text snapshot and screenshot evidence per captured page; strip account identifiers from shareable evidence. Research-only raw evidence stays private. Record block/failure/shortfall explicitly, retry a transient failure once without changing query, never silently replace failed locations. Stop bulk capture if controls or rank evidence are unreliable. Aim to finish collection within one day; disclose exact date range if it spans days. No repeated-date panel.

## Classification

Mutually exclusive primary types, applied to the landing page's function and operator, with a short evidence note and supporting URL. Classify every unique business domain using its About/contact/locations/legal page where needed; homepage/snippet alone is not proof of independence. Mixed-purpose platforms may need page-level decisions. Use pending review while working, and unresolved Other with reason at freeze if bounded checks cannot settle a case. Never force an ambiguous entity into SME.

1. Independent/local business: direct seller/provider, with positive evidence of owner-operated or independently operated local/regional business and no evidence of national chain, franchise or corporate group control. Multiple local branches do not automatically exclude it. This is an observable local/independent proxy, not legally verified SME size; public findings must use independent/local, with this caveat adjacent to any SME language.
2. Large/chain business: direct provider under a national/international chain, franchise/network brand, or documented large corporate group. A local franchise page is included here consistently even if its franchisee may legally be an SME. Record franchise and regional-group ambiguity; independent regional businesses with unverified size stay in the local proxy only when independence is evidenced.
3. Commercial intermediary: a third-party directory, review, booking, marketplace, lead-generation or aggregator page helping users discover/evaluate/contact/book/buy from businesses it does not itself operate. Includes a single-business profile on such a platform; directness is about the destination host. Supplier plus marketplace hybrids use the result landing page's primary role. Affiliate-supported editorial lists remain publisher unless the primary role is directory/transaction matching.
4. Publisher/editorial: news, magazine, guide or editorial article, including council/tourism editorial lists. A council/tourism searchable business directory is intermediary by function, with non-commercial operator subtype disclosed rather than assumed commercial. For strict commercial totals, non-commercial directories go Other/non-commercial directory.
5. Forum/community: discussion/social/community destinations. Reddit always this type, separately reported by domain, never included in commercial intermediary totals. A business-owned Facebook/Instagram profile remains hosted social (subtype business profile), not an owned business domain.
6. Other: government/institutional/service information, non-commercial directories, inaccessible/uncertain role or ownership after bounded review. Keep uncertainty reason separate from genuinely other content.

Store subtype, ownership confidence (confirmed/provisional/unresolved), classification evidence URL, rationale, reviewer and review timestamp. Check a deterministic 10% of result rows plus every ambiguous case a second time; single-researcher review must not be called independent double-coding. Bound difficult domain checks to approximately two minutes, then record uncertainty. Report unresolved share and a sensitivity range for independent/local share (confirmed numerator through confirmed plus unresolved-business numerator), without guessing ownership. No legal employee/turnover verification project.

Use registrable domains, consolidating www and ordinary subdomains with a Public Suffix List-aware parser or explicit reviewed mapping; preserve original host. Do not merge separate registrable domains based on suspected ownership. Concentration is domain concentration, not corporate ownership concentration. Hosted-business platforms need a recorded mapping decision.

## Analysis fixed before collection

Unit: eligible organic position. Overall category shares = category positions / all observed eligible positions. Report n/N alongside percentages and missingness out of planned 1,000. Top-three denominator is all captured eligible ranks 1–3, planned 300. Sector denominators planned 200 and 60. All five terms and all 20 places are equally represented by design, no traffic or population weights. If a query is incomplete, report complete-query sensitivity from these same observations.

Commercial domain counts rank descending by positions, alphabetical domain tie-break. Top 3/5/10 concentration = positions belonging to the leading k commercial intermediary domains / all commercial intermediary positions. Also show their share of all eligible positions and the number of distinct queries they appear in / valid queries. If fewer than k domains exist, show available count; zero commercial positions gives N/A concentration. Preserve duplicate-domain positions for exposure count; query reach counts a domain at most once per query.

Compare count of unique independent/local domains, their total positions and positions per domain with unique commercial intermediary domains and their repeat frequency. Do not compare a count of domains against a count of positions without explicit labels. Show sector shares and sector-leading intermediaries from the same data. Publishers and forums remain separate; a broad third-party subtotal must specify constituent categories and cannot silently treat all Other as third-party. No causal crowding-out, click/traffic share, national market-share, monopoly or regulatory conclusions. No significance claims from this purposive sample.

## Pilot gate and time box

Pilot succeeds when ten queries have ordered evidence, exclusion rules work, search-state controls are documented, and classification can be completed within the budget. Log necessary amendments, freeze final sample/method with timestamp and SHA256 hashes, then collect remaining 90. Freeze observations before analysis. Corrections create a new version and change note. Avoid new infrastructure beyond simple capture records, classification registry and reproducible arithmetic.

