Google Ads

Google Ads

Strategic guide for building new Google Ads campaigns. Research first, consult intelligently, validate everything, and use the blueprint flow for controlled creation.

Requirements

• Hyper MCP installed and connected. https://app.hyperfx.ai/mcp
• Google Ads integration connected at https://app.hyperfx.ai/integrations.

If google_ads_list_accounts is not in the tool list, stop and tell the user to enable Hyper MCP and connect Google Ads.

Out of scope

• Ongoing optimization (search-term cleanup, bid adjustments, ad testing, restructuring existing campaigns) → defer to a future google-ads-operator skill.
• Creative generation (headlines, descriptions, images) → [ad-creative-generation](../ad-creative-generation).
• Cross-platform campaign launches → use this skill for Google, then invoke meta-ads / tiktok-ads separately.

Tool surface

| Tool | Purpose | | --- | --- | | google_ads_list_accounts | Discover accessible accounts (and MCC sub-accounts). | | google_ads_execute_gaql | Run a GAQL query (conversion actions, search term reports, etc.). | | google_ads_search_locations | Resolve human-readable location names to IDs. | | google_ads_list_assets, google_ads_upload_image_asset | Manage image assets for Display / PMax. | | google_ads_preview_blueprint, google_ads_create_from_blueprint | Validate + create Search/Display campaigns. | | google_ads_preview_pmax_blueprint, google_ads_create_from_pmax_blueprint | Validate + create Performance Max campaigns. |

Phase 1: Initial Setup

Call google_ads_list_accounts() to list accessible accounts.

• If multiple: ask the user to select one.
• If single: inform the user and proceed.

CRITICAL: Verify account access before any operations.

Phase 2: Discovery & Research (MANDATORY)

Research Steps

1. Get the real domain (ask; don't infer).
2. Scan the site end-to-end: home, product/service, pricing, about, FAQs, locations, contact, landing pages.
3. Understand funnel & goals: primary conversions, CTAs, forms/checkout, thank-you pages.
4. Extract messaging: value props, differentiators, proof, offers.

Conversion Tracking Check

Before asking questions, run the purpose-built diagnostic (simpler than a hand-rolled GAQL query):

google_ads_diagnose_conversion_tracking(customer_id="")

This returns all conversion actions, their status, and any tracking signal issues in one call. If the user's MCP doesn't expose google_ads_diagnose_conversion_tracking, fall back to GAQL:

google_ads_execute_gaql(
  customer_id="",
  query="""
    SELECT conversion_action.id, conversion_action.name,
           conversion_action.status, conversion_action.type
    FROM conversion_action
    WHERE conversion_action.status = 'ENABLED'
  """
)

> google_ads_execute_gaql vs google_ads_run_gaql: execute_gaql works on manager accounts (MCC) and sub-accounts. run_gaql is only available on non-manager accounts. Use execute_gaql consistently — it works everywhere.

Market & Keyword Research

• Inspect SERPs, competitors, and themes.
• Propose keyword candidates (intent-aligned).
• Identify initial negatives.

Confirm Criticals

• Daily budget (+currency).
• Served geos.
• Constraints.
• Tracking status.

Phase 3: Consultation

Act as a partner, not order-taker:

• Present findings (site + GAQL + market).
• Recommend bidding (default Smart Bidding when tracking exists) with trade-offs.
• Propose structure (campaign → themed ad groups → keywords + match types).
• Suggest locations via google_ads_search_locations(customer_id="...", location_names="New York"). Note: location_names is a single string, not an array — pass a city, state, country, or postal code and the tool returns matching geo target IDs.
• Set budget expectations via benchmark ranges.
• Show reasoning for each choice.

Phase 4: Pre-Creation Summary (Must Be Approved)

Campaign Strategy for [Business Name]

Sources: [URL], GAQL
Conversion Setup: [GAQL findings]
Primary Goal: [objective + why]
Bidding: [strategy + why]
Budget: $[X]/day (expectations)
Locations: [targets + rationale]
Keyword Themes: [themes + match types + 2-3 examples]
Messaging: [angles pulled from site]
Trade-offs/Risks: [bullets]

Approve to proceed?

Wait for explicit approval. No emojis. No assumptions.

Phase 5: Campaign Creation (Blueprint System)

> CRITICAL: Always use the blueprint system for campaign creation. It validates locally, fills smart defaults, resolves locations, and rolls back on failure.

Workflow: Preview → Confirm → Create

1. Build the blueprint JSON from research.
2. Call the preview tool to validate and show the user what will be created.
3. Get explicit user approval.
4. Call the create tool.

