Adapted from the GoMarble Paid Media Claude Skills repository (https://github.com/gomarble-ai/gomarble-ai-paid-media-claude-skills) by its author, shared under GoMarble's attribution license. Works with any data source: a connected ads MCP such as GoMarble, CSV exports, screenshots, or pasted tables.

A consolidated set of skills for analyzing and optimizing Google Ads accounts. These skills are framework-agnostic - the analytical logic (Q1–Q5 query tiers, auction-pressure diagnosis, PMax maturity gates, item classification) applies regardless of where the data comes from.

The skills define what to know, what data to look for, how to interpret it, and how to recommend action. They do not depend on any specific tool integration.

Read-and-recommend only. These skills never make changes to the user's Google Ads account. Claude reads data and produces recommendations in plain language; the user applies the recommended changes themselves in the Google Ads UI.

---

Data Sources Are Not All Equal

Data can come from four kinds of sources, but they have very different completeness. Some skill rules silently break on weaker sources - most notably, PMax asset performance labels and PMax search term insights are NOT in default Google Ads CSV exports. Always identify the source first and consult the Data Source Compatibility Matrix below before applying any rule.

Priority	Source	Notes
1 (preferred)	A connected MCP server exposing Google Ads data (e.g. GoMarble) with GAQL access	All rules work. Query each Google Ads resource via GAQL.
2	CSV / Excel exports from Google Ads UI	Most basic rules work, but PMax assets, PMax search insights, and Quality Score components require separate UI reports the user often doesn't realise they need.
3 (narrow questions only)	Screenshots of report tables	Use only for spot questions on metrics visible in the screenshot. Skip deep analysis.
4 (last resort)	Direct copy-paste of metric values	Treat as user-asserted; ask for the underlying export when stakes are real.

---

How to Use These Skills

1. Identify the user's intent - are they asking about Search, Shopping, Performance Max, or keyword research? Use the routing table below.
2. Run the Data Inventory step (Step 0 below) - confirm what source is being used and what's actually available before invoking any skill.
3. Gather the required data - ask the user to share the data the skill needs. The "Minimum required data" section of each skill lists what's needed.
4. Apply the analytical framework - classify, diagnose, decide based on the rules in the relevant skill.
5. Always check guardrails before recommending any action.
6. Output recommendations in the structured format defined per skill.

If required data is missing or unverifiable, say so explicitly and ask the user for it. Never fabricate metrics or invent thresholds.

---

Step 0: Data Inventory (Mandatory Before Any Skill)

Run this once at the start of every analysis. It takes one short turn and prevents wasted effort - especially for PMax, where users often share a campaign CSV without realising they're missing the Asset Group Asset and Search Term Insights reports.

1. Identify the data source - MCP / CSV / screenshot / copy-paste. If MCP with GAQL access is connected, use it. If the user attached a file, inspect its columns/fields before asking for more.
2. State what's available and what's missing for the requested skill, based on the Data Source Compatibility Matrix below.
3. For each missing signal, decide:
   • Blocking - cannot proceed without it; ask user to fetch (e.g. Asset Group Asset report for PMax).
   • Downgrade - can proceed with caveats; flag as low-confidence in the output.
   • Skip - rule doesn't apply to this source (e.g. micros conversion on CSV data already in account currency); note it and move on.
4. Confirm account currency before showing any monetary value - never assume USD.
5. Confirm the date range in the data matches what the skill expects (30 days for Search; 14d or 30d for Shopping depending on volume; 30d for PMax with optional 60d for trend).

Output one short paragraph before applying the skill. Example:

"Using your GoMarble MCP via GAQL. For the PMax audit (Skill 3), I'll pull (1) campaign performance, (2) asset_group_asset for performance labels, (3) campaign_search_term_insight for PMax categories, (4) device + geo segments. Account currency confirmed as USD."

---

Data Source Compatibility Matrix

Use to plan which signals you can get from which source before running any skill.

Signal	MCP (GoMarble etc. with GAQL)	CSV (Google Ads UI exports)	Screenshot
Campaign performance (cost, conversions, conversion value, clicks, impressions, CTR, avg CPC)	✅ via standard metrics.* fields	✅ default Campaign report	✅ if shown
Cost in account currency	✅ but cost_micros returned by API - divide by 1,000,000	✅ already in account currency - skip the micros rule	✅
Average CPC	✅ in account currency (no conversion)	✅ in account currency	✅
Search Impression Share, IS Lost (Rank), IS Lost (Budget)	✅ via metrics.search_impression_share, search_rank_lost_impression_share, search_budget_lost_impression_share	✅ but column names vary ("Search lost IS (rank)" / "Impr. share lost (rank)") - match by meaning	⚠️
Top of Page Rate / Absolute Top of Page Rate	✅ via metrics.top_impression_percentage, absolute_top_impression_percentage	✅ default Search columns	⚠️
Search term report	✅ via search_term_view resource	✅ separate "Search Terms" report - not in the Campaigns export	❌
Keyword-level performance	✅ via keyword_view resource	✅ separate "Keywords" report	⚠️
Quality Score components (Expected CTR / Ad Relevance / Landing Page)	✅ via ad_group_criterion.quality_info.*	⚠️ requires Quality Score columns enabled in the Keywords report	❌
Bid strategy + target	✅ via campaign.bidding_strategy_type + target_cpa.target_cpa_micros / target_roas.target_roas	⚠️ requires Bid Strategy + Target columns enabled (often not default)	⚠️
Daily budget utilization	✅ via segments.date aggregation	⚠️ requires day-by-day segmented export	❌
Shopping item-level performance	✅ via shopping_performance_view	✅ separate "Products" report	❌
Merchant Center approval rate	❌ requires manual check in Merchant Center	❌ same	❌ same
PMax asset performance labels (BEST / GOOD / LOW / PENDING / UNSPECIFIED)	✅ via GAQL on asset_group_asset with performance_label	❌ NOT in default exports - requires Asset Report (separate UI report)	❌
PMax search term insights (aggregated categories)	✅ via GAQL on campaign_search_term_insight	❌ NOT in default exports - requires Search Terms Insights from the PMax campaign page	❌
PMax asset group structure (asset group ID, name, status)	✅ via asset_group	⚠️ partial - visible in PMax campaign page but not always in exports	⚠️
Device performance	✅ via segments.device	✅ separate "Devices" segment export	⚠️
Geographic performance	✅ via geographic_view	✅ separate "Locations" report	⚠️
Time-of-day / day-of-week	✅ via segments.day_of_week / hour	✅ separate "Time" segment export	⚠️
Auction Insights (competitor share)	⚠️ limited via API (auction_insight_domain)	✅ separate "Auction Insights" UI report	⚠️
Conversion type / value source	✅ via conversion_action resource + segments.conversion_action_category	⚠️ requires "Conversion action" segment + "All conv. value" enabled	❌
Keyword Planner data (volume, top-of-page bid, competition)	✅ via google_ads_keyword_discover / google_ads_keyword_metrics (GoMarble)	✅ via Keyword Planner export	❌

✅ = available and reliable. ⚠️ = available but requires extra export / column / segment enabled. ❌ = not available from this source.

