Doctrine

src/lib/adops/doctrine/google-ads.md · 11 MACHINE-CHECKABLE RULES

Google Ads — Best Practices Doctrine

Purpose. This is the playbook the audit engine diffs a live account against. The prose

sections encode account-management doctrine the LLM reasons over; the final RED FLAGS

section is a machine-checkable checklist the deterministic pre-check parses into guaranteed

recommendations. Marketers edit this file; the engine and the LLM both obey it.

---

Account structure

Google Ads has a fixed hierarchy: Account → Campaign → Ad Group → (Keywords + Ads). Structure is the single biggest lever on Quality Score, budget control, and reporting clarity — budget, bid strategy, geo and schedule live at the campaign level; relevance (keyword ↔ ad ↔ landing page) is decided at the ad group level.

• One theme per ad group. Every keyword in an ad group maps to one searcher intent so a

single ad set is genuinely relevant to all of them. Mixed-intent ad groups depress Quality Score.

• Segment campaigns by the dimension you must control independently — budget, geography,

vertical margin, brand vs. non-brand. Two keyword sets that need different budgets belong in different campaigns, not different ad groups.

• Separate brand and non-brand — always. Brand terms convert at high rates and low CPCs and

will flatter the blended numbers of prospecting campaigns, hiding poor performance. For an affiliate lead-gen book, this is doubly true: brand traffic is house money; non-brand is where margin is won or lost.

• STAG over SKAG. Group 3–15 tightly related keywords per ad group so Responsive Search Ads

have query variety to optimize against. Single-keyword ad groups splinter conversion data below the volume Smart Bidding needs to learn.

Budget & bid strategy

Match bid strategy to conversion volume — automated strategies need data to learn:

1. Manual CPC / Maximize Clicks — bootstrap only; migrate off as soon as conversions accrue. 2. Maximize Conversions (no target) — tracking is live but volume is thin. 3. Target CPA — only once a campaign reliably produces ~15–30 conversions / 30 days. 4. Target ROAS — needs conversion value tracking and ~30+ valued conversions / 30d.

Do not skip rungs. A low-volume campaign jumped straight onto tCPA starves the learning phase and produces volatile spend.

• Budget-limited is a signal, not a steady state. A campaign flagged "limited by budget"

while converting profitably is leaving volume on the table — raise the budget or reallocate from an underperformer. A budget-limited campaign that is NOT converting should be fixed (targeting, bids) before it gets another dollar.

• Daily budgets can overspend up to ~2× on a given day (Google averages to daily × 30.4).

Size against the monthly figure.

• Reallocate toward marginal CPA, not average CPA — move budget to campaigns where the

next dollar still clears the allowable, away from campaigns past their efficient frontier.

Keyword hygiene

• Broad match only with Smart Bidding and a strong negative list. Broad without negatives is

the classic lead-gen budget-waster (searchers wanting free, jobs, DIY).

• Shared negative lists are mandatory — a global junk list applied across campaigns, plus

brand negatives on non-brand campaigns and vice versa so the two never cannibalize.

• Mine search terms weekly on high-spend accounts. Converting queries become keywords;

wasteful queries become negatives.

• Prune zero-impression and long-term zero-conversion keywords — they add management surface

and mask the real term mix.

Ad copy

• ≥ 2–3 enabled RSAs per ad group is the durable floor; a single-ad ad group has no A/B

signal and goes dark if that ad is disapproved. An enabled ad group with zero serving ads is dead spend surface.

• 10–15 distinct headlines, ≥ 4 descriptions per RSA; include the theme keyword in a couple

of headlines. Pin only what compliance requires.

• Disapproved ads in enabled campaigns are emergencies — for regulated lead-gen verticals

(insurance, finance) a policy flag can silence a whole ad group overnight. Replace or appeal immediately.

• CTR is the canary. Search campaigns sustaining under ~1% CTR at meaningful volume have an

ad-relevance or query-mix problem — refresh copy and check the term report before touching bids.

