Copy-paste recipes for the 6-7 reports every site owner actually wants:
DAU/MAU/WAU, retention cohort, top pages by source, channel attribution,
conversion funnel, geo breakdown, device split. Each recipe is a fully-formed
Data API request. Trigger with "GA4 DAU", "GA4 retention", "GA4 top pages",
"GA4 funnel", "GA4 channel report", "common GA4 reports".
Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Copy-paste recipes for the 6-7 reports every site owner actually wants:
DAU/MAU/WAU, retention cohort, top pages by source, channel attribution,
conversion funnel, geo breakdown, device split. Each recipe is a fully-formed
Data API request. Trigger with "GA4 DAU", "GA4 retention", "GA4 top pages",
"GA4 funnel", "GA4 channel report", "common GA4 reports".
Recipes for the reports that get asked for ~95% of the time. Each one is a complete runReport you can paste, change PROPERTY_ID, and run. Prerequisite: ga4-auth-setup done.
The setup block (same for every recipe):
import os
from google.analytics.data_v1beta import BetaAnalyticsDataClient
from google.analytics.data_v1beta.types import (
RunReportRequest, DateRange, Metric, Dimension,
FilterExpression, Filter, OrderBy,
)
PROPERTY = f"properties/{os.environ['GA4_PROPERTY_ID']}"
client = BetaAnalyticsDataClient()
1. Daily Active Users (DAU) — 30-day rolling
req = RunReportRequest(
property=PROPERTY,
date_ranges=[DateRange(start_date="30daysAgo", end_date="yesterday")],
metrics=[Metric(name="activeUsers")],
dimensions=[Dimension(name="date")],
order_bys=[OrderBy(dimension=OrderBy.DimensionOrderBy(dimension_name="date"))],
)
resp = client.run_report(req)
for r in resp.rows:
print(f"{r.dimension_values[0].value}{r.metric_values[0].value}")
Why yesterday, not today: today's number is incomplete and will keep climbing through the day. For a clean rolling DAU, end the window at yesterday.
2. MAU / WAU — rolling unique users
GA4 doesn't expose MAU as a single metric — you compute it from the same activeUsers rolled up over a wider date range. The trick: a single-row report with no date dimension returns the unique count over the entire window (de-duplicated across days).
GA4 has a built-in cohort exploration in the UI but the Data API doesn't expose it cleanly. The workaround: query DAU per week and compute rolling overlap. The cheap approximation:
# Weekly active users for the last 8 weeks
req = RunReportRequest(
property=PROPERTY,
date_ranges=[DateRange(start_date="56daysAgo", end_date="yesterday")],
metrics=[Metric(name="activeUsers")],
dimensions=[Dimension(name="isoYearIsoWeek")],
order_bys=[OrderBy(dimension=OrderBy.DimensionOrderBy(dimension_name="isoYearIsoWeek"))],
)
resp = client.run_report(req)
for r in resp.rows:
print(f"{r.dimension_values[0].value}{r.metric_values[0].value}")
For true cohort retention (e.g. "of users acquired in week N, what % came back in week N+1, N+2, N+3"), you need event-level data — use ga4-bigquery-export and write the cohort SQL directly. The Data API can't express the join.
GA4 funnels via API: query each step as a separate runReport filtered by the event that defines the step, then divide.
defstep_users(event_name, days_ago=7):
returnint(client.run_report(RunReportRequest(
property=PROPERTY,
date_ranges=[DateRange(start_date=f"{days_ago}daysAgo", end_date="yesterday")],
metrics=[Metric(name="activeUsers")],
dimension_filter=FilterExpression(filter=Filter(
field_name="eventName",
string_filter=Filter.StringFilter(
match_type=Filter.StringFilter.MatchType.EXACT,
value=event_name,
),
)),
)).rows[0].metric_values[0].value)
# Example funnel: landed → engaged → signed up → purchased
steps = [
("session_start", step_users("session_start")),
("user_engagement", step_users("user_engagement")),
("sign_up", step_users("sign_up")),
("purchase", step_users("purchase")),
]
top = steps[0][1] or1print(f"{'Step':<20}{'Users':>10}{'% of top':>10}")
for name, count in steps:
print(f"{name:<20}{count:>10,}{count/top:>9.1%}")
Limitation: this counts users who fired the event at any point in the window, NOT users who progressed through the funnel in order. For ordered funnels (true sequencing), use BigQuery export or the GA4 UI's Exploration → Funnel report.
A common signal: if one country dominates with low engagement + high bounce, it's often bot traffic from that country's cloud-host hubs (Singapore, Vietnam, China data centers are the usual suspects).
When the Data API isn't enough
Three reasons to graduate to BigQuery export:
Sampling — your queries hit resp.metadata.data_loss_from_other_row=True. Sampled = approximate. BQ export = exact.
Custom event analytics — joining event-level data across sessions, computing retention cohorts, building attribution models. SQL is the only sensible tool.
Cost — Data API has daily quotas; BQ is pay-per-query (free for small properties, cheap up to ~100M events/day).
See ga4-bigquery-export for the setup.
Related skills
ga4-auth-setup — prerequisite
ga4-data-api-query — the underlying API the recipes here use
ga4-realtime-api — for "right now" data instead of any of the above
ga4-bigquery-export — when these recipes hit their limits