---

Google Ads CSV Column Translation

When the source is a CSV export from the Google Ads UI, column names rarely match the API field names used throughout the skills. Use this table to translate. Column names vary slightly by language, account version, and export template - match by meaning, not exact string.

Skill / API reference	Typical Google Ads CSV column
metrics.cost_micros	"Cost" (already in account currency in CSV - micros rule does not apply)
metrics.conversions	"Conversions"
metrics.conversions_value	"Conv. value" or "All conv. value"
metrics.clicks	"Clicks"
metrics.impressions	"Impr."
metrics.ctr	"CTR"
metrics.average_cpc	"Avg. CPC"
metrics.search_impression_share	"Search impr. share"
metrics.search_rank_lost_impression_share	"Search lost IS (rank)" / "Impr. share lost (rank)"
metrics.search_budget_lost_impression_share	"Search lost IS (budget)" / "Impr. share lost (budget)"
metrics.top_impression_percentage	"Top impr. rate" / "Search top IS"
metrics.absolute_top_impression_percentage	"Abs. top impr. rate" / "Search abs. top IS"
metrics.value_per_conversion (ROAS = this / cost)	"Conv. value / cost" or compute as Conv. value ÷ Cost
campaign.bidding_strategy_type	"Bid strategy type" (must enable)
target_cpa.target_cpa_micros	"Target CPA"
target_roas.target_roas	"Target ROAS"
campaign.budget_amount_micros	"Budget" or "Daily budget"
ad_group_criterion.quality_info.quality_score	"Quality Score" (must enable in Keywords report)
ad_group_criterion.quality_info.creative_quality_score	"Ad relevance" (must enable)
ad_group_criterion.quality_info.post_click_quality_score	"Landing page experience" (must enable)
ad_group_criterion.quality_info.search_predicted_ctr	"Expected CTR" (must enable)
asset_group_asset.performance_label	❌ NOT in standard CSV - requires Asset Report from the PMax campaign
campaign_search_term_insight.*	❌ NOT in standard CSV - requires Search Terms Insights from PMax campaign page

Rule for Claude: if the source is CSV, costs are already in account currency - never apply the divide-by-1,000,000 rule. That rule only fires on raw API/MCP data.

---

Routing - Which Skill to Use

User asks about	Use Skill
Search campaigns, keywords, RSAs, text ads, search terms	Search Optimization
Shopping campaigns, product feed, Merchant Center, SKUs	Shopping Optimization
Performance Max, PMax, asset groups	PMax Optimization
Keyword research, new keyword ideas, search volume, CPC estimates, market expansion	Keyword Research
Change requests (create campaign, add keyword, change bid, pause, etc.)	Apply the relevant analysis skill first, then recommend the change in plain language with rationale. Claude does not execute changes on the user's ad account - the user applies recommendations manually in the Google Ads UI.

If the user's intent is ambiguous or spans multiple campaign types, ask which to focus on first.

---

Universal Guardrails - Apply Before Every Recommendation

These rules apply across every skill below. Read them before generating any recommendation.

Budget Reallocation Prohibition

You cannot allocate, shift, increase, reduce, or "optimize" budget for: keywords, search terms, match types, devices (unless via bid adjustments in Search), locations (unless via bid adjustments or exclusions), audiences (unless via bid modifiers), asset groups (PMax has no per-asset-group budgets).

Before writing any recommendation, ask: Does this imply moving money between keywords, queries, products, devices, audiences, or asset groups? If yes → rewrite using the valid controls below.

Valid budget controls:

Campaign Type	Valid Controls
Search / Shopping	Campaign daily budget, bid strategy targets (tCPA / tROAS / max CPC), pause or enable campaigns or ad groups
Performance Max	Campaign daily budget, bid strategy targets (tCPA / tROAS), pause asset groups or campaign

Mandatory rewrites:

• ❌ "Shift budget from keyword A to keyword B"

• ✅ "Pause keyword A. Increase campaign daily budget by 10–20% to give keyword B more eligible demand."

• ❌ "Allocate more to [placement / device / demographic / product / asset group]"

• ✅ "Create a new campaign targeting [dimension] with its own budget, or use bid adjustments / exclusions."

Metric Selection

First infer the account's conversion type:

• Ecommerce: Use ROAS, conversion value, cost per purchase. Never use CPA alone for revenue decisions.
• Lead gen: Use CPA, cost per lead. Never use ROAS unless offline revenue is imported.
• Footfall / awareness / local-service: Replace CPA/ROAS with cost per footfall action, cost per call, cost per direction, CPM, reach, frequency. State explicitly that purchase-derived CPA/ROAS do not apply.
• Unclear: Ask the user. Do not assume.

Scaling Limits

• Max budget increase: 20% per move for Search/Shopping; up to 50% per move for mature PMax.
• Never scale if: CPA > target, ROAS < target, or query / search term hygiene not done.
• Never scale based on: impression share alone, CPC alone, or early data (< 7 days).
• Frame scaling as: "incremental test with downside risk."

Learning & Stability

• Do not stack changes (budget + bids + structure simultaneously).
• Do not judge performance on: < 30 conversions for Search, < 50 conversions for PMax/Shopping.
• Avoid changes that reset learning unless absolutely required.

Immediate STOP Triggers

Halt any recommendation that attempts:

• "Scale keywords / products / asset groups"
• "Reallocate spend between queries"
• "Optimize placements in PMax"
• "Increase bids and budget at the same time"
• "Scale > 20% in one step"
• "Fix low impression share by increasing budget blindly"

Required Diagnostics Before Any Action

Before any scale/pause/restructure suggestion, confirm you know:

1. Search term quality (presence of Q4 / Q5 queries - defined in the Search skill)
2. CPA / ROAS vs target
3. IS Lost split: Rank vs Budget
4. CPC trend direction
5. Conversion volume sufficiency
6. Bidding strategy compatibility

If diagnostics are incomplete → say insufficient data, ask for what's missing, and do not recommend.

---

Metric Glossary - Canonical Definitions

Used across every skill below.

