GA4 limits reference card

What this is

Every limit and quota Google Analytics 4 enforces that changes how you implement, register, or query data — collection caps, custom-definition quotas, cardinality, export ceilings, and API tokens — on one printable card. Values are for standard properties with 360 differences called out; each number was read from the official page listed under Sources on the date in the header.

Collection limits

Enforced at collection time, not by the interface. Over-limit names and values do not arrive intact.

Custom definitions and reporting quotas

• Deleting (archiving) a custom definition at the quota does not free the slot immediately — allow about 48 hours before registering a replacement.
• Terminology: “conversions” were renamed key events; “conversions” now refers to Google Ads conversions.

Data retention

Retention governs event-level and user-level data — what explorations and funnel reports query. Standard aggregated reports are not affected. The retention page adds that “Large and XL properties are limited to 2 months”; how that classification applies to a given property is worth checking on the retention settings page itself.

Cardinality and the (other) row

• A high-cardinality dimension is one that exceeds 500 unique values in one day.
• GA4 applies a cardinality limit of 50,000 values, beyond which cardinality control kicks in and detail collapses into the (other) row.
• Exact per-report row limits are not published for standard properties — the docs give only one illustrative figure (100k rows for the table behind Pages and screens). Treat per-report limits as undocumented.
• 360 expanded data sets raise a report’s underlying table to up to 2 million rows: max 100 expanded data sets per property, about 48 hours to apply, and the report must have at most 6 dimensions and 12 metrics.

BigQuery export

• Streaming export is best-effort: no completeness SLO, gaps possible, and it excludes some traffic-source data for new users and sessions — use the daily export for attribution-sensitive work.
• Streaming cost: about $0.05 per GB, roughly 600,000 events per GB.
• What happens when a standard property exceeds the 1M/day cap is not precisely documented; the docs point to filtering the export to stay under it.

Structure and API quotas

Data API v1 core-method tokens (Realtime quotas mirror these; daily quotas reset at midnight PST):

Measurement Protocol limits

• 25 events per request; POST body under 130 kB.
• Per event: 25 parameters; event and parameter names at most 40 characters; values at most 100 characters (500 on 360).
• 25 user properties per request; names at most 24, values at most 36 characters.
• timestamp_micros may backdate events up to 72 hours.

Gotchas

1. Long event names look wrong or collide. Cause: the 40-character cap is enforced at collection, so anything longer does not arrive as sent. Fix: keep names short and lint them in CI; in BigQuery, LENGTH(event_name) = 40 is a strong truncation signal.
2. A report dimension collapses into (other). Cause: the dimension went high-cardinality (over 500 unique values in a day) and the 50,000-value cardinality limit applied. Fix: move detail to BigQuery, stop registering high-cardinality parameters as custom dimensions, or use 360 expanded data sets.
3. Deleted a custom dimension but cannot add its replacement. Cause: archived definitions free their quota slot only after about 48 hours. Fix: plan registrations ahead; audit the 50-slot budget before adding.
4. Explorations show less history than standard reports. Cause: retention (default 2 months) applies to the event-level data explorations query, not to aggregated reports. Fix: set retention to 14 months (standard maximum) on day one.
5. Streaming-export rows are missing traffic-source data. Cause: documented exclusion for new-user and new-session attribution fields in streaming export. Fix: run attribution queries on the daily export tables.
6. Measurement Protocol batches silently underdeliver. Cause: more than 25 events per request, body over 130 kB, or timestamps older than 72 hours. Fix: chunk requests and validate against the limits above before sending.

Quick recipes

Find what is eating the 1M/day export cap:

SELECT event_name, COUNT(*) AS events
FROM `project.analytics_123456.events_*`
WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', DATE_SUB(CURRENT_DATE(), INTERVAL 1 DAY))
GROUP BY event_name
ORDER BY events DESC;

Flag event names at the 40-character cap (likely truncated upstream):

SELECT DISTINCT event_name, LENGTH(event_name) AS len
FROM `project.analytics_123456.events_*`
WHERE _TABLE_SUFFIX = FORMAT_DATE('%Y%m%d', DATE_SUB(CURRENT_DATE(), INTERVAL 1 DAY))
  AND LENGTH(event_name) >= 36
ORDER BY len DESC;

Guard parameter names and values client-side before they hit the caps:

const LIMITS = { name: 40, value: 100 }; // 360 properties: value 500
function safeParams(params) {
  return Object.fromEntries(
    Object.entries(params).map(([k, v]) => [
      k.slice(0, LIMITS.name),
      typeof v === 'string' ? v.slice(0, LIMITS.value) : v,
    ])
  );
}
gtag('event', 'search', safeParams({ search_term: query }));

Measurement Protocol batch skeleton (25 events max, body under 130 kB, backdating within 72 h):

curl -s "https://www.google-analytics.com/mp/collect?measurement_id=G-XXXXXXX&api_secret=SECRET" \
  -H "Content-Type: application/json" \
  -d '{
    "client_id": "123456.7654321",
    "events": [
      { "name": "tutorial_begin", "params": { "source": "crm" } }
    ]
  }'

Related

• Hub: GA4 guides
• GA4 event naming cheat sheet
• Server-side GTM setup checklist

Sources

Official pages these values were read from (2026-07-19):

• Event collection limits
• Custom dimensions and metrics
• GA4 360 comparison
• Configuration limits
• Data retention
• High-cardinality dimensions
• BigQuery Export
• Account and property limits
• Data API v1 quotas
• Measurement Protocol reference

The January–June 2026 GA4 release notes contain no changes to the limits above. Items the docs leave unstated (per-report row limits, over-cap export behavior) are flagged inline rather than guessed.