Niche and Intent Research

1) The problem this lesson owns

A keyword list is not a market. Several queries can have attractive impressions while much of the traffic is branded, some queries are omitted for privacy, and the best-performing page solves a broader problem than the proposed niche. COM-005 turns that ambiguity into an auditable evidence map.

2) Exact COM-005 boundary

This lesson owns Niche-Intent Evidence Map v1. Website Audit Engine belongs to COM-004 and Multi-model Editorial Cell belongs to COM-006. COM-005 does not prove total addressable market, profitability, or directly observed human intent, and it performs no live campaign or production mutation.

3) FACT: what Search Console actually observes

The Performance report exposes metrics such as clicks, impressions, CTR and average position, with dimensions such as Query and Page. These are observations about Google Search performance for the property, not a complete market-demand measure and not a direct intent label.

4) FACT: the query table is incomplete

Google documents that some queries are anonymized for privacy and that table data can be truncated. A missing row therefore does not prove missing demand, and visible rows cannot be treated as the entire market. Preserve these boundaries in data_limitations.

5) FACT: branded/non-branded is not intent

Search Console can separate branded and non-branded queries, but the classification is informational and can be imperfect. Rising branded traffic may indicate brand familiarity; it does not by itself prove new-market demand or commercial intent.

6) Separate observation, inference and business hypothesis

query_observation and page_observation are observations. intent_hypothesis is an inference. commercial_fit_signal is a business-fit hypothesis/evidence field. Merging them into one confidence number destroys traceability.

7) A query string alone does not prove intent

A phrase such as “best invoice reconciliation software” is directional evidence, not an oracle. Intent needs page and business context. If the workflow infers transactional intent solely from wording, the correct state is HOLD_INTENT_OVERCLAIM.

8) Page evidence prevents false commercial fit

A cluster can have strong clicks and impressions while landing on a page outside the proposed offer. Query demand and offer fit are different questions. That mismatch should drive RESEARCH_PAGE_OFFER_MISMATCH or equivalent research, not automatic pursuit.

9) Do not use average position as an oracle

Average position is reporting context and reflects the topmost result under Search Console aggregation rules. Favorable position plus high impressions still does not prove profitability or intent. Interpret it beside clicks, impressions and context.

10) Compare like with like

If one period is Germany/Mobile and another is US/Desktop, a change in impressions is not clean market-growth evidence. date_range and country_device_filter are part of the contract; materially different filters produce HOLD_FILTER_MISMATCH.

11) Evidence strength without a magic weighted score

MAKNOW does not hide decisive gaps inside a blended score. Strength depends on explicit boundaries: limitations recorded, intent supported by query+page+business context, commercial fit known, and provenance complete. More impressions cannot compensate for a critical evidence failure.

12) Core Evidence Contract

The artifact has 21 dimensions, 19 of them critical. The compact table below shows a traceable baseline; use the JSON/CSV artifact for the complete field set.

13) Correct baseline

The baseline records query/page metrics with declared date, filters and source; limitations are explicit; intent remains a hypothesis supported by page/business context; and commercial fit is aligned. PURSUE is therefore allowed because the evidence chain is bounded, not because one keyword looks attractive.

14) Tempting shortcuts and repairs

Three common shortcuts are deleting anonymization/truncation limits, treating branded traffic as proof of new-market demand, and inferring intent only from query wording. Repair starts at the broken field: restore limitations, align filters, add page/offer context, then recompute the decision.

15) Lab and transfer

The lab contains eight offline fixtures and 48 deterministic assertions. In transfer, impressions rise while clicks stay flat, branded share jumps, and one high-intent cluster maps outside the proposed offer. Rebuild evidence strength and the PURSUE/HOLD/RESEARCH/REJECT decision from the changed evidence instead of copying the baseline.