Currency Note: Micros (API/MCP only)

This rule applies only to raw API / MCP data. If the source is a CSV export from the Google Ads UI, costs are already shown in account currency - skip this rule.

When data is pulled from API exports (MCP / GAQL), cost and bid fields are often returned in micros (1,000,000 micros = 1 unit of account currency). If you see fields named cost_micros, cpc_bid_micros, or amount_micros, divide by 1,000,000. average_cpc is typically already in account currency.

Source check before applying:

• MCP returning *_micros fields → divide by 1,000,000.
• CSV with a "Cost" or "Avg. CPC" column → already in account currency; do nothing.
• Screenshot showing a currency symbol → already converted; trust it.

Always confirm account currency before showing monetary values - never assume USD.

Direct Metrics

Metric	Notes
Cost	Total spend in account currency
Conversions	Includes all configured conversion actions unless filtered
Conversion value	Revenue for value-based conversions
Clicks
Impressions
Average CPC	Already in account currency
CTR	Decimal (0.05 = 5%)
Search IS	Search impression share
IS Lost (Rank)	Share of impressions lost due to ad rank (quality + bid)
IS Lost (Budget)	Share of impressions lost because campaign hit daily budget
Top of Page Rate	Share of impressions appearing above organic results
Absolute Top Rate	Share of impressions appearing as the first ad

Calculated Metrics

Metric	Formula
CPA	cost / conversions
ROAS	conversion value / cost (decimal multiple, e.g. 3.5 = 3.5×)
CVR	conversions / clicks × 100
AOV	total conversion value / total conversions
CPC trend WoW	(this_week_cpc − last_week_cpc) / last_week_cpc × 100
CPC trend MoM	(this_month_cpc − last_month_cpc) / last_month_cpc × 100
Spend share	entity_cost / total_cost × 100

Account-Type Inference

If the user hasn't stated account type, infer from data:

• Ecommerce: Purchase conversion actions present AND non-zero conversion value → use ROAS, AOV, CPA.
• Lead gen: Conversions present AND conversion value = 0 → use CPA only; ROAS unreliable unless offline conversion import is configured.
• Mixed: Group campaigns by primary conversion action and apply each group's metrics separately.
• Unclear: Ask the user. Don't default.

Maturity & Statistical-Significance Gates

Gate	Threshold	Rationale
Smart Bidding minimum (Search)	≥ 30 conversions / 30 days	Below this, tCPA/tROAS lacks signal
Smart Bidding minimum (PMax / Shopping)	≥ 50 conversions / 30 days	Wider channel mix needs more signal
PMax maturity gate	Age ≥ 30 days (ideally 60) OR conversions ≥ 50	Required before any PMax optimization
Segment-level recommendation	≥ 50 clicks AND ≥ 5 conversions	Below this → flag as low-confidence
Scaling step cap	≤ 20% per move (Search/Shopping); ≤ 50% per move on mature PMax
Lookback for Shopping	14 days if ≥ 100 purchases/month, else 30 days

PMax vs Search ROAS Comparison

Apply only after the PMax maturity gate passes.

PMax ROAS Ratio	Assessment	Default Action
< 0.5× Search ROAS	Poor	Monitor closely; consider pausing if not improving
0.5 – 0.9× Search ROAS	Below par	Optimize assets and negatives before any scaling
≥ 0.9× Search ROAS	Healthy	Optimize; eligible for scaling
> Search ROAS	Outperforming	Prioritize scaling

PMax Asset Performance Labels

Google labels each PMax asset (it does NOT expose cost/conversion data per asset).

Label	Interpretation	Default Action
BEST	Top performer in its asset group	Create variations with similar style
GOOD	Acceptable	Keep, monitor
LOW	Underperforming	Remove; replace with BEST variations
PENDING / UNSPECIFIED	Insufficient data	Wait 14+ days

Shopping Item Classification (used in the Shopping skill)

Classification	Trigger
KILL	cost ≥ 2× AOV AND conversions = 0
DOWNGRADE	ROAS < 0.7× target AND cost ≥ 0.5× AOV
PROMOTE	conversions ≥ 2 AND ROAS ≥ target

Quality Score Components (when CPC is high)

Decompose Quality Score into:

• Expected CTR
• Ad relevance
• Landing page experience

Identify which component is dragging and recommend fixes for that component. Do not recommend bid changes alone to compensate for low Quality Score.

---

Depth-of-Analysis Framework - Apply on Audits & Optimization Reviews

Apply on: audits, performance reviews, optimization requests, strategy questions, account restructuring. Skip for: simple data fetches, campaign listings, single-metric lookups.

1. Hierarchical Drill-Down

Never recommend on campaign-level aggregates alone. Drill deeper:

• Asked about a campaign → check ad group performance within it.
• Asked about an ad group → check keyword + search term performance within it.
• Asked about PMax → check asset group performance, search term categories, URL expansion config.

Quantify the driver: "Ad Group X accounts for 65% of campaign spend but CPA is 2× the campaign average."

2. Objective-Contextual Evaluation

Evaluate each campaign against its bid strategy target, not just absolute performance.

Bid Strategy	Evaluate against	Red flag
tCPA	Actual CPA vs target CPA	CPA 30%+ above target
tROAS	Actual ROAS vs target ROAS	ROAS 30%+ below target
Maximize Conversions (no target)	CPA trend direction	Rising CPA with no constraint
Manual CPC	CPC vs Quality Score implied CPC	Paying premium due to low QS

Flag unconstrained strategies: if tROAS target is 0.5× on a campaign achieving 3×, the algorithm isn't actually constrained. If tCPA is 3× above actual CPA, same problem. Call it out.

3. Breakdown & Segmentation Analysis

Before concluding, check at least 2 of these dimensions:

• Device performance (mobile vs desktop CPA gap)
• Search Partners conversion rate vs Search network (quantify the gap)
• Geographic variations (regions with high spend and poor CVR)
• Day-of-week patterns
• Cross-reference bid adjustments against actual segment performance

Statistical significance: Do not make segment-level recommendations from < 50 clicks or < 5 conversions. Flag as low-confidence.

4. Temporal & Trend Analysis

Never treat the date range as a single number. Compare:

• Last 7 days vs previous 7 days (WoW)
• Separate day-of-week seasonality from structural trends
• Identify direction for each key metric: improving, declining, stable

Distinguish: "metric is bad" vs "metric is getting worse" - direction matters more than absolute value.

Connect the chain: impression share declining → CPC increasing → competitive pressure, not quality degradation.

