Panel & Market Methodology

What a market is

A market is one vertical × one destination — private banking in Hong Kong, UHNW villa rental in Thailand. Every observation, panel, query and reading is scoped to that pair. Nothing is shared implicitly across markets.

How the query panel is built — three layers, each human-validated

The panel is never a single generation. It is built in three layers, with a human gate after every layer — operator and analyst review; nothing advances unvalidated.

• Structured baseline (~80 queries) — generated under an operator-written instruction across the seven intent categories, covering the market's sub-activities; reviewed line by line, corrected, and re-delivered for sign-off.
• Deep-research long-tail (~80–100 candidates → keep 20–25) — a long autonomous research pass surfaces researched long-tail queries: micro-locations, buyer situations, role-confusion, trust and due-diligence angles. Humans curate the list for market relevance and coverage; the default rule is keep 20–25 maximum.
• Generic completion — research output skews long and thoughtful; real users type short, classic queries. The operator adds back the generic phrasings people actually use.

A full market panel typically contains around one hundred governed queries. Panel depth varies by market; Method & Standards reports the dated median for the current 100-market set. The wider query universe, the panel used for a capture and the usable coverage of a reading are distinct. Reading Basis states the coverage applicable to each reading.

The balance doctrine — localization is deliberate, not default

A run is already geo-scoped by its capture location, so repeating the destination name in every query is redundant and unrealistic. The standing rule: ~25–35% of queries name the destination; ~65–75% are generic. The destination name is kept only on specific micro-places, marquee events, authority heads (“best / top / leading [service] [city]”), and the four spotlights — except where the destination name is the market's identity.

The seven intent categories

Every query carries exactly one category: AUTHORITY · DISCOVERY · COMMERCIAL · EXPERIENCE · COMPETITIVE · TRUST · SEGMENT. Categories are part of the governed metadata: they scope what a reading may claim and keep panels comparable across markets.

Spotlights — four flagship queries per market, per language

Each market designates four spotlight queries — typically three broad authority or commercial heads and one long-tail trust query — tagged with priority and role, in every language the market runs in. A flagship that exists only in English produces no spotlight reading for a French run.

Languages and geography — three distinct notions

Capture location is the geography of the run, validated at country level before any run. Market languages are configuration metadata. Run language is chosen per run — an English run and a French run are separate, dated observations. Parity across languages is an exact 1:1 translation by intent: the same queries, faithfully translated, never a reinterpretation; proper names stay verbatim.

The actor panel — real, named, observable

No actor enters the registry without its verified bare domain (name.tld — no www, no protocol, no path). Registry identities help attribute observed rankings, citations and mentions; they do not form a whitelist deciding who may appear. A discovered actor without a verified domain is completed first or not added to the registry. The observed surface may include operators, publishers, platforms, institutions, intermediaries and unexpected actors. Their identity and role are resolved without treating every visible actor as a direct commercial competitor. Registry actors are researched and verified against the real local market — not generated.

Activation — ten gates and a versioned reference panel

A market goes live only after ten server-side activation gates pass — validated location, country, language, geographic scope, minimum enabled actors and queries, at least one core actor, no identity conflicts. Activation writes a versioned panel snapshot: the reference panel configuration is recorded at that moment. Adding actors later requires an explicit re-snapshot — the panel version changes, and the relevant version remains part of the reading’s evidence basis. A snapshot records that configuration; it neither limits the actors an observation may reveal nor, by itself, establishes comparability.

Run metadata — search conditions and reading scope

Each search run records its market, panel snapshot version, executed query set, capture location, language, device, engine and date. A fixed series signature supplies version context; comparability also requires compatible measures, panels, capture conditions and actual coverage. DNC Lens and Model Watch instead identify their narrative source, mode or surface and capture window. A single run remains one dated observation; a period reading brings comparable observations together.

How readings are analyzed

Organic rankings count blue links only; the engine's own blocks (local pack, People-Also-Ask, knowledge panels, video) are archived per query but excluded from ranking computations. AI Overview citations come only from the AI Overview rail, with capture artifacts filtered at the extraction root. Metrics are computed per the governed metric dictionary; cards are written under family contracts and pass deterministic validations before they may exist.

Who decides what

The machine proposes and executes; humans validate at fixed gates — baseline review, long-tail curation, spotlight choice, activation. Every write is journaled with its actor. The methodology itself is versioned: this page states the doctrine; the operator runbook records its current execution.