↓ メインコンテンツへスキップ

google-analytics

Google Analytics 4 (GA4) のアクセス解析データをローカル SQLite データベースに収集・分析するスキル。ページビュー、アクティブユーザー、滞在時間、流入元、外部リンククリックを保存し、SQL でサイト成果を検証します。

注記:以下の SKILL.md プロンプト本文は、LLM の実行精度を最大化するため英語で管理されています。

Google Analytics 4 SQLite Ingestion & SQL Analytics #

The google-analytics skill ingests Google Analytics 4 (GA4) traffic, reading depth, acquisition channels, event streams, and outbound clicks into a local SQLite analytics database (google_analytics.db or $XDG_DATA_HOME/google-analytics/analytics.db) without data loss, preserving raw JSON payloads on all records, and providing a fast SQL interface for website analytics.

Available scripts #

  • scripts/google_analytics.py: Automated sync, reporting, and annotation CLI for Google Analytics 4. Executed via uv run scripts/google_analytics.py (requires Google Cloud ADC or OAuth credentials).
  • scripts/test_google_analytics.py: Unit and regression test suite validating schema, query extraction, and CLI flags.

⚡ Quick Start & Primary Actions #

All operations are driven via the bundled Python CLI script:

bash
# 1. Authorize OAuth 2.0 (with analytics.edit & readonly scopes)
uv run scripts/google_analytics.py auth --port 8080

# 2. Discover accessible GA4 properties
uv run scripts/google_analytics.py properties

# 3. Create Deployment / Milestone Annotation (Cloud API + Local SQLite)
uv run scripts/google_analytics.py annotate \
  --title "Major Release / Architecture Overhaul" \
  --date 2026-08-18 \
  --commit abc1234 \
  --description "Milestone description and release context."

# 4. Incremental Sync (Updates newest days + 3-day latency lookback overlap)
uv run scripts/google_analytics.py sync --db path/to/database.db

# 5. Full Historical Backfill (Ingests up to 14 months of daily granular data)
uv run scripts/google_analytics.py sync --full --db path/to/database.db

# 6. Run Pre-Built Reports
uv run scripts/google_analytics.py report overview --db path/to/database.db
uv run scripts/google_analytics.py report top-pages --db path/to/database.db
uv run scripts/google_analytics.py report channels --db path/to/database.db
uv run scripts/google_analytics.py report geo --db path/to/database.db
uv run scripts/google_analytics.py report events --db path/to/database.db
uv run scripts/google_analytics.py report outbound --db path/to/database.db
uv run scripts/google_analytics.py report milestone-impact --db path/to/database.db

# 7. Execute Ad-Hoc SQL Query
uv run scripts/google_analytics.py query "SELECT page_path, total_views, total_users, avg_dwell_sec, avg_bounce_pct FROM v_page_performance LIMIT 10" --db path/to/database.db

If --db is omitted, the script defaults to google_analytics.db in the current working directory.


🗄️ Database Schema & Relational Structure #

The database maintains 7 relational tables and 7 analytical views. Detailed DDL and schema definitions are in references/schema.md.

Tables #

  1. daily_pages: Granular daily page metrics by URL, country, device, and traffic source.
    • Key columns: id (PK), property_id, date, page_path, page_title, country, device_category, source_medium, screen_page_views, active_users, sessions, user_engagement_duration, bounce_rate, raw_json, synced_at.
  2. daily_traffic: Acquisition channels and source/medium pairs.
    • Key columns: id (PK), property_id, date, session_source_medium, session_default_channel_group, country, device_category, sessions, active_users, new_users, engaged_sessions, user_engagement_duration, bounce_rate, raw_json, synced_at.
  3. daily_events: User interaction event stream (scroll, click, first_visit, user_engagement, page_view).
    • Key columns: id (PK), property_id, date, event_name, page_path, country, device_category, event_count, total_users, raw_json, synced_at.
  4. outbound_clicks: External link exit destinations and click counts.
    • Key columns: id (PK), property_id, date, link_url, page_path, country, event_count, total_users, raw_json, synced_at.
  5. properties: Verified GA4 property metadata, timezone, and settings.
    • Key columns: property_id (PK), name, account_id, display_name, industry_category, time_zone, currency_code, service_level, raw_json, last_synced_at.
  6. site_milestones: Release milestones and publication events.
    • Key columns: commit_hash (PK), event_date, title, description, category, scope, author, created_at.
  7. sync_history: Audit log of sync executions and row counts.
    • Key columns: id (PK), property_id, sync_type, start_date, end_date, pages_synced, traffic_synced, events_synced, outbound_synced, status, error_message, started_at, finished_at.

