Google Ads Audit

A one-time, account-wide Google Ads health audit using live data via Cogny's MCP server. Covers every campaign type, scores the account out of 100, and routes you to the right campaign-type skill for deeper drill-down.

Requires: Cogny Agent subscription ($9/mo) — Sign up

Works with

Call list_integrations once. Audit Google Ads in full; if Microsoft Ads is connected, add its overview, pacing and waste numbers as a separate block. Never sum the two.

Step 0: Cogny check

If no mcp__cogny__* tools exist (no list_integrations, no cogny_status), print this and stop:

Google Ads Audit needs your marketing data through Cogny's MCP gateway.

1. Sign up at https://cogny.com and connect Google Ads (and Microsoft Ads, GA4, Shopify or
   WooCommerce if you use them).
2. Add Cogny to .mcp.json in your project:
   {"mcpServers": {"cogny": {"type": "http", "url": "https://app.cogny.com/mcp"}}}
3. Restart Claude Code; it opens a browser for the OAuth login on first use.

Nothing to run without Cogny.

If Cogny is connected but mcp__cogny__mcp__Google_Ads__tool_execute_gaql is missing, say "connect Google Ads in Cogny" (or "reconnect Google Ads in Cogny" if it's listed but unhealthy or a read returns 401/permission denied) and stop. Do not guess at numbers.

Usage

/google-ads-audit — full account audit /google-ads-audit campaigns — campaign structure + performance only /google-ads-audit budget — budget pacing only /google-ads-audit tracking — conversion tracking sanity only /google-ads-audit negatives — account negatives + exclusions only

Approach

This is the account-level audit — it finds structural problems and wasted spend across the whole account. For campaign-type depth (asset groups, keywords, feeds, audiences) it points you to /pmax-audit, /search-campaign-audit, /shopping-campaign-audit, or /demand-gen-audit.

All money fields come back in micros — divide cost_micros etc. by 1,000,000.

An empty GAQL result (no rows, or the tool returns no output) means zero: no matching entities or no activity in the window. Report it as 0, not as an error or missing data.

Steps

1. Account discovery

tool_list_accessible_accounts

If multiple accounts are accessible, list them and ask which to audit. Then read the account name and currency:

SELECT customer.id, customer.descriptive_name, customer.currency_code, customer.time_zone
FROM customer

Report every Google Ads money figure in customer.currency_code.

2. Campaign overview by channel type (last 30 days)

SELECT customer.currency_code, campaign.name, campaign.status,
  campaign.advertising_channel_type, campaign.bidding_strategy_type,
  campaign_budget.amount_micros, metrics.cost_micros, metrics.conversions,
  metrics.conversions_value, metrics.clicks, metrics.impressions
FROM campaign
WHERE segments.date DURING LAST_30_DAYS
ORDER BY metrics.cost_micros DESC

Build a per-channel-type summary (Search / Performance Max / Shopping / Display / Video / Demand Gen): total spend, conversions, CPA, ROAS (conversions_value / cost).

Microsoft Ads (if connected): mcp__cogny__mcp__Bing_Ads__tool_list_accounts, then tool_get_campaigns (account_id, campaign_types: ["Search", "Shopping", "PerformanceMax", "Audience"]) and tool_get_report (account_id from tool_list_accounts, report_type: "campaign", start_date, end_date). Every tool_get_report call needs account_id. Build the same per-type summary from Spend, Conversions, Revenue, CampaignType. Many accounts import Google campaigns into Microsoft Ads; flag imported campaigns with zero conversions or a CPA more than 2x the Google equivalent.

Flag:

• Campaigns spending >5% of account budget with zero conversions
• MAXIMIZE_CONVERSIONS / TARGET_CPA campaigns with <15 conversions/30d (too little

data for the bidding strategy to learn)

• Multiple campaigns of the same type competing for the same theme (self-competition)
• Enabled campaigns with zero spend (broken, or budget/targeting starved)

3. Conversion tracking sanity

You must run BOTH queries below before flagging anything in this section. Structure alone (which actions exist, which are primary_for_goal) tells you nothing about what is actually driving the bidding signal. A "Primary" action with zero recent volume is harmless; a "Secondary" action can be a deliberate dedup choice. Always cross-reference the structure against actual conversion volume per action.

3a. Structure — which conversion actions exist:

SELECT conversion_action.name, conversion_action.status, conversion_action.type,
  conversion_action.category, conversion_action.counting_type,
  conversion_action.primary_for_goal, conversion_action.include_in_conversions_metric
FROM conversion_action

3b. Volume — which actions are actually firing (last 30 days):

SELECT segments.conversion_action_name, segments.conversion_action_category,
  metrics.all_conversions, metrics.all_conversions_value,
  metrics.conversions, metrics.conversions_value
FROM customer
WHERE segments.date DURING LAST_30_DAYS

An empty result here means 0 conversions for every action in the window. Go straight to 3c and report it as zero volume, not as a failed query.

