match-types

name

match-types

description

For a Google Ads account, judge whether spend is over-concentrated in one keyword match type (Broad/Phrase/Exact) and whether any type is sitting untested at $0 — then flag rebalance vs test-a-type, with each type's cost / conversions / cost-per-conversion side by side. Use this skill whenever the user asks about match types, broad vs phrase vs exact, "am I too heavy on broad", whether to test a match type, match-type efficiency or budget concentration, or audits how keywords are bid — even if they don't say "match type". Judges SPEND CONCENTRATION + per-type efficiency ONLY; per-keyword relevance belongs to `search-terms/relevance` and search-term mining to `search-terms/n-grams`.

Match Types — Efficiency & Concentration

Goal (job-to-be-done)

Across the account's keywords, answer the client's question: is the budget over-concentrated in one match type (usually Broad), and is any match type untested (sitting at $0)? Put Broad / Phrase / Exact side by side — cost, conversions, cost-per-conversion, share of spend — and decide whether to rebalance spend or test a type. The unit of analysis is the match type; the output is a concentration verdict + the move to make per type.

• Who: media buyer / PPC manager. When: recurring account audit; the "match types" item of the account-audit cluster.
• Decision it drives: whether to shift budget off an over-weighted type, and which untested type to pilot — with the two real numbers (share + efficiency) in hand.
• The differentiator: the blend guardrail. Porter cannot split a single qualified action (MQL/Opp) by match type, so per-type conversions can be a blend that inverts the true ranking. This skill reports efficiency as directional only when that risk is present — it never asserts "most/least efficient" on a blended number.

Scope

• ✅ Spend concentration + per-type efficiency: cost, conversions, cost/conv, spend-share by Broad/Phrase/Exact; budget concentration %; untested types.
• ❌ Per-keyword relevance (does this keyword fit its term/ad?) → search-terms/relevance.
• ❌ Search-term mining (n-grams, new negatives, harvest) → search-terms/n-grams.
• ↔ Cross-reference search-terms/performance, which carries match type as a breakdown — use it to drill into which terms sit behind a flagged type.

Components (read these references as needed)

• Tools / data plan: references/tools.md — the keyword-grain query by match type + the conversion-action inventory probe.
• Framework / rubric: references/framework.md — the brain: concentration threshold, untested test, and the blend guardrail.
• Output schema: references/output.md — the JSON this skill emits.

Operate

Input: per match type (BROAD / PHRASE / EXACT) — cost_micros, clicks, conversions from the keyword view; plus a second probe (conversion-action inventory) to learn whether conversions blends more than one primary action. Brand context: which terms are branded, so they can be excluded.

Process: apply the rubric in references/framework.md. Group the keyword rows by keyword_info_match_type; compute cost, conversions, cost/conv and spend-share per type from the base counts (never trust a native ratio field). First exclude branded terms (../../../_framework/brand-vs-nonbrand.md) — branded keywords are usually Exact and flatter that type. Flag concentrated if one type holds ~≥70% of spend; flag untested for any type at $0. Run the blend guardrail before judging efficiency: if the account has >1 primary conversion action, mark every type directional and never claim one is "most/least efficient" — cost/conv is a blend that can invert the true qualified ranking.

Emit the JSON in references/output.md:

• synthesis — three strings: headline (the concentration + the move), diagnosis (which way the account leans — over-concentrated vs untested — and whether efficiency is trustworthy), action (the one move now, where / what / why).
• match_types[] — one per type: cost, conversions, cost_per_conversion, spend_share, verdict, directional, and a recommendation {where, what, why}.

A renderer (the orchestrator's formats/*) turns this JSON into the human document. Emit pure data — no emojis, tables, markdown, or colors in the output.

Voice (don't copy the rules, link them): write every narrative line per ../../../_framework/writing.md — the question the data answers yes/no; the metric+share carried as data, never spelled out in prose; first sentence answers it, then names the driver. Plain language for a non-technical owner ("almost all the budget is on one setting"), the technical term in parentheses.

Example (illustrative — NOT rules)

• Concentrated: Broad holds 78% of spend, Exact 18%, Phrase 4% → concentrated on Broad → rebalance: too much budget rides one setting; shift some toward the proven type.
• Untested: Phrase sits at $0 across the account → untested → test_type: pilot Phrase on the best ad groups to learn before committing budget.
• Blend trap (verified 2026-06-23): Broad looks best on all-conv ($373/conv) but worst on Cost/MQL ($1,515) once you split by qualified action — Porter can't make that split per type, so every type is directional: true and the skill never says "Broad is most efficient".