📊 Analytical SQL Views #

View NameDescriptionKey Columns
v_daily_summaryDaily aggregated traffic metricsdate, total_sessions, total_active_users, total_page_views, total_engagement_min, avg_bounce_pct
v_page_performancePage rollup with views, active users, dwell time, and bounce ratepage_path, page_title, total_views, total_users, total_sessions, avg_dwell_sec, total_dwell_min, avg_bounce_pct
v_channel_performanceAcquisition channel breakdownchannel_group, source_medium, total_sessions, total_users, total_new_users, total_engaged_sessions, engagement_rate_pct, avg_dwell_sec, engaged_dwell_sec, total_dwell_min, avg_bounce_pct
v_geo_breakdownCountry traffic and dwell timecountry, total_sessions, total_users, total_page_views, avg_dwell_sec, avg_bounce_pct
v_events_summaryAggregate event countsevent_name, total_events, total_users
v_outbound_linksOutbound destination rankingslink_url, total_clicks, total_users, referring_pages_count
v_milestone_impactPre vs. Post milestone comparisonmilestone_title, milestone_date, cohort, days_tracked, total_views, total_users, total_sessions, avg_engagement_sec, avg_bounce_pct

🔍 SQL Analytics Recipes #

Pre-tested SQL query recipes are documented in references/queries.md.

1. Top Landing Pages by Active Dwell Time #

sql
SELECT
    page_path,
    page_title,
    total_views,
    total_users,
    avg_dwell_sec || 's' AS avg_dwell,
    total_dwell_min || 'm' AS total_dwell,
    avg_bounce_pct || '%' AS bounce_pct
FROM v_page_performance
ORDER BY total_dwell_min DESC
LIMIT 15;

2. Category / Subdirectory Rollup #

sql
SELECT
    CASE
        WHEN page_path LIKE '/docs/%' THEN 'Docs'
        WHEN page_path LIKE '/blog/%' THEN 'Blog'
        WHEN page_path = '/' THEN 'Homepage'
        ELSE 'Other'
    END AS category,
    COUNT(DISTINCT page_path) AS page_count,
    SUM(total_views) AS total_views,
    SUM(total_users) AS total_users,
    ROUND(SUM(total_dwell_min), 1) AS total_dwell_min
FROM v_page_performance
GROUP BY category
ORDER BY total_views DESC;

🧭 Analytical Guardrails & Engagement Methodology #

When analyzing GA4 metrics, apply these four structural checks to avoid misinterpreting raw totals:

  1. Two-Stage Engagement Decomposition (avg_dwell_sec vs. engaged_dwell_sec + Interaction Density):
    • GA4 calculates user_engagement_duration as the elapsed time between session_start and the final engagement event. Single-page visits that exit without firing a follow-up event record 0s duration.
    • Always report Engaged Session Dwell (engaged_dwell_sec) alongside raw session dwell (avg_dwell_sec), and pair time metrics with Interaction Density (scroll completion and outbound_clicks per user) using Recipe #11 in references/queries.md.
  2. Unengaged Direct Anomaly Gate (Bot & Scraper Filtering):
    • Headless crawlers, link previewers, and cloud monitors typically omit the HTTP Referer header (landing in Direct) and exit before firing a second event (<2s dwell and <15% engagement rate).
    • Before reporting site-wide dwell or bounce averages, inspect Direct traffic for localized (month, country, device_category) spikes with near-zero engagement (Recipe #9 in references/queries.md) and segment human discovery metrics separately from unengaged Direct spikes.
  3. Generative Search Channel Split (Organic vs. AI Assistants vs. Social):
    • GA4’s built-in session_default_channel_group groups AI assistants (chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com) into generic Referral or Organic Search.
    • Use session_source_medium pattern matching (Recipe #8 in references/queries.md) to separate Generative AI, Organic Search, Social Referrals, and Web Referrals.
  4. Custom Telemetry Discovery Gate (daily_events Inventory):
    • Standard page views (v_page_performance) query only daily_pages. Any custom instrumentation (such as scroll_25, scroll_50, scroll_75, code_copy, or form_submit) lands in daily_events.
    • Always run an event inventory query (WHERE event_name NOT IN ('page_view', 'user_engagement', 'session_start', 'first_visit'), Recipe #10 in references/queries.md) before concluding page-level analysis so custom events are joined into your findings.

⚠️ Cross-Tool Alignment: GA4 Organic Search vs. Search Console Property Totals #

When cross-referencing GA4 acquisition metrics with Google Search Console data:

  1. Multi-Engine Organic Reach: GA4 Organic Search aggregates landing sessions and active users across all search engines (Google Search, Bing, DuckDuckGo, Ecosia, Yahoo, AI search engines). Search Console exclusively measures Google Search impressions and clicks.
  2. Session Arrivals vs. SERP Clicks: GA4 tracks 100% of landing sessions without query-level privacy truncation. In contrast, Search Console API keyword exports filter out “anonymized queries”.
  3. Property-Level Reconciliation: For organic traffic reporting, GA4 v_channel_performance (filtering for channel_group = 'Organic Search') naturally aligns with Search Console property-level totals (Google Search Console property-level totals and GSC Web UI Performance cards), while Search Console’s search_performance table provides the granular ranking breakdown for identifiable keywords.

📚 Progressive Disclosure & References #

  • Full DDL Schema Reference: references/schema.md — Complete SQL table definitions, column types, constraints, and views.
  • SQL Query Cookbook: references/queries.md — Tested SQL recipes for reading depth, acquisition channels, and exit destinations.
  • Authentication Guide: references/setup_auth.md — Google Cloud ADC login, API enablement, and GA4 property permissions.