Then join: for every conversion action with non-trivial volume, note whether it is Primary or Secondary, and whether metrics.conversions (the bidding signal) or only metrics.all_conversions (the reporting signal) is counting it.

3c. Reconcile against account totals. Pull account-level metrics.conversions and metrics.conversions_value for the same window — the sum across Primary actions in 3b should match (within rounding). If it does, the metrics.conversions figure is the real booking/purchase/lead action and bidding is grounded in real value. State this explicitly in the report before making any tracking claim.

Flag (each must be backed by a number from 3b, not a guess from 3a):

• No primary action has volume in the last 30 days — bidding really is flying

blind. (Note: a Primary action with zero recent volume is harmless, not broken.)

• A high-volume real revenue action (purchase, booking, qualified lead) is Secondary

while a low-value proxy (page view, "Lokala åtgärder", "Detail product page view") is the only Primary — that is the "bidding on page views" pattern.

• MANY_PER_CLICK counting on a lead-gen action with meaningful volume (double-counts

leads; usually should be ONE_PER_CLICK).

• Two near-duplicate Primary actions both firing on the same event (e.g. on-site

booking + GA4 booking both Primary → double-counted bid signal).

• REMOVED / HIDDEN actions still receiving conversions.

3d. Enhanced when connected: check against real sales. For ecommerce, pull the same 30 days from the store: Shopify mcp__cogny__mcp__Shopify__tool_list_orders (query_filter: "created_at:>=YYYY-MM-DD financial_status:paid", paginate with after) or WooCommerce mcp__cogny__mcp__WooCommerce__get_sales_report (date_min, date_max). For lead gen or as a second view, find the property with GA4 tool_list_account_summaries, then tool_run_report (property_id, date_ranges, dimensions: ["sessionGoogleAdsCampaignName"], metrics: ["sessions", "keyEvents", "purchaseRevenue"]). Keep sessions in the list: GA4 drops rows where every requested metric is 0, so without it campaigns with zero key events and zero revenue vanish instead of showing as 0. Report as a ratio, never as a correction: "Google Ads purchase value = 1.4x Shopify paid revenue for the period." If Ads-reported purchase value exceeds total store revenue, that's a tracking finding (double counting or a wrong value), backed by both numbers. Paid-order counts from the store are the ground truth; GA4 and Ads both attribute.

Do NOT flag:

• A gap between all_conversions and conversions, by itself. That gap is just the

pile of Secondary actions (store visits, GA4 echoes of an on-site action, local actions, contacts, etc.) — it is often correct and intentional. Only flag if a high-value, non-duplicate action is sitting in Secondary.

• View-through / page-view actions marked primary_for_goal = true if they have

~0 conversions in 3b — they cost nothing, so they are not polluting bidding.

• A Secondary GA4 purchase when on-site purchase/booking is Primary — that is the

textbook dedup pattern, not a bug.

4. Budget pacing (last 7 days)

SELECT campaign.name, campaign.status, campaign_budget.amount_micros,
  metrics.cost_micros, metrics.impressions
FROM campaign
WHERE segments.date DURING LAST_7_DAYS AND campaign.status = 'ENABLED'

For each campaign compute average daily spend vs daily budget.

Flag:

• Campaigns consistently capped at ~100% of daily budget while converting profitably

(budget-limited — leaving volume on the table)

• Campaigns spending <50% of budget (delivery problem: bids, targeting, or quality)

5. Account negatives & exclusions

Negatives live in three places. Query all three before saying an account has none.

Shared lists:

SELECT shared_set.name, shared_set.type, shared_set.status, shared_set.member_count
FROM shared_set

Campaign-level negatives:

SELECT campaign.name, campaign.advertising_channel_type,
  campaign_criterion.keyword.text, campaign_criterion.keyword.match_type
FROM campaign_criterion
WHERE campaign_criterion.type = 'KEYWORD' AND campaign_criterion.negative = TRUE
  AND campaign.status = 'ENABLED'

Ad-group-level negatives:

SELECT campaign.name, ad_group.name,
  ad_group_criterion.keyword.text, ad_group_criterion.keyword.match_type
FROM ad_group_criterion
WHERE ad_group_criterion.type = 'KEYWORD' AND ad_group_criterion.negative = TRUE
  AND campaign.status = 'ENABLED'

Brand terms in non-brand campaigns (keywords and the search terms from step 6):

SELECT campaign.name, ad_group_criterion.keyword.text, ad_group_criterion.keyword.match_type,
  metrics.cost_micros, metrics.conversions
FROM keyword_view
WHERE segments.date DURING LAST_30_DAYS AND campaign.status = 'ENABLED'
  AND ad_group_criterion.status = 'ENABLED'

Treat a campaign as brand if its name says so (e.g. contains "brand"). In every other campaign, match keyword text and search terms against the brand name and its common misspellings.