Quality Score

Quality Score (1–10) = expected CTR + ad relevance + landing-page experience. Spending keywords at QS ≤ 3 overpay on every click and almost always signal a structural mismatch — wrong ad group, weak ad, or a generic landing page. Fix the structure, not the bid.

Conversion tracking

Everything above assumes conversions are measured. An account spending with zero recorded conversions is flying blind — either tracking is broken or the account is pure waste; both are critical. Verify at least one conversion action fired in the window before trusting any other number.

---

RED FLAGS

Machine-checkable checklist. Each bullet is one rule the deterministic pre-check

(redflags.ts) evaluates against the AdSpec snapshot. Grammar:

- rule_id — when: <condition> — recommend: <RecommendationType> — action: <slug> — risk: <low|medium|high> — why: <rationale>

recommend MUST be one of the nine canonical types

(budget_change | pause | enable | keyword_add | keyword_negative | bid_change | new_creative | structure | other).

Metrics are trailing-30-day unless suffixed. Thresholds here are the documented defaults.

• gads-no-conversion-tracking — when: account.cost_30d > 0 AND account.conversions_30d == 0 — recommend: other — action: add_conversion_tracking — risk: high — why: An account spending with zero recorded conversions is unmeasurable; this blocks every downstream optimization.
• gads-spend-zero-conversions — when: campaign.status == enabled AND campaign.cost_30d >= 0.20 * account.cost_30d AND campaign.conversions_30d == 0 — recommend: pause — action: pause_campaign — risk: high — why: A campaign consuming ≥20% of account spend with zero conversions is bleeding budget; pause and diagnose before it spends more.
• gads-disapproved-ad — when: campaign.status == enabled AND ad.policy_status == disapproved — recommend: new_creative — action: replace_disapproved_ad — risk: high — why: A disapproved ad in an enabled campaign silences delivery and, in regulated verticals, risks account-level policy strikes.
• gads-adgroup-no-active-ad — when: ad_group.status == enabled AND enabled_ad_count == 0 — recommend: new_creative — action: add_rsa — risk: high — why: An enabled ad group with no serving ad cannot show; it is dead spend surface.
• gads-adgroup-single-ad — when: ad_group.status == enabled AND enabled_ad_count == 1 — recommend: new_creative — action: add_rsa — risk: medium — why: One RSA gives no A/B signal and the ad group goes dark if that ad is disapproved.
• gads-zero-impression-keyword — when: keyword.status == enabled AND keyword.impressions_30d == 0 — recommend: pause — action: prune_keyword — risk: low — why: Zero-impression keywords add management surface and mask the real query mix; prune or restructure.
• gads-low-quality-score-spend — when: keyword.status == enabled AND keyword.quality_score <= 3 AND campaign.cost_30d > 0 — recommend: structure — action: restructure_ad_group — risk: medium — why: A spending keyword at QS ≤ 3 overpays on every click; fix the ad-group/ad/landing-page mismatch, not the bid.
• gads-premature-smart-bidding — when: campaign.bid_strategy IN (target_cpa, target_roas) AND campaign.conversions_30d < 15 — recommend: bid_change — action: change_bid_strategy — risk: medium — why: tCPA/tROAS need ~15–30 conversions/30d to stabilize; below that the learning phase starves and spend goes volatile.
• gads-budget-limited-converting — when: campaign.is_budget_limited == true AND campaign.conversions_30d > 0 — recommend: budget_change — action: increase_budget — risk: medium — why: A budget-capped campaign that is converting is leaving profitable volume unclaimed; raise or reallocate.
• gads-budget-limited-not-converting — when: campaign.is_budget_limited == true AND campaign.conversions_30d == 0 — recommend: other — action: fix_targeting_before_budget — risk: medium — why: Adding budget to a campaign that isn't converting scales waste; fix targeting first.
• gads-low-ctr-high-impressions — when: campaign.impressions_30d >= 10000 AND campaign.ctr_30d < 0.01 — recommend: new_creative — action: improve_ad_strength — risk: medium — why: Search CTR under 1% at ≥10k impressions signals weak ad relevance or a polluted query mix; refresh copy and mine the term report.