Working budget: method and pilot 3–4h; remaining collection 4–5h; classification/QA 4–5h; analysis 1–2h; publication-ready writing/figures 2–4h. Total 14–20h across at most three working days. At a collection blocker, preserve work and identify the smallest viable resolution; do not substitute invented data or expand the study. Publication-ready handover includes methods, dataset/dictionary, exact-count findings, chart-ready tables, limitations, citation text and evidence-led outreach draft angles. Actual website publication and sending pitches require the owner's separate authorisation. No schema/CWV/AI audits or future studies 10/11.

## Final-method clarifications from pilot (v1.0)

These rules override ambiguities above. Lock time is in methodology-lock-v1.0.json. The fixed sample is unchanged. All 100 queries, including the ten pilot queries, will be freshly collected under v1.0 to keep the final run distinct from exploratory evidence.

- The category independent/local is an operational proxy, not a verified SME population. Positive owner/family/independence evidence or explicit local trading identity plus reviewed operator/team/contact information is required. Use evidence_strength=ownership_explicit or local_operator_supported; report the explicit-ownership subset separately as a classification sensitivity, derived from the same results. If provider role is clear but independence/scale cannot be established, use Other/unresolved-business and retain direct_provider=true. Do not infer independence from a geographic keyword in a domain.
- Apply large/chain ahead of independent: documented 250+ employees, 10+ branded outlets/offices, national/international multi-site chain/franchise, or subsidiary of such a group. The ten-location boundary is a disclosed practical coding convention, not the legal SME definition. Record regional_multi_site where fewer locations and positive local/independent evidence support local coding. Remote national delivery alone does not establish company size. A national online seller with insufficient size/ownership evidence is unresolved-business. Membership of Interflora, a professional association or a supplier network alone does not make a separately branded local operator a chain.
- Direct-provider content, including its own advice article, follows the provider's operator class. An independent publisher's review/listicle is editorial; a searchable or bookable third-party business catalogue is intermediary even if its listings involve expert reviews (for example Michelin's restaurant finder). Ordinary editorial articles on the same domain may have a different class. Preserve page-specific overrides.
- Commercial intermediary means a for-profit discovery, lead, booking, marketplace or review service. Free consumer access is not evidence of non-commercial status. Government-backed/professional-body accreditation and membership directories are Other/non-commercial-directory unless there is evidence the discovery service itself operates commercially. Identify contractor-operated schemes explicitly and do not equate a public endorsement with public ownership. Any unresolved commercial status is Other/unresolved-intermediary with sensitivity range for commercial share/concentration if material.
- Social business profiles remain forum/community with subtype business-profile; discussion threads use subtype discussion. This is a host-based destination definition and does not imply the business lacks control of its social content.
- Exclude every grouped result module, including Find results on, Discussions and forums, Latest in local food and embedded video/news cards. Standalone results from those same hosts remain eligible. No measure of total screen real estate, all Reddit exposure or Maps visibility is implied.
- Ranking authority is the retained rendered DOM snapshot under Web results headings, reviewed against visible result presentation. Store standalone URL/title/order and excluded module headings. Full-page screenshots from this browser can show stitching repetitions; retain them as supplementary evidence but never infer duplicate ranks from screenshot stitching. Use normal viewport screenshot for final captures. On an unfamiliar layout, flag for manual review instead of silently extracting zero results.
- Retain raw URL and a separate normalised destination URL without fragments/known tracking parameters; do not drop query parameters that identify content. Exact-card duplicates within one page count once; separately presented organic cards from one host remain separate. Keep page number. Never label cumulative eligible rank as Google's absolute rank among modules.
- Domain normalisation uses an explicit reviewed registrable-domain mapping, preserving source host; standard UK second-level endings (co.uk/org.uk/ac.uk/gov.uk) and ordinary TLDs are unambiguous in the pilot. Inspect hosted/private suffix cases manually. Separate registrable domains such as tripadvisor.com/co.uk remain distinct; cross-country aliases may be noted but are not silently merged into top-k domain figures.
- Collection stays in the observed desktop 1280×720 browser, signed in, English, UK preference, pws=0, without changing account settings. Each capture must visibly show Results are not personalised and preserve footer location. The pilot location was PR3, UK, from the IP address. This single-vantage geography remains a disclosed limitation. Do not save account identifiers in shareable evidence.
- Pilot classification checking found viable primary-site evidence but some HTTP blocks and weak ownership disclosures. Use at most one alternative read or linked About/team page before uncertainty coding; revisit only high-impact cases in the final QA budget. Do not turn this into corporate ownership research. Report unresolved counts and sensitivity alongside point estimates.
- Deterministic QA sample: the rank-1 row of every query (100 of 1,000) plus all flagged ambiguous classifications/rankings. Check duplicate counts, 100 unique prescribed queries, exact ranks, denominator arithmetic, category completeness and domain mapping before final freeze.