Flag:

• No negatives at any of the three levels on Search campaigns
• No negative keyword lists shared across Search campaigns (campaign- or ad-group-level

negatives only — works, but hard to maintain)

• Brand terms bidding or matching inside non-brand campaigns, with their spend (inflates

non-brand performance and cannibalises the brand campaign); the fix is the brand name as a negative in non-brand campaigns

• For Performance Max / Display: missing account-level placement and content

exclusions (brand-safety + waste risk)

Microsoft Ads: tool_get_negative_keywords (account_id, entity_type: "Campaign", entity_ids) covers campaign-level negatives only, not shared lists.

6. Wasted spend scan (last 30 days)

SELECT search_term_view.search_term, campaign.name, campaign.advertising_channel_type,
  metrics.cost_micros, metrics.conversions, metrics.clicks
FROM search_term_view
WHERE segments.date DURING LAST_30_DAYS AND metrics.cost_micros > 0
ORDER BY metrics.cost_micros DESC
LIMIT 50

Sum the cost of terms with zero conversions and meaningful spend — this is the headline "wasted spend" number and the source of the largest estimated_impact_usd.

Microsoft Ads: tool_get_report with account_id and report_type: "search_query"; total zero-conversion Spend separately and report it in the Microsoft account's currency.

7. Score and report

Google Ads Audit — [Account Name]
Health Score: X/100  ·  30-day spend: [X]  ·  Conversions: [X]  ·  CPA: [X]
[Microsoft Ads, if connected: spend · conversions · CPA, own currency, not scored]
[Ads-reported vs store/GA4 revenue ratio, if a store or GA4 is connected]

Score breakdown:
  Account structure ...... X/20
  Conversion tracking .... X/25   (weighted highest — everything depends on it)
  Budget efficiency ...... X/20
  Wasted spend control ... X/20
  Negatives & exclusions . X/15

By campaign type:
| Type          | Spend | Conv | CPA | ROAS | Drill-down skill        |
|---------------|-------|------|-----|------|-------------------------|
| Performance Max | ... | ...  | ... | ...  | /pmax-audit             |
| Search          | ... | ...  | ... | ...  | /search-campaign-audit  |
| Shopping        | ... | ...  | ... | ...  | /shopping-campaign-audit|
| Demand Gen      | ... | ...  | ... | ...  | /demand-gen-audit       |

🔴 Critical
- ...
🟡 Important
- ...
🟢 Optimization
- ...

Top 3 Actions:
1. [Highest $ impact — estimated savings/lift]
2. ...
3. ...

Run next: [the campaign-type skill for the highest-spend type]

8. Record findings

For every actionable issue, call mcp__cogny__create_finding:

{
  "title": "$1,840/mo wasted on zero-conversion search terms",
  "body": "23 search terms spent $1,840 over 30 days with 0 conversions, led by 'free [product]' ($420) and '[competitor] alternative' ($310). Add as negative keywords and create a shared negative list.",
  "action_type": "negative_keyword",
  "expected_outcome": "Reclaim ~$1,800/mo of spend for converting queries",
  "estimated_impact_usd": 1840,
  "priority": "high"
}

Cloud uses create_ticket with the same content, fields mapped per section 2 of the routing guide.

estimated_impact_usd is in USD. If customer.currency_code isn't USD, convert the monthly figure at a current rate and state it in the body ("SEK 18,400/mo ≈ $1,700 at 10.8 SEK/USD"). Keep the account currency everywhere else in the report.

Action types: campaign_optimization, budget_adjustment, bidding_strategy, conversion_tracking, negative_keyword, account_structure.

Recording findings

Use create_ticket when the workspace has it (Cloud) and create_finding otherwise (Solo). The fields differ; see section 2 of ../marketing-engineer/references/mcp-routing.md.

Critical rules

1. Conversion tracking is checked first and weighted highest. A great account on

broken tracking is a broken account — say so loudly.

1. Never claim tracking is broken from structure alone. primary_for_goal = true

on an action only matters if that action has volume. Pull conversion volume per action (step 3b) and reconcile against account totals (step 3c) before writing any finding about conversion tracking, CPA basis, or bidding signal quality. If metrics.conversions reconciles to a real revenue/booking action, say so — don't wave at the all_conversions gap as if it implies missing primaries.

1. Quote real numbers, never benchmarks alone. "CPA is $48" beats "CPA looks high."
2. One finding per issue, each with a dollar figure. No vague advice.
3. This skill reads only. It never pauses campaigns or changes budgets. Write tools on

any platform are used only on the user's explicit request, and the skill asks first.

1. Platform numbers stay per platform. Google, Microsoft and store figures are shown

side by side in their own currencies, never added together.