---
name: google-ads-weekly-digest
description: Writes a weekly Google Ads performance digest from live data via the Caybl.ai connector - week-over-week and month-over-month changes against target, statistically flagged anomalies, the account changes that explain them, and a short "act this week / watch" list. Use for "weekly report", "what changed this week", "anything unusual in the account", "Monday Google Ads update", or as a scheduled weekly routine.
license: MIT
compatibility: Requires the Caybl.ai MCP connector (https://api.caybl.ai/mcp) with Google Ads connected.
metadata:
  author: Caybl.ai
  version: "1.0"
  adapted-from: mardab96/google-ads-skills (weekly-operator-digest-google-ads, anomaly-detection-google-ads), MIT
---

# Google Ads weekly digest

## Requires

The Caybl.ai MCP connector with Google Ads connected (https://caybl.ai/google-ads-mcp). Works well as a scheduled task (e.g. every Monday 8am).

## Workflow

1. **Fixed cut-off.** Use the last complete Monday-Sunday week (or the user's reporting week). Say which dates.
2. **Daily data, 90 days back.** `run_gaql_report`:
   `SELECT campaign.name, segments.date, metrics.cost_micros, metrics.clicks, metrics.impressions, metrics.conversions, metrics.conversions_value FROM campaign WHERE segments.date BETWEEN '<90 days ago>' AND '<last Sunday>'`
3. **Changes.** `run_gaql_report` on `change_event` for the last 30 days (`LIMIT 1000`) - budget, bid strategy, status, keyword and ad changes. **Read changes before explaining any movement.** A known change outranks any statistical story.
4. **Deltas.** This week vs last week, and month-to-date vs the same days last month - absolute and %, and against target where one exists.
5. **Anomalies.** For each campaign, compare each day with the same weekday over the baseline (30 and 90 days): z-score below 2 = normal, 2-3 = investigate, above 3 = significant. Rules:
   - fewer than 7 baseline days -> no z-score,
   - don't judge ratio metrics (CPA, conversion rate) at daily grain below ~5 conversions a day - judge them weekly,
   - "unexplained" is a valid answer.
6. **Attribute** each material movement to a change, a mix shift (one campaign growing or shrinking), seasonality, or "unexplained".
7. Flag campaigns that **stopped spending** or **hit their budget cap** every day.

## Output

1. **Headline** (2-3 sentences): spend, conversions and CPA/ROAS this week vs last, vs target.
2. **Table:** | Campaign | Spend | Delta WoW | Conv. | Delta WoW | CPA / ROAS | vs target | Note |
3. **Anomalies:** each with metric, size, z-score band and the likely cause (or "unexplained").
4. **Changes made this week** and their visible effect so far.
5. **Act this week** (max 3, specific) and **Watch** (things not yet worth acting on).

## Guardrails
- Don't call something an anomaly on tiny volume - say the data is too thin.
- Don't recommend changes during a Smart Bidding learning period without saying so.
- Report numbers as they are; never smooth or round away a bad week.