Build a runReport request against the GA4 Data API v1 — pick valid metric/dimension
combinations, set date ranges that respect data-freshness limits, apply filters,
paginate large result sets, handle sampling thresholds. Trigger with "query GA4",
"GA4 Data API", "runReport", "fetch GA4 metrics", "GA4 pageviews", "GA4 sessions".
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Build a runReport request against the GA4 Data API v1 — pick valid metric/dimension
combinations, set date ranges that respect data-freshness limits, apply filters,
paginate large result sets, handle sampling thresholds. Trigger with "query GA4",
"GA4 Data API", "runReport", "fetch GA4 metrics", "GA4 pageviews", "GA4 sessions".
The Data API v1 is the canonical read path for GA4. One endpoint (runReport) covers most use cases. Two paths matter for picking the right query: dimensions describe rows (date, page, source), metrics describe values (sessions, users, events). Not every combination is valid — see "Compatibility" below.
Prerequisite: auth working (see ga4-auth-setup).
The minimum viable query
from google.analytics.data_v1beta import BetaAnalyticsDataClient
from google.analytics.data_v1beta.types import (
RunReportRequest, DateRange, Metric, Dimension,
)
client = BetaAnalyticsDataClient()
req = RunReportRequest(
property="properties/123456789", # YOUR property ID (digits only)
date_ranges=[
DateRange(start_date="30daysAgo", end_date="today"),
],
metrics=[Metric(name="activeUsers")],
dimensions=[Dimension(name="date")],
)
resp = client.run_report(req)
for row in resp.rows:
date = row.dimension_values[0].value # YYYYMMDD string
users = row.metric_values[0].value # numeric stringprint(f"{date}: {users}")
That's the full skeleton. Everything below extends this shape.
The 12 metrics worth knowing
Metric
What it counts
Notes
activeUsers
Unique users with engagement in the window
The "users" people mean by default
newUsers
First-seen users in the window
totalUsers
All users (engaged or not) — superset of activeUsers
sessions
Sessions started in the window
Re-engages after 30min inactivity
engagedSessions
Sessions ≥10s OR ≥2 pageviews OR ≥1 conversion
The "good" sessions
screenPageViews
Pageviews + app screenviews combined
What people mean by "pageviews"
eventCount
Total event count (every event, not just page_view)
Often misleadingly large
bounceRate
(sessions - engagedSessions) / sessions
Lower is better
averageSessionDuration
Avg seconds per session
Across sessions, not engagedSessions
eventsPerSession
eventCount / sessions
conversions
Events flagged as conversions in the property setup
Property-specific
totalRevenue
Sum of purchase event revenue
Currency = property default
bounceRate and averageSessionDuration are ratios — don't SUM them across rows; they're already aggregated within each row's group.
The 12 dimensions worth knowing
Dimension
Cardinality
When to use
date
Low (1/day)
Time series
dateHour
Med
Intra-day patterns
pagePath
High
Top-pages reports
pageTitle
High
When path is opaque (e.g. SPA hash routes)
sessionSource / sessionMedium
Med
Attribution
sessionDefaultChannelGrouping
Low (~12 channels)
High-level traffic source breakdown
country / region / city
Med / Med / High
Geo
deviceCategory
Low (desktop/mobile/tablet)
browser / operatingSystem
Med
Tech audit
landingPage
High
Entry-page reports
eventName
Med
Event-level breakdowns
customEvent:<name>
Property-specific
If you defined custom dimensions in the property setup
Compatibility — not every (dim, metric) combo is valid
GA4 enforces a compatibility matrix at the API level. If you ask for sessions + customEvent:purchaseId together you may get an empty result or a 400 INVALID_ARGUMENT. Two rules cover ~90% of cases:
User-scoped vs session-scoped vs event-scoped dimensions don't always mix with each other's metrics. Stick to dimensions in the same scope as your headline metric where possible.
High-cardinality custom dimensions can trigger sampling. GA4 will silently sample if a single query touches more than the property's data-quota threshold; the response includes metadata.dataLossFromOtherRow=true. Check it.
If you're unsure, query the compatibility metadata endpoint:
Filters are nested expressions. The common case: filter rows by a dimension value.
from google.analytics.data_v1beta.types import (
FilterExpression, Filter, FilterExpressionList,
)
# Just pages under /docs/
docs_only = FilterExpression(
filter=Filter(
field_name="pagePath",
string_filter=Filter.StringFilter(
match_type=Filter.StringFilter.MatchType.BEGINS_WITH,
value="/docs/",
case_sensitive=False,
),
),
)
# AND combine: organic search AND not from referrer "spam.com"
combined = FilterExpression(
and_group=FilterExpressionList(expressions=[
FilterExpression(filter=Filter(
field_name="sessionMedium",
string_filter=Filter.StringFilter(
match_type=Filter.StringFilter.MatchType.EXACT,
value="organic",
),
)),
FilterExpression(not_expression=FilterExpression(filter=Filter(
field_name="sessionSource",
string_filter=Filter.StringFilter(
match_type=Filter.StringFilter.MatchType.EXACT,
value="spam.com",
),
))),
]),
)
req = RunReportRequest(
property="properties/123456789",
date_ranges=[DateRange(start_date="30daysAgo", end_date="today")],
metrics=[Metric(name="sessions")],
dimensions=[Dimension(name="pagePath")],
dimension_filter=docs_only,
)
Use metric_filter for filtering by metric (e.g. only rows where sessions > 100). Same shape.
Date ranges
Form
Meaning
"2026-05-01"
Absolute (ISO date)
"30daysAgo"
Relative — N days before today
"yesterday", "today"
Named relative
"NdaysAgo" to "today"
Standard rolling window
GA4 has 48-hour data freshness — today's numbers fluctuate; yesterday's settle ~24h after midnight in the property's timezone; numbers older than 48h are stable. Don't draw conclusions from "today" alone.
Multiple date_ranges in one request gives you a comparison report:
The response will have dateRange as an extra dimension on each row.
Pagination
req = RunReportRequest(
# ... as above
limit=10_000, # max 250_000 per request
offset=0,
)
resp = client.run_report(req)
# resp.row_count is the TOTAL matching rows; resp.rows is the current pagewhile resp.row_count > req.offset + len(resp.rows):
req.offset += len(resp.rows)
resp = client.run_report(req)
# process resp.rows
For result sets over ~1M rows, use ga4-bigquery-export instead.
Sampling — always check
resp = client.run_report(req)
if resp.metadata.data_loss_from_other_row:
print("WARNING: data was sampled. Tighten date range, drop high-cardinality dimensions, or use BigQuery export for unsampled data.")
If sampled, results are statistically valid but not exact. For exact counts, BigQuery export is the only path.
Common errors
Error
Cause
Fix
400 INVALID_ARGUMENT: dimension X is incompatible with metric Y
Compatibility matrix violation
Use check_compatibility to find a valid combination
400 The request must contain at least one valid dimension
All dimensions in the list are invalid (typo, deprecated name)
