Google Ads Builder Branded Search

Google Ads — Builder: Branded Search

A focused Search campaign to (1) defend your own brand at near-certain ROAS and (2) optionally conquest competitors. Pushes cleanly via Editor CSV. Builds on builder-search patterns with brand-specific rules.

Operating rules

• Read brand_terms and competitor_terms from account-context.yaml (set by setup). Read guardrails.
• Write to the working directory: blueprint.xlsx + campaign-spec.json

(campaign_type: "branded_search", ad_groups[]).

• Coordinate brand exclusion: ensure PMax/Shopping have brand exclusion applied so they don't claim

brand conversions (clean attribution). Honor the pmax-brand-exclusion guardrail.

Model dispatch (run cheap, decide expensive) — see ${CLAUDE_PLUGIN_ROOT}/references/model-tier-dispatch.md

• Scout (haiku) — per-URL 200 checks; char-count validation; brand-misspelling expansion.
• Routine (sonnet) — pulling competitor domains from auction insights as conquesting candidates. Dispatch as general-purpose; return raw candidates, don't decide.
• Judge (main session) — STEP 1-2 ad-copy (brand-led, trademark-aware — never a competitor TM in ad text), STEP 3 separation, bidding, the brand-exclusion coordination call. Trademark safety is judgment; never delegate it.

STEP 1 — Brand-defense ad group(s)

• Keywords from brand_terms (own brand + variants + common misspellings), Exact + Phrase. No broad.
• Copy uses your own brand prominently, your USPs, and your real offers. Target high impression share at

low CPC (brand intent converts cheaply).

• RSA: up to 15 headlines (≤30) / 4 descriptions (≤90), keyword-led, with CTA.

STEP 2 — Competitor conquesting — optional, trademark-aware, a SEPARATE campaign

• Only if competitor_terms exist and the user wants it. **Emit it as its OWN campaign-spec, not an ad group

inside the brand-defense campaign — because bid strategy is campaign-level and conquesting needs Maximize Conversions**, not the defense campaign's Target Impression Share. (You do NOT want to bid to dominate a competitor's SERP at absolute-top 90% — expensive, low-intent.)

• Keywords from competitor_terms (Phrase/Exact).
• Trademark rules: do NOT put a competitor's trademark in the ad TEXT (headlines/descriptions/paths) —

only as keywords. Emit the competitor terms as a spec-level trademark_avoid[] so the validator BLOCKS any in ad text. Lead with YOUR differentiators ("wholesale pricing", "free shipping", a price extension with your real prices on the competitor SERP), not the competitor's name. Expect lower CTR/QS, higher CPC.

• Add YOUR brand_terms as negatives so conquesting never overlaps the brand-defense campaign.

STEP 3 — Negatives & separation (Search levels: campaign OR ad-group)

• Negatives can sit at campaign level OR ad-group level in Search (unlike PMax = campaign-only). Use

campaign-level for account-wide junk (job-seeker, DIY, location) and ad-group-level only where it does real work — here, negate brand_terms in the conquesting ad group so defense and conquesting never cross-serve. Don't duplicate the same negative at both levels; pick the narrowest level that achieves it.

• Keep brand-defense and conquesting ad groups cleanly separated.
• Extensions, likewise, can be campaign OR ad-group level in Search (PMax forces campaign-level). Default

to campaign-level; only push sitelinks/callouts down to ad-group level if the groups are different enough to need distinct ones (e.g. brand-specific vs conquesting sitelinks) — and only if it's worth the upkeep.

STEP 4 — Bidding (Search ≠ PMax — pick a brand-appropriate strategy, NOT a "learning ramp")

• Check what the account already does for brand first. If a brand campaign exists, read its

bidding_strategy_type (a healthy brand campaign commonly runs TARGET_IMPRESSION_SHARE) and match it — don't reinvent.

• Brand defense → Target Impression Share (absolute top of page, ~90%, with a max-CPC bid limit). The goal

is to dominate your own brand SERP cheaply, not to maximize value. Brand has history, so do NOT frame it as "maximize conversion value, no target, learning" — that's a cold-prospecting default, wrong for brand. (Maximize Conversions is an acceptable alternative if you'd rather optimize to conversion volume.)

• Conquesting → Maximize Conversions (watch CPA, don't over-invest). It's colder and pricier than brand.
• Bid strategy is CAMPAIGN-level. Brand defense and conquesting want different strategies, so once

conquesting has real budget, split it into its own campaign rather than forcing both under one strategy.

STEP 4b — Conversion goal (use the account's REAL action — never invent one)

Pull the enabled conversion actions (conversion_action.name, primary_for_goal) and use the account's actual primary purchase action by name — e.g. a store's only purchase action may be the store-channel purchase (Shopify's "Google & YouTube" channel tracks website purchases, Search-driven ones included), so the Search campaign optimizes on THAT, not a made-up name. Set conversion_goals.scope: "campaign". If no web-purchase primary exists for Search, flag it for the operator to create.

STEP 5 — Output + verification

Emit campaign-spec.json (branded_search, ad_groups[] as in builder-search) + blueprint.xlsx. Set campaign.status: "paused", conversion_goals.scope: "campaign". Build the extensions block the same way as builder-pmax STEP 7b — sitelinks, callouts, structured snippets (the store carries many brands, so a Brands snippet IS correct here, unlike a single-brand PMax), and a concrete prices asset built from real store data (don't leave it a placeholder — pull "From" prices per category and a live URL each; a price extension is especially strong on a brand/conquesting SERP). On a conquesting campaign, emit a trademark_avoid list (the competitor terms) at the spec level — the validator BLOCKS any ad text containing one, enforcing the no-competitor-TM-in-copy rule.

Render the human workbook from the spec (the shared renderer handles ad_groups → an AG sheet per ad group with keywords, RSA, negatives, paths): python ${CLAUDE_PLUGIN_ROOT}/skills/builder-pmax/scripts/spec_to_xlsx.py --output blueprint.xlsx

Verify with validate_spec.py: RSA limits/counts (Google-accurate counting, ${CLAUDE_PLUGIN_ROOT}/references/google-ads-formatting.md), keyword/negative match types (no broad), no trademark_avoid term in any ad text (enforced), brand_terms negated in conquesting groups, Final URLs live (200). Hand the JSON to pusher → Editor CSV (spec_to_editor_csv.py) + ready-to-copy blocks (spec_to_paste.py).

Coordination note for the operator

Branded Search + PMax brand exclusion are two halves of one decision: PMax excludes brand traffic, this campaign captures it deliberately. Surface this so reporting and credit stay clean.

To build / refine later

• [ ] Auto-generate brand misspellings; pull competitor domains from auction insights.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: chanktb
• Source: chanktb/claude-google-ads
• License: MIT
• Homepage: https://khuetran.com

Install and usage instructions live in the source repository linked above.