The "Claude Ads" project is a comprehensive skill designed for Claude Code, focusing on paid advertising audit, optimization, and strategic planning. It provides an advanced analytical framework for major ad platforms including Google Ads, Meta Ads (Facebook/Instagram), YouTube Ads, LinkedIn Ads, TikTok Ads, Microsoft Ads (Bing), and Apple Ads.

The core methodology of Claude Ads is built around a local-first, data-driven analysis architecture. It operates entirely on the user's local machine via the Claude Code CLI, ensuring that no ad account data is transmitted to external servers, thereby maintaining data privacy. The analysis is performed on user-provided data, which can include ad account exports, screenshots, or directly pasted metrics from ad platform dashboards.

Architecture and Workflow:
The system's architecture is orchestrated by the primary /ads command, which intelligently routes requests to specialized sub-skills and agents.

1. Orchestrator (/ads): This serves as the central command dispatcher, interpreting user input and delegating tasks.
2. Sub-skills: These are specialized modules responsible for deep, single-domain analysis. Examples include /ads google for Google Ads, /ads meta for Meta Ads, /ads creative for cross-platform creative analysis, and /ads budget for budget review. Each sub-skill generates structured outputs relevant to its domain.
3. Agents: For comprehensive analyses, particularly the full multi-platform audit (/ads audit), the system employs parallel subagent delegation. This involves spawning multiple specialized agents concurrently to maximize analytical speed and breadth. For instance, the /ads audit command leverages six parallel subagents: audit-google, audit-meta, audit-creative, audit-tracking, audit-budget, and audit-compliance. Additionally, four specialized agents are dedicated to AI-powered creative generation.
4. Reference Data (RAG Pattern): The system integrates 25 built-in reference files, employing a Retrieval-Augmented Generation (RAG) pattern. These files contain critical data such as 2026-current benchmarks, bidding decision trees, platform specifications, compliance requirements, conversion tracking guides, and MCP integration guides. This data is loaded on-demand, ensuring that only relevant information is retrieved and utilized for each specific analysis, enhancing context and accuracy.
5. Industry Templates: For strategic planning (/ads plan), the skill utilizes 12 industry-specific templates. These templates offer predefined frameworks including platform mix recommendations, campaign architecture, creative strategies, targeting guidelines, budget allocations, and Key Performance Indicator (KPI) targets tailored to various business types (e.g., SaaS, e-commerce, local service, B2B enterprise).

Key Features and Technical Details:

• Comprehensive Audit Checks: Over 250 distinct audit checks are implemented across all supported platforms. Each check is assigned a weighted severity score, contributing to an overall Ads Health Score.
  • Platform-Specific Checks: E.g., Google Ads includes 80 checks covering Search, PMax, AI Max, Demand Gen, CTV, YouTube; Meta Ads has 50 checks for Pixel/CAPI, creative diversity, structure, and audience.
  • Cross-Platform Checks: These include privacy infrastructure assessment (e.g., Consent Mode V2, CAPI, Events API, AdAttributionKit), creative diversity, and refresh cadence.
• Ads Health Score (0-100): A sophisticated weighted scoring algorithm is used to compute an overall health score for the ad account. This algorithm applies severity multipliers to identified issues, categorizing performance into grades (A to F) that indicate required intervention levels.
• Industry Detection: The system can auto-detect the business type based on ad account signals (e.g., product feeds, conversion events, platform mix, targeting patterns). This enables the loading of industry-specific benchmarks and templates for more relevant recommendations.
• Quality Gates: Hard rules are enforced during audits to prevent suboptimal or non-compliant recommendations. Examples include:
  • Prohibiting Broad Match without Smart Bidding in Google Ads.
  • A "3x Kill Rule" to flag and recommend pausing campaigns where Cost Per Acquisition (CPA) exceeds 3x the target.
  • Budget sufficiency checks (e.g., Meta >= 5x CPA/ad set, TikTok >= 50x CPA/ad group).
  • Learning phase protection to prevent edits during active learning periods.
  • Automatic compliance checks for Special Ad Categories (housing, credit, finance) and privacy infrastructure (Consent Mode V2, CAPI).
  • Andromeda creative diversity checks to flag Meta accounts with insufficient genuinely distinct creatives.
• PPC Financial Modeling (/ads math): Provides a calculator for key PPC metrics such as CPA, Return on Ad Spend (ROAS), break-even points, budget forecasting, and Lifetime Value to Customer Acquisition Cost (LTV:CAC).
• A/B Test Design (/ads test): Facilitates the design of A/B tests by providing frameworks for hypothesis formulation, significance determination, sample size calculation, and duration planning.
• PDF Report Generation (/ads report): Professional PDF audit reports can be generated for client deliverables. These reports feature the health score gauge, platform comparison charts, pass/fail distributions, formatted tables, and a zero-overlap layout.
• Optional Live Data Integration: While the primary analysis is offline, Claude Ads can be optionally paired with MCP (Marketing Cloud Platform) servers for direct API access to platforms like Google Ads (via mcp-google-ads) or Meta Ads (via Adspirer MCP). This enables real-time data fetching without manual exports, though the analytical processing remains local to the user's machine.