5. Cross-Entity Dependency Recognition

• Brand Search ↔ PMax overlap: Is PMax capturing brand terms? Check PMax search term categories.
• Non-brand Search → Remarketing: Does non-brand search activity feed remarketing lists?
• Shopping ↔ Search overlap: Same products advertised in both? Check for cannibalization.

Flag dependencies: "PMax and Brand Search are both converting on brand terms. Total reported conversions across both campaigns likely exceed actual unique conversions. Before scaling PMax based on its ROAS, factor in cannibalization from Brand Search."

6. Platform Mechanic Awareness

Signal	Mechanic	Consequence
Broad match + rising CPC	Query expansion	Algorithm widens to less relevant queries → QS drops → CPC rises → tCPA struggles
Low QS + high CPC	Ad relevance	Paying premium for same position - fix relevance, not just bid
tCPA + low conversion volume	Smart Bidding data	Insufficient signal - consider Manual CPC
PMax + declining Brand Search IS	Brand cannibalization	PMax claiming brand conversions → inflated PMax ROAS
Maximize Conversions + no target	Unconstrained spend	Algorithm spends full budget at any CPA - add a tCPA target

7. Proactive Signal Detection

After answering the user's question, surface 2–3 additional concerns or opportunities they didn't ask about.

Format: "Outside your question, I noticed [signal] in [entity] - this suggests [predicted consequence] within [timeframe]."

8. Attribution & Measurement Rigor

Don't take reported conversion numbers at face value.

Always check:

• Attribution model in use (last click, data-driven, etc.)
• Conversion lag - are recent days' conversions still incomplete?
• Share of conversions that are estimated (modeled) vs observed
• Mix of micro- vs macro-conversions

Frame as caveat: "Before scaling based on the reported 5× ROAS, note that [attribution concern] means the true incremental ROAS is likely lower."

---

Skill 1: Search Campaign Optimization

A systematic approach to Google Ads Search via query classification, auction-pressure diagnosis, and rule-based decision logic.

Workflow: Data → Classification → Pressure Analysis → Decision → Action.

When to use this skill

• Audit or optimize a Search campaign
• Diagnose why CPA / ROAS / impression share is changing
• Decide whether to scale, hold, or cut a Search campaign
• Identify wasted spend in search terms

Data to gather

Ask the user to share - or pull from whatever data source is available - the following for the relevant Search campaigns over the last 30 days, filtered to ENABLED campaigns only.

Campaign-level auction metrics:

• Search impression share
• IS Lost (Rank)
• IS Lost (Budget)
• Top of Page Rate
• Absolute Top of Page Rate
• Average CPC
• Cost, conversions, conversion value, clicks, impressions

Search-term report: Each search term with impressions, clicks, cost, conversions, conversion value, sorted by cost descending (top 200 by spend).

Keyword-level data: Keyword text, match type, keyword impression share, IS Lost (Rank), avg CPC, conversions, cost.

CPC trend: Average CPC for the last 7 days AND the prior 7 days, so WoW change can be calculated.

User inputs to request (if not already known):

• Target CPA (or target ROAS for ecommerce). Default: historical 30-day average.
• Brand name (so brand vs non-brand terms can be classified)
• Competitor names (optional)
• Priority geographies (optional)

Step 1: Classify Each Search Term (Q1 – Q5)

Tier	Name	Condition
Q1	Profitable	Conversions ≥ 2, OR CPA ≤ target, OR ROAS ≥ target
Q2	Promising	Conversions = 0 AND cost < 0.5× target CPA AND no Q5 signals
Q3	Unproven	Conversions = 0 AND cost between 0.5×–1× target CPA
Q4	Losing	Cost ≥ 1× target CPA with 0 conversions, OR CPA ≥ 1.5× target, OR ROAS ≤ 0.7× target
Q5	Invalid	Query contains: free, cheap, diy, how to, job, salary, course, pdf, meaning, used, repair

Q5 terms are eliminated regardless of cost. Calculate spend share per tier.

Intent signal patterns (for classifying Q1 vs Q5 when ambiguous):

• High intent (→ Q1): buy, purchase, order, pricing, price, cost, near me, delivery, shipping, discount, coupon, subscribe, sign up, get started, demo, trial, quote, consultation, book, schedule
• Mid intent (→ Q2 / Q3): best, top, review, vs, versus, compare, alternative, recommendation, worth it
• Low intent (→ Q4 / Q5): what is, how to, tutorial, guide, learn, definition, meaning, example
• Negative intent (→ Q5): free, diy, repair, fix, used, second hand, jobs, career, salary, template, sample

Step 2: Classify Auction Pressure

Rank Pressure

Rank Pressure = HIGH if any true
IS Lost (Rank) ≥ 30%
Search IS ≤ 50% AND IS Lost (Rank) > IS Lost (Budget)
Top of Page Rate ≤ 60% on Q1 queries
Absolute Top ≤ 20% on Q1 queries

Rank Pressure = LOW if all true
IS Lost (Rank) ≤ 20%
Top of Page Rate ≥ 70%
Absolute Top ≥ 30% on Q1 queries

Budget Pressure

Budget Pressure = HIGH if all true
IS Lost (Budget) ≥ 20%
IS Lost (Budget) > IS Lost (Rank)
CPA ≤ target CPA
≥ 60% of spend from Q1–Q3

Budget Pressure = LOW if any true
IS Lost (Budget) < 10%
CPA > target CPA
Q4 + Q5 spend ≥ 25%

Step 3: Decision Matrix

Query Mix	Rank Pressure	Budget Pressure	Action
Q1-heavy	Low	High	Increase budget 10–20%
Q1-heavy	High	Low	Increase bids 5–10% OR improve ad relevance
Q1-heavy	High	High	Fix rank first, then budget
Q1-heavy	Low	Low	Hold - no changes needed
Q2–Q3 heavy	Low	High	Limited budget test
Q4+ present	High	Any	Cap or cut spend
Q5 present	Any	Any	Negate immediately

Search IS alone: Low Search IS means nothing on its own. Only act when Q1 share ≥ 50% AND IS Lost (Rank or Budget) explains it.

IS Lost (Rank) logic:

• If majority Q1, CPA ≤ target, Rank Pressure HIGH → increase bids 5–10% or improve ad relevance.
• If CPA > target → do NOT raise bids. Instead tighten queries: promote Q1 to Exact, kill Q3/Q4 leakage.

IS Lost (Budget) logic:

• If CPA stable, Q1–Q3 share ≥ 60%, Budget Pressure HIGH → increase budget 10–20%.
• Otherwise fix queries first.