Search & Display Campaigns

Preview: google_ads_preview_blueprint(blueprint={...}) Create: google_ads_create_from_blueprint(blueprint={...})

{
  "customer_id": "1234567890",
  "name": "Campaign Name",
  "budget_amount_micros": 50000000,
  "advertising_channel_type": "SEARCH",
  "bidding_strategy": "MAXIMIZE_CLICKS",
  "status": "PAUSED",
  "location_names": ["New York", "Los Angeles"],  // array in blueprints; note: google_ads_search_locations takes a single string for manual lookup
  "conversion_action_ids": ["1234567890"],
  "ad_schedules": [
    {"day_of_week": "MONDAY", "start_hour": 9, "end_hour": 17}
  ],
  "sitelinks": [
    {"link_text": "Free Trial", "final_urls": ["https://example.com/trial"]}
  ],
  "callouts": [{"callout_text": "Free Shipping"}],
  "ad_groups": [
    {
      "name": "Ad Group 1",
      "keywords": [
        {"text": "marketing software", "match_type": "PHRASE"},
        {"text": "competitor brand", "match_type": "EXACT", "negative": true}
      ],
      "ads": [
        {
          "headlines": ["Headline 1", "Headline 2", "Headline 3"],
          "descriptions": ["Description 1", "Description 2"],
          "final_urls": ["https://example.com"]
        }
      ],
      "audiences": [{"audience_id": "1234567890"}]
    }
  ]
}

For Display campaigns, set "advertising_channel_type": "DISPLAY" and use display_ads instead of ads:

"display_ads": [{
  "headlines": ["Headline 1", "Headline 2", "Headline 3"],
  "long_headline": "Longer headline up to 90 characters",
  "descriptions": ["Description 1", "Description 2"],
  "business_name": "Business Name",
  "final_urls": ["https://example.com"],
  "marketing_image_asset_ids": ["1234567890"],
  "square_marketing_image_asset_ids": ["1234567891"],
  "logo_asset_ids": ["1234567892"]
}]

> CRITICAL (Display): All three image asset types are REQUIRED: > - marketing_image_asset_ids — landscape 1.91:1 (e.g. 1200×628). > - square_marketing_image_asset_ids — square 1:1 (e.g. 1200×1200). > - logo_asset_ids — square 1:1 or landscape 4:1.

Performance Max Campaigns

Preview: google_ads_preview_pmax_blueprint(blueprint={...}) Create: google_ads_create_from_pmax_blueprint(blueprint={...})

{
  "customer_id": "1234567890",
  "name": "PMax Campaign",
  "budget_amount_micros": 50000000,
  "bidding_strategy": "MAXIMIZE_CONVERSIONS",
  "status": "PAUSED",
  "location_names": ["United States"],
  "conversion_action_ids": ["1234567890"],
  "asset_groups": [
    {
      "name": "Asset Group 1",
      "headlines": ["Headline 1", "Headline 2", "Headline 3"],
      "long_headlines": ["Longer headline up to 90 characters"],
      "descriptions": ["Description 1", "Description 2"],
      "business_name": "Business Name",
      "final_urls": ["https://example.com"],
      "marketing_image_asset_ids": ["1234567890"],
      "square_marketing_image_asset_ids": ["1234567891"],
      "logo_asset_ids": ["1234567892"]
    }
  ]
}

> CRITICAL (PMax): All three image asset types are REQUIRED with correct aspect ratios. The preview tool validates ratios before creation.

Blueprint Features

| Feature | Details | | --- | --- | | Smart defaults | Budget name, start date, network settings, ad group type auto-filled. | | Location resolution | Pass location_names (human-readable) instead of IDs. | | Client-side validation | Char limits, match types, image ratios checked before any API call. | | Batch operations | Keywords, extensions, ad schedules batched for efficiency. | | Cleanup on failure | Entire campaign removed if any step fails after campaign creation. | | Ad extensions | Sitelinks, callouts, structured snippets — created and linked automatically. | | Ad scheduling | Day-parting at campaign level via ad_schedules. | | Audiences | Target or exclude audiences per ad group. | | Conversion actions | Link specific conversion actions via conversion_action_ids. |

Bidding Strategies

| Strategy | Fields | Notes | | --- | --- | --- | | MANUAL_CPC | — | Manual cost-per-click. | | MAXIMIZE_CLICKS | — | Default, good starting point. | | MAXIMIZE_CONVERSIONS | target_cpa_micros (optional) | Requires conversion tracking. | | MAXIMIZE_CONVERSION_VALUE | target_roas (optional) | Requires conversion values. | | TARGET_CPA | target_cpa_micros (required) | Sets a target cost per acquisition. | | TARGET_ROAS | target_roas (required) | Sets a target return on ad spend. |

> For PMax: TARGET_CPA and TARGET_ROAS are translated to constraints within MAXIMIZE_CONVERSIONS / MAXIMIZE_CONVERSION_VALUE.

Technical Rules

• Budget in micros: $50 → 50,000,000.
• Create PAUSED; activate only after approval.
• Location resolution: use location_names in blueprint (auto-resolved) or google_ads_search_locations for manual lookup.
• RSA limits: headlines ≤30 chars (min 3), descriptions ≤90 chars (min 2).
• Creative source: only from the actual site; no generic claims.
• Image assets must already exist in the account (use google_ads_list_assets to find them).

