Agency OS · Architecture & Training
Sections

Live Jessie skill

Google Ads Reporting.

Explains paid search performance using Source Asset backed Google Ads evidence, GA4 outcome context, approved client context, and AI Gateway synthesis.

When It Routes

Live Paid search skills/google-ads-reporting/SKILL.md

A task asks for Google Ads, paid search, PPC, spend, conversion, CPA, ROAS, search terms, or campaign reporting.

Flow

1
Task intake

Jessie checks the Asana task for client, account, period, comparison, and paid-search questions.

2
Toolkit

The run receives scoped Google Ads MCP tools, GA4 support, Browser Run landing-page checks, context retrieval, AI Gateway, and Asana write-back access.

3
Context

Jessie loads approved goals, budget decisions, target CPA/ROAS caveats, campaign tests, and consultant notes.

4
Evidence

Jessie retrieves account, campaign, ad group, search-term, keyword, bidding-strategy, conversion-action, and optional landing-page evidence.

5
Insight

The LLM interprets spend, traffic, efficiency, and conversion movement through Cloudflare AI Gateway.

6
Review output

Jessie attaches an internal Google Ads report and posts a review-required Asana comment.

Context Jessie Looks At

  • Approved client context for targets, budget moves, campaign priorities, product/service focus, and measurement caveats.
  • Task notes that explain known tests, expected CPA movement, or campaign constraints.
  • Linked strategy, plan, or previous performance context when available.

Data And Tools

  • Google Ads account, campaign, ad group, search-term, keyword, bidding-strategy, conversion-action, impression-share, and change evidence where API queries succeed.
  • Admin Registry Source Assets for non-secret customer ID resolution; credentials stay in Cloudflare Secrets Store or Worker secret bindings.
  • GA4 paid search sessions, conversions, landing-page outcomes, and corroborating movement.
  • Bounded Browser Run landing-page checks when useful.

LLM Insight Stage

Google Ads reporting uses an AI Gateway synthesis stage after deterministic evidence assembly. The LLM is expected to interpret tradeoffs in light of approved context, not only metric deltas.

Outputs

  • Internal Google Ads HTML report attached to the source Asana task.
  • Spend, traffic, efficiency, campaign, ad group, search-term, landing-page, and change-history sections.
  • Likely drivers, confidence, lag timing, and consultant follow-up checks.

Guardrails

  • No budget, bid, keyword, asset, conversion, tracking, website, or account changes.
  • Read-only GAQL queries are scoped to the current run and capped before data returns to the execution harness.
  • No direct client publishing.
  • Provider failures block evidence-backed recommendations rather than becoming invented recommendations.

Example Prompt

Please do a Google Ads report for Plunkett Homes last month.

Source References