Top / Absolute Top:

• Use only on Q1 queries.
• If CVR at Top ≥ 1.15× rest → defend the position.
• No CVR delta = ego tax. Ignore.

Step 4: CPC Inflation Diagnosis

When CPC is rising WoW:

Symptom	Root Cause	Action
CPC ↑ AND IS Lost (Rank) ↑	Competition increased	Bid increase allowed ONLY if CPA ≤ target
CPC ↑ AND new Q4/Q5 terms appearing	Match type leakage	Add negatives, tighten match types
CPC ↑ AND CTR ↓	Ad relevance degraded	Improve ad copy, NOT bids

Only competition justifies bid increases. Cases 2 and 3 require query or creative fixes.

Step 5: Scaling Prerequisites

Scale ONLY if all true:

• Q1 queries isolated (Exact or tight Phrase match)
• CPA stable or improving over 14+ days
• IS Lost (Budget) ≥ 20%
• CPC WoW% increase < CVR WoW% increase (efficiency improving faster than costs)

If CPC rising faster than CVR → false scale signal. Do not increase budget.

Step 6: Brand & Competitor Logic

Brand defense (when brand CPC ↑, Absolute Top ↓, competitor overlap ↑):

• Raise bids only on exact-match brand terms
• Tighten match types on brand campaigns
• Add competitor terms as negatives in non-brand campaigns

Competitor campaigns: Default state is OFF. Enable only if CPA ≤ 1.2× target AND Rank Pressure = LOW. Otherwise pause.

Recommendations to Make

When the user is ready to act, recommend the change in plain terms. The recommendation itself is the deliverable - the user applies the change manually in the Google Ads UI. Examples:

• "Increase the daily budget on Campaign X from $Y to $Y × 1.15 (15% increase). Hold here for 14 days and re-evaluate."
• "Raise the keyword-level max CPC on these 5 Q1 keywords by 5–10%. Keep the other keywords' bids unchanged."
• "Add the following terms to the campaign's negative keyword list: free, diy, how to, repair, [...]"
• "Create a new ad group containing the top 8 Q1 search terms as Exact-match keywords. Add those same terms as negatives in the original ad group to prevent overlap."

Weekly Execution Order (immutable)

1. Prune Q4/Q5 queries
2. Reclassify remaining queries
3. Recompute auction pressure
4. Make bid / budget moves
5. Fix ad relevance

Prohibitions

• Never increase budget to fix low IS without classifying queries first
• Never increase bids with Q4/Q5 terms present
• Never chase Absolute Top without CVR proof
• Never let Broad match dominate auction signals
• Never increase bids AND budget simultaneously

Output Format

### Current State
- Total spend: $X over 30 days
- Query distribution: Q1 (X%), Q2–Q3 (Y%), Q4–Q5 (Z%)
- CPA: $XX (Target: $XX) - [XX% over/under target]
- Rank Pressure: HIGH / LOW
  - IS Lost (Rank): X%
  - Top of Page Rate: X%
  - Absolute Top: X%
- Budget Pressure: HIGH / LOW
  - IS Lost (Budget): X%
  - CPA vs Target: [status]
  - Q1–Q3 spend: X%

### Query Classification Summary
- Q1 (X% of spend): top profitable terms
- Q2–Q3 (Y% of spend): top promising/unproven terms
- Q4–Q5 (Z% of spend): terms to negate; estimated savings $X/day

### CPC Inflation Analysis
- Current CPC: $X
- Prior period CPC: $X
- Change: X% WoW
- Diagnosis: [Case 1 / 2 / 3]
- Root cause: [...]

### Recommended Action (from decision matrix)
Execution order:
  1. ...
  2. ...
  3. ...

Tactical moves:
  - Bid/budget adjustments with entity references
  - Negatives to add (list)
  - Isolation recommendations

### Expected Outcomes (7-day horizon)
- CPA: $XX → $XX
- IS Lost (Rank): X% → X%
- IS Lost (Budget): X% → X%
- Q1 spend share: X% → X%

Minimum required data (by source)

Source	What you need
MCP	Last 30 days, filtered to ENABLED Search campaigns: (1) campaign-level metrics (cost, conversions, conv. value, clicks, impressions, search IS, IS Lost Rank/Budget, Top of Page Rate, Abs Top Rate, avg CPC) via the campaign resource; (2) search-term performance via search_term_view (top 200 by cost); (3) keyword-level via keyword_view; (4) CPC trend by segmenting last 7d vs prior 7d.
CSV	Requires three separate exports from the Google Ads UI: (1) Campaign Performance report with auction metrics columns enabled; (2) Search Terms report (top 200 by cost); (3) Keywords report. Plus a week-over-week breakdown if not in the campaign export.
Screenshot / copy-paste	Workable only for a quick sanity-check on one campaign. The Q1–Q5 classification needs the full search-term list.

Tool hints

The Q1–Q5 classification engine needs the full search-term list, not just top performers. With MCP, query search_term_view and paginate to 200 rows by cost desc. With CSV, the user must run a separate Search Terms export - not the Campaigns export. Auction metrics (Search IS, IS Lost Rank/Budget, Top of Page Rate, Abs Top Rate) are in the Campaigns export but require the auction columns to be enabled - many users miss this and share a CSV with only spend / conversions.

---

Skill 2: Shopping Campaign Optimization

Product-led optimization: validate feed → classify items → manage structure → scale or cut.

When to use this skill

• Audit or optimize Shopping / Performance Max for Retail campaigns
• Identify wasted spend at the SKU level
• Decide which products to promote, downgrade, or kill
• Diagnose why Shopping ROAS is below target

Step 0: Feed Health (BLOCKING)

Before any optimization, verify:

• ≥ 80% of products approved in Merchant Center
• Purchase conversion tracking is active (conversion value > 0 in data)
• Prices and availability are accurate (requires Merchant Center check or manual verification)

If any check fails → STOP all optimization. Fix the feed first. Ask the user to confirm Merchant Center status before continuing.

Data to gather

Lookback window: 14 days if the account has ≥ 100 purchases/month, otherwise 30 days.

Ask the user to share - or pull from whatever data source is available:

Account-level AOV: total conversion value ÷ total conversions over the lookback window.

Item-level performance: Per product: product ID, product title, brand, cost, conversions, conversion value, clicks, impressions. Sort by cost descending (top 200).

Campaign-level: Campaign ID, name, status, daily budget, cost, conversions, conversion value, filtered to Shopping campaigns and ENABLED status.

Search terms: Search term, cost, clicks, conversions for the Shopping campaigns (top 200 by cost).