Campaign Workflow

Initial Setup → Research (site + GAQL + market) → Analyze (goals, audience, keywords) → Consult (options + trade-offs) → Recommend (structure + bids) → Confirm (summary approved) → Build Blueprint → Preview → User Approval → Create (PAUSED) → Activate (post-approval).

GAQL Resource Compatibility

Critical: Not all Google Ads resources can be joined in a single query.

searchtermview

Use segments.keyword.info.* for keyword text and match type — never ad_group_criterion.*:

SELECT
    search_term_view.search_term,
    customer.descriptive_name,
    segments.keyword.info.text,
    segments.keyword.info.match_type,
    campaign.name,
    ad_group.name,
    metrics.clicks,
    metrics.impressions,
    metrics.conversions,
    metrics.cost_micros
FROM search_term_view
WHERE segments.date DURING LAST_7_DAYS
ORDER BY metrics.impressions DESC

ad_group_criterion is incompatible with search_term_view as the FROM resource. Selecting ad_group_criterion.* fields will always fail with PROHIBITED_RESOURCE_TYPE_IN_SELECT_CLAUSE.

keyword_view

For keyword performance (Quality Score, bid estimates):

SELECT
    ad_group_criterion.keyword.text,
    ad_group_criterion.keyword.match_type,
    metrics.clicks,
    metrics.impressions,
    metrics.historical_quality_score
FROM keyword_view
WHERE segments.date DURING LAST_30_DAYS

Multi-Account Workflows (MCC)

When a user provides a manager account (MCC) with multiple sub-accounts:

Batching Rule

Never query more than 5 sub-accounts in a single run. Each google_ads_execute_gaql call goes through a proxy with a timeout. Querying 10+ accounts sequentially in one loop reliably causes 504 timeouts.

Process in batches of 5 and aggregate results before proceeding:

Batch 1: accounts[0:5]   → collect results
Batch 2: accounts[5:10]  → collect results
Batch 3: accounts[10:15] → collect results
...merge all batches...

Announce progress to the user: "Processing accounts 1-5 of 23...".

504 Timeout Recovery

If a google_ads_execute_gaql call times out (504 / "Google Ads API timed out after retries"):

1. The tool already retries 3× internally — do not retry immediately.
2. Reduce the scope: narrow the date range, add a LIMIT, or split the account list further.
3. Tell the user which accounts succeeded and which failed; do not silently drop accounts.

Cross-Account Search Term Report (Most Common Use Case)

For MCC search term reports across many sub-accounts:

1. Get all accounts: google_ads_list_accounts()
   Filter to sub-accounts: [a for a in result.accounts if not a["manager"]]
2. For each batch of 5 accounts:
   - Run search_term_view GAQL (using segments.keyword.info.*, NOT ad_group_criterion).
   - Collect results tagged with customer.descriptive_name.
3. Merge all batches into a single dataset.
4. Write to Google Sheets or CSV file for the user.

Minimum required columns for a search term report:

• Search Term, Account Name, Keyword, Match Type, Campaign, Ad Group.
• Clicks, Impressions, Conversions, Cost.
• Add Negative (Yes/No — your recommendation).

Critical Safety Rules

Never:

• Assume URL/budget/location IDs.
• Skip research.
• Invent keywords/copy.
• Apply rigid "best practices" without context.
• Build without buy-in.
• Skip the preview step before creation.
• Join ad_group_criterion with search_term_view in GAQL.
• Query more than 5 accounts in a single batch.

Cached Data (optional, when google_ads_query_insights is exposed)

If the MCP exposes google_ads_query_insights, use it for large multi-account performance queries — it reads from a local cache refreshed hourly and avoids API timeouts entirely.

Query Performance

Call google_ads_query_insights — its built-in description includes the exact table name, schema, and example queries. Read the tool description first, then pass a SQL query targeting that table. Typical pattern:

google_ads_query_insights(
  query="SELECT date, campaign_name, SUM(cost_micros) as spend, SUM(clicks) as clicks
         FROM 
         WHERE customer_id = '1234567890'
           AND date >= CURRENT_DATE - INTERVAL '7 days'
         GROUP BY date, campaign_name ORDER BY date DESC"
)

Replace ` with the table name shown in the googleadsquery_insights` tool description. Cache is refreshed hourly — no manual sync needed.

> If the tool returns a "no data cached" error, check the suggestion field in the response — it will contain the correct workspace-specific table name. Retry the query using that suggested table name instead.

Key columns available (verify in tool description):

• Hierarchy: date, customer_id, campaign_id, campaign_name, campaign_status
• Volume: impressions, clicks, cost_micros (÷ 1,000,000 for dollars)
• Conversions: conversions, conversions_value, conversion_rate
• Efficiency: ctr, average_cpc, average_cpm

Supports standard SQL aggregations and window functions. For cross-account queries, omit the customer_id filter and GROUP BY it instead.

Source & license

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

• Author: hyperfx-ai
• Source: hyperfx-ai/marketing-skills
• License: MIT
• Homepage: https://www.hyperfx.ai/mcp

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