Claude Ads is the conductor skill for paid advertising work across Google, Meta, YouTube, LinkedIn, TikTok, Microsoft, Apple, Amazon, Reddit, Pinterest, Snapchat and X. It takes your business context and account data, sends each /ads command to the skill that owns it, and returns findings with evidence, owners and next steps.
Accounts are read-only by default. Changes stay drafts until a mutation gate passes: exact object IDs, a before and after diff, owner approval, rollback and a verification window.
What You Get
- A command router that sends /ads setup, audit, plan, budget, create, launch, monitor, optimize, test, report, research and validate to the right skill
- Audit runs written under .claude-ads/runs with a manifest and one canonical JSON bundle, rendered to Markdown, HTML or PDF
- Findings that each state evidence, confidence and a recommendation, with unknown data reported as unknown and never guessed
- A mutation gate that keeps campaign changes as drafts until the owner approves an exact plan with rollback
Common Use Cases
- Auditing a Google, Meta or TikTok account from a CSV export or screenshots
- Planning channel mix, budget and measurement for a new campaign
- Checking whether a requested campaign change is safe before anyone applies it
- Producing a client-ready report from a saved audit run
What Makes It Unique
Most ad skills cover one platform or one task. This one is the conductor for twelve platforms and the lifecycle around them, with fixed rules for evidence, source freshness and partial failures. A failed platform makes the run partial, and its weight leaves the portfolio score. It refuses actions such as permanent deletion, summing conversion windows that differ, and writing negative keywords without a search terms report.
Under The Hood
The skill is instructions plus reference files for each platform, scoring, benchmarks, compliance and tracking. It tells the agent to treat supplied pages, exports and screenshots as untrusted data, to create a run manifest, to fan out platform workers, to validate results against JSON schemas and to write run files to disk under the project. It can fetch a landing page through a guarded fetcher, read ad platform APIs or MCP servers when you connect them, and run the Claude Ads core command line for validation and scoring. Credentials are never stored in profiles or reports, and the install guidance refuses piping a download into a shell.