Daily budget utilization: Daily cost over the last 14 days for each Shopping campaign.

Ask for target ROAS. Default: historical 30-day ROAS.

Step 1: Item-Level Classification

Sort items by cost (highest first). For each item:

Classification	Condition	Action
KILL	Cost ≥ 2× AOV AND conversions = 0	Exclude item ID immediately. No learning window.
DOWNGRADE	ROAS < 0.7× target AND cost ≥ 0.5× AOV	Lower bid or exclude. Spending but underperforming.
PROMOTE	Conversions ≥ 2 AND ROAS ≥ target	Isolate into its own product group / campaign to protect budget.

Items not matching any rule: monitor, no action needed.

Step 2: Product Group Structure

• Maximum 3 product groups per campaign.
• Split ONLY when economics differ (different margins, price bands, seasonal risk).
• If groups don't have meaningfully different ROAS targets → consolidate.

Step 3: Search Term Negatives

Add negatives ONLY for wrong-intent terms:

• Price objection: free, cheap, budget, discount code, coupon
• DIY / repair: diy, repair, fix, how to, tutorial
• Used / second-hand: used, second hand, refurbished, pre-owned
• Products not in feed: compare search terms to product titles; negate terms for products you don't sell
• Irrelevant attributes: colors / sizes not offered

Do NOT negate high-intent commercial terms even if they haven't converted yet.

Step 4: Budget & Bid Decisions

SCALE (increase budget 10–20%)

All must be true:

• Campaign ROAS ≥ target
• Budget utilization ≥ 90% on 10+ of the last 14 days
• Top 20% of SKUs by conversions account for ≥ 60% of spend (winners are getting the budget)

CUT / CAP (reduce budget or exclude SKUs)

Any true:

• Campaign ROAS < target
• SKUs with ROAS < 0.7× target receiving increasing spend share WoW

→ Reduce budget OR exclude underperforming SKUs first, then reassess.

Step 5: Device & Location (last)

Only analyze if segment spend ≥ 2× AOV (enough data to judge).

• Exclude device/location ONLY if ROAS ≤ 0.6× account average AND no operational explanation (e.g. mobile site is broken).
• Check for technical issues before excluding (site speed, checkout flow).

Step 6: Campaign Kill Conditions

Pause the Shopping campaign if:

• ROAS < 0.6× target after total spend ≥ 2× AOV
• Shopping ROAS consistently worse than Search ROAS over 30+ days

Prohibitions

• Never optimize by CTR alone - Shopping is ROAS-driven.
• Never let cheap, low-converting SKUs dominate spend.
• Never scale without confirming winners are getting the budget.

Output Format

### Feed Health
- Merchant Center approval rate: X%
- Conversion tracking: [active / inactive]
- Status: [OK to proceed / blocked]

### Item Classification Summary
- KILL: N items, $X estimated monthly savings
- DOWNGRADE: N items, $X estimated savings
- PROMOTE: N items, suggested isolation strategy
- Monitor: N items

### Search Term Negatives
- Recommended additions: [list]
- Estimated savings: $X/day

### Budget Recommendation
- Current: $X/day
- Recommended: $X/day (+/- X%)
- Rationale: [SCALE / CUT / HOLD with reasons]

### Expected 7-day ROAS Impact
- Current ROAS: X.XX (Target: X.XX)
- Projected ROAS after kills/negatives: X.XX

Minimum required data (by source)

Source	What you need
MCP	(1) Account-level AOV (total conv. value / total conversions over lookback). (2) Item-level performance via shopping_performance_view (top 200 by cost). (3) Campaign-level Shopping metrics filtered to ENABLED. (4) Search terms for Shopping campaigns. (5) Daily cost segmentation for last 14 days via segments.date. (6) Merchant Center approval rate - manual user check, not in MCP.
CSV	Three separate exports: (1) Campaign Performance for Shopping campaigns; (2) Products report for item-level performance; (3) Search Terms report for Shopping. Plus a daily-segmented export for budget utilization. Merchant Center approval rate is a manual check the user must do in Merchant Center.
Screenshot / copy-paste	Not workable for full Shopping audit - too many entity-level joins required.

Tool hints

The feed-health gate (Step 0) blocks all optimization if Merchant Center approval rate is unknown. This is never in MCP or CSV - always ask the user to confirm. The Products report at item level is the second most-missed export (after PMax assets) - many users share a Campaigns export and ask for product-level recommendations that can't be made from it.

---

Skill 3: Performance Max (PMax) Optimization

PMax is machine-learning-led: multi-channel (Search, Display, YouTube, Gmail, Discover), asset-driven, with limited granular control. It requires maturity before optimization.

When to use this skill

• Audit or optimize an existing PMax campaign
• Decide whether PMax is worth keeping vs. shifting spend to Search/Shopping
• Plan scaling for a profitable PMax campaign
• Diagnose why PMax ROAS is below Search ROAS

Maturity Gate (BLOCKING)

Campaign must meet one of:

• Age ≥ 30 days (ideally 60)
• Total conversions ≥ 50

If neither met → WAIT. Do not optimize. ML has not stabilized.

Exception: if the campaign is clearly burning money (ROAS < 0.3× target after spend ≥ 3× AOV), recommend pausing - even pre-maturity.

Data to gather

Ask the user to share - or pull from whatever data source is available - over the last 30 days (use 60 days for trend analysis):

Campaign performance: Campaign ID, name, status, start date, impressions, clicks, CTR, cost, conversions, conversion value. Filter to PMax campaigns.

Asset group details: Campaign ID, asset group ID, name, status.

Asset performance labels: For each asset (image, video, headline, description) within each asset group, the performance label: BEST / GOOD / LOW / PENDING / UNSPECIFIED.

Device performance: Cost, conversions, conversion value, clicks, impressions per device (mobile / desktop / tablet) for each PMax campaign.

Geographic performance: Cost, conversions, conversion value per location (top 200 by cost).

Time series: Daily cost, conversions, conversion value over the last 60 days for trend analysis.

Search term insights (PMax-aggregated): Category labels, clicks, impressions, conversions, cost. Note: PMax provides aggregated categories only, not individual terms.

PMax vs Search comparison: Cost, conversions, conversion value for both PMax AND Search campaigns over the same window.

Ask for: Target ROAS, target CPA, initial budget (for scaling caps), priority locations.

Phase 1 - Evaluate Performance

Step 1: PMax vs Search Comparison

Calculate ROAS for each channel type and compare:

PMax ROAS Ratio	Assessment	Action
< 0.5× Search ROAS	Poor	Monitor closely; consider pausing if not improving
0.5 – 0.9× Search ROAS	Below par	Optimize assets and negatives before any scaling
≥ 0.9× Search ROAS	Healthy	Proceed with optimization; eligible for scaling
> Search ROAS	Outperforming	Prioritize scaling

Also check for brand cannibalization: review PMax search term categories. If brand terms appear, correlate PMax conversion growth with Brand Search impression share decline over the same window. Flag inflated PMax ROAS.

Step 2: Asset Performance Analysis

Group by asset type (images, videos, headlines, descriptions). For each asset:

Label	Action
BEST	Create variations with similar style/messaging. This is working - make more like it.
GOOD	Keep, monitor. No changes needed.
LOW	Remove from asset group. Replace with variations of BEST performers.
PENDING / UNSPECIFIED	Wait for more data (typically 14+ days).

Step 3: Device & Location

Device:

Device CPA vs Account	Action
> 1.5× account CPA	Check site/funnel usability on that device first. If no issues → consider -50% bid adjustment or exclusion.
> 2× account CPA	Strong candidate for exclusion after confirming no technical issues.
CVR < 0.5× account CVR	Check landing page experience on that device.

Location:

Location CPA vs Account	Action
> 1.5× account CPA	Consider excluding
> 2× account CPA	Exclude
Spend but 0 conversions	Wait for more data, OR exclude if spend ≥ 2× AOV

Always check for technical/usability issues before excluding a device or location.

Insights Tab Review (qualitative)

If the user has access to the PMax Insights tab in the Google Ads UI:

• Audience Insights: which audience types are getting majority of clicks (cannot see conversion performance directly - infer by click volume).
• Asset Insights: overlap of assets and audiences. Use this to plan new asset groups for high-performing segments.

Phase 2 - Scale (only if Phase 1 confirms profitability)

Scaling Prerequisites

All must be true before scaling:

• Campaign profitable for 14–30 consecutive days
• ROAS ≥ target (default 300% if not specified)
• CPA ≤ target
• Performance trends stable or improving WoW

Budget Scaling

Increase daily budget by 20–50% every 14–30 days.

Scenario	Action
Profitable 14+ days, trends stable	Increase 20–50%
Post-scaling CPA > 1.2× target	PAUSE scaling, optimize assets/negatives, reassess after 30 days
Post-scaling ROAS < 0.8× target	PAUSE scaling, optimize, reassess after 30 days
Consistently profitable 90+ days	Can exceed 2–3× initial budget cap

Hard cap: do not exceed 2–3× initial monthly budget until 90+ days of consistent profitability.

Location Expansion

• Only after core locations profitable for 30+ days.
• Expand to similar / adjacent markets.
• Create separate asset groups for new locations if targeting differs.
• Monitor new locations independently for 30 days before further expansion.

Audience Expansion

• Start with Observation mode for new audience signals.
• Monitor performance for 14+ days.
• Switch to Targeting mode only if performance holds.
• Do NOT force audience preferences without data.

Asset Expansion

Based on BEST-performing assets from evaluation:

Asset Type	Expansion Rule
Images	Add 5–10 total, similar style to BEST, different angles/contexts
Headlines	Variations of BEST - same theme, different wording. Test different CTAs.
Descriptions	Variations of BEST - same value props, different framing
Videos	Test 30-second formats if shorter videos are BEST
New asset groups	For high-performing audience segments identified in Insights

PMax Limitations (do NOT recommend these)

• No keyword targeting
• No budget per asset group
• No placement control
• No demographic control
• No channel-level budget control
• Search term data is aggregated categories only, not individual terms

Prohibitions

• Never optimize before 30 days OR 50 conversions (unless burning money)
• Never scale before 14 days of profitability
• Never increase budget > 50% in one step
• Never scale beyond 2–3× initial budget before 90 days of profitability
• Never continue scaling if CPA > 1.2× target OR ROAS < 0.8× target
• Never remove BEST or GOOD performing assets
• Never force device/location exclusions without checking for technical issues first

Output Format

### Current State
- Campaign age: X days (Ready: Yes/No)
- Total conversions: X (Minimum met: Yes/No)
- Status: [Ready for optimization / Still stabilizing]

### Performance vs Search
- PMax ROAS: X.XX (Target: X.XX)
- Search ROAS: X.XX
- Ratio: X.XX (PMax is XX% of Search performance)
- Assessment: [Outperforming / Acceptable / Underperforming / Poor]

### Asset Group Analysis
- BEST assets: [list with action - create variations]
- GOOD assets: [list - keep, monitor]
- LOW assets: [list - remove and replace]
- PENDING assets: [list - wait]

### Device Performance
- Per-device CPA vs account; flagged exclusion candidates

### Location Performance
- Top locations by spend; flagged exclusion candidates

### Scaling Recommendation
- Days profitable: X
- Recommendation: [SCALE / OPTIMIZE FIRST / PAUSE]
- If SCALE: current $X/day → recommended $X/day (+X%); cap at $X/month
- If OPTIMIZE FIRST: reason and required fixes
- If PAUSE: reason

### Expansion Opportunities (if applicable)
- Locations, audiences, asset variations to test

### Expected Outcomes (30-day horizon)
- CPA: $X → $X
- ROAS: X.XX → X.XX
- Monitoring priorities

Minimum required data (by source)

PMax has the largest data-source gap of any skill. Many users share a campaign-level export and ask for asset and search-term recommendations that simply cannot be made from it.

Source	What you need
MCP (with GAQL)	(1) Campaign performance over 30d (60d for trend) via campaign; (2) Asset group structure via asset_group; (3) Asset performance labels via GAQL on asset_group_asset with performance_label field; (4) Search term insights via GAQL on campaign_search_term_insight; (5) Device segment via segments.device; (6) Geographic segment via geographic_view; (7) PMax vs Search comparison - pull both with same date range.
CSV	Requires four to five separate exports that users frequently miss: (a) Campaign Performance for PMax campaigns; (b) Asset Report from the PMax campaign page (gives BEST / GOOD / LOW labels); (c) Search Terms Insights from the PMax campaign page (gives aggregated categories - NOT the regular Search Terms report); (d) Devices segment; (e) Locations segment. Plus a Search-campaigns export with same window for the PMax vs Search ROAS comparison.
Screenshot / copy-paste	Not workable for full PMax audit. Possibly workable for a single yes/no question (e.g. "is my PMax mature enough to optimize?").

Before running this skill on CSV data, explicitly check whether the user has the Asset Report and Search Term Insights. If they don't, ask them to pull both before continuing - or fall back to the MCP path.

Tool hints

PMax's "limitations" cut both ways: the platform exposes less, but Google Ads API still has the resources Claude needs. The two resources that turn an incomplete PMax audit into a complete one are asset_group_asset (for performance labels) and campaign_search_term_insight (for the aggregated category data). Both require GAQL. If the connected MCP supports arbitrary GAQL (e.g. GoMarble's google_ads_run_gaql), use it. The CSV alternative - running two separate UI reports from the PMax campaign page - is the most common point of friction in PMax audits and worth flagging early in the conversation.

---

Skill 4: Keyword Research

When a decision depends on search volume, CPC (top-of-page bid), or competition, real keyword data is required. Without that data, do not estimate volume or CPC - say insufficient data and ask the user to share it.

When to use this skill

Trigger	Action
"Do keyword research" / "Suggest keywords" / "Find buyer keywords"	Discover new keyword ideas + validate
"Are these keywords good?" / "Which should I keep?"	Validate against volume + CPC
"Which keyword is better?" / "Prioritize these"	Compare metrics across candidates
"Create a new search campaign" / "Launch Google Search ads"	Research → validate BEFORE proposing structure
"Expand to a new country" / "International Google Ads"	Pull metrics with the new geo + language
"Can we scale?" / "What CPCs should we expect?"	Pull top-of-page bid forecasts

When NOT to use this skill

Keyword Planner data is NOT required for:

• Creative work (RSA headlines, descriptions, ad strength)
• Structure work (match type explanations, campaign naming, account audits)
• Post-performance analysis (CPA / ROAS analysis, search-term-report mining for existing campaigns)

Data to gather

For each candidate keyword, the user needs to provide (or pull from any keyword-research tool - Keyword Planner, third-party tools, etc.):

• Search volume (monthly searches in the target geo + language)
• Top-of-page bid range (Low / Medium / High, or explicit min–max CPC)
• Competition (Low / Medium / High - used only as a tie-breaker)

If targeting a specific geography or language, confirm the metrics are scoped to that region.

Step 1: Generate Candidate Keywords

Generate using buyer-intent patterns only.

Approved patterns:

• [service] + agency
• [service] + management
• [service] + pricing
• [service] + cost
• [service] + consultant
• [software] + for + [industry]
• [competitor] + alternative

Disallowed patterns:

• how to
• what is
• guide
• tips
• tutorial
• free
• examples

Step 2: Evaluate Each Keyword

Criterion	Rule
Intent	Buyer intent must be explicit in the keyword text. Unclear → REJECT.
Search Volume	Volume < 50 → REJECT. Volume ≥ 50 → PASS.
Top-of-Page Bid	High → STRONG APPROVE. Medium → APPROVE. Low → APPROVE only if intent is very strong. Zero / Missing → REJECT.
Competition	Tie-breaker only. Never reject solely due to high competition.

Step 3: Final Decision Rule

IF
  buyer intent = YES
  AND volume ≥ 50
  AND top-of-page bid ≠ Low / Zero
THEN
  APPROVE
ELSE
  REJECT

Step 4: Match Type Assignment

For approved keywords only:

Match Type	When to Use
Exact	Very strong intent, specific wording
Phrase	Core scalable keywords
Broad	Do NOT suggest by default

Output Format (per keyword)

Keyword:
Match Type:
Volume:
Top of Page Bid:
Competition:
Decision: APPROVE / REJECT
Reason:

Example:

Keyword: google ads agency for ecommerce
Match Type: Phrase
Volume: 90
Top of Page Bid: High
Competition: Medium
Decision: APPROVE
Reason: Clear buyer intent + active advertiser bidding

Core Principles

1. Intent > Volume (always)
2. Top-of-page bid = strongest commercial signal
3. Informational keywords = auto-reject
4. Fewer correct keywords > many random ones

Minimum required data (by source)

Source	What you need
MCP	Per candidate keyword (scoped to target geo + language): monthly search volume, top-of-page bid range (low / high), competition level. Via tools like google_ads_keyword_discover (seed → ideas) and google_ads_keyword_metrics (validate specific keywords).
CSV	A Keyword Planner export covering the target geo + language. Standard Planner CSV columns map directly: "Avg. monthly searches" → volume; "Top of page bid (low range)" / "(high range)" → top-of-page bid; "Competition" → competition.
Screenshot / copy-paste	Workable if the user pastes the Planner table directly with all three signals.

Tool hints

This skill cannot run on guesses - never estimate volume or CPC from training data. If MCP keyword tools are available, they're the cleanest source. If only a CSV export is available, confirm it's scoped to the target geo and language before applying the decision rule (a US-scoped Planner export gives the wrong CPCs for a UK launch).

---

Cross-Cutting Notes

How to handle missing data

If the user can't provide the data a skill needs, do not invent values. State explicitly:

"To answer this properly I need [specific metrics]. You can pull these from [Google Ads UI report / a CSV export / a connected data tool]. Without them I can outline the framework but not give a recommendation."

How to handle change requests

When the user asks Claude to change something in their account (e.g. "raise the bid", "pause the campaign"):

1. Run the relevant analysis skill to validate the change is justified.
2. Apply the guardrails (especially scaling caps and the "no simultaneous bid + budget" rule).
3. Recommend the change in plain language with the exact entity, the current value, the proposed value, the rationale, and what to monitor afterwards.

Claude does not execute changes on the user's ad account. These skills are read-and-recommend only - the user applies recommendations manually in the Google Ads UI.

How to handle multiple campaign types in one account

If the user's account has Search + Shopping + PMax all active, ask which to focus on first. Don't analyze all three at once unless explicitly requested - it dilutes the recommendation quality.

How to handle account-type ambiguity

Always classify the account before applying KPIs:

1. Check the conversion actions present (purchase / lead / store visit / call).
2. Check whether conversion value is populated.
3. If unclear, ask the user.

If the account is footfall / awareness / local-service, replace CPA/ROAS in headline metrics with: cost per footfall action, cost per call, cost per direction, CPM, reach, frequency. State explicitly that purchase-derived CPA/ROAS do not apply.

What good looks like

• The data source is named first, and every rule that the source cannot support is skipped out loud instead of guessed.
• Each recommendation maps to a control that exists in the Google Ads UI: campaign budget, bid strategy target, pause or enable. No budget is "shifted" between keywords, devices or asset groups.
• Nothing scales until search-term hygiene is done, CPA or ROAS is on target, and conversion volume clears the significance floor.
• Auction pressure is diagnosed as rank or budget before anyone touches bids or budgets.
• When diagnostics are incomplete the output says insufficient data and asks for the missing report, rather than recommending.