Sessions and Attribution

Sessions in GA4 are not collected — they are derived. GA4 collects individual events, and sessions are reconstructed from those events during processing. This distinction explains many of the differences between GA4 and Universal Analytics session counts.

How GA4 sessions work

A session in GA4 is a group of events from a single user within a continuous time window. Sessions are identified by the ga_session_id event parameter — an integer timestamp of when the session started.

A new session begins when:

• A user arrives with no active session (first visit, or after a 30-minute inactivity gap)
• A user arrives with a new campaign source (utm_source, utm_medium, etc.), even within the 30-minute window
• Midnight passes (in some configurations — controlled in Admin → Reporting identity)

A session ends when:

• The user is inactive for 30 minutes (default, configurable)
• The session has lasted 24 hours (hard limit)

Session timeout

The default 30-minute timeout can be changed at Admin → Data streams → [stream] → Configure tag settings → Session timeout. Range: 1 minute to 7 hours, 55 minutes.

Shorter timeout = more sessions, higher bounce rates, shorter average duration. Longer timeout = fewer sessions, lower bounce rates, longer average duration.

Most implementations should keep the default unless users routinely have inactivity gaps longer than 30 minutes during legitimate sessions (e.g., reading long documents, watching videos).

Session identifiers

When a new session begins, GA4 fires session_start with:

• ga_session_id — a numeric timestamp (the session identifier)
• ga_session_number — which session this is for the user (1, 2, 3, etc.)

Both parameters attach to every subsequent event in the session. The canonical unique session identifier in BigQuery is the combination of user_pseudo_id + ga_session_id.

Engaged sessions

An “engaged session” meets at least one of:

• Lasted longer than 10 seconds with the page in focus
• Had 2 or more page views or screen views
• Had at least 1 key event

Engagement rate = engaged sessions ÷ total sessions. This is GA4’s primary engagement signal.

Tip

GA4 has a bounce rate field — it is the inverse of engagement rate. A GA4 bounce is a non-engaged session, not a single-page session. Do not compare GA4 bounce rates to Universal Analytics bounce rates; the definitions are incompatible.

Session counting differences from Universal Analytics

Expect different session counts between GA4 and UA for the same site and date range.

Midnight session splits — UA starts a new session at midnight by default. GA4 does not. A user active from 11:45 PM to 12:15 AM = 2 UA sessions, 1 GA4 session.

Cross-device stitching — When User ID is configured, GA4 merges sessions across devices. UA counted cross-device visits as separate sessions. GA4 may count fewer sessions per user.

Bot filtering — GA4 automatically filters known bot traffic. UA required an opt-in setting.

Do not use session count as a migration validation metric between GA4 and UA. Use specific, well-defined event counts instead.

Attribution in GA4

Attribution assigns credit for a key event to a traffic source. GA4 supports multiple attribution models.

Attribution models

GA4 currently supports two attribution models:

Model	How credit is assigned
Data-driven (default)	ML distributes fractional credit based on actual impact of each touchpoint
Last click	100% to the last touchpoint before key event

Note: First click, Linear, Time decay, and Position-based models were removed from GA4 in November 2023. For multi-touch attribution analysis, use BigQuery SQL-based models or the GA4 BI Connector to implement custom attribution.

Data-driven attribution requires sufficient volume — Google recommends 30+ key events per day. Below that threshold, it falls back to last click.

Configure the default model in Admin → Attribution settings.

Lookback windows

Event type	Default lookback window
Acquisition events (first_visit, first_open)	30 days
All other key events	90 days

Change in Admin → Attribution settings. Shorter windows concentrate credit on recent channels.

Attribution changes apply retroactively

Unlike Universal Analytics, GA4 attribution model changes restate historical data in the Advertising reports. Changing from last-click to data-driven will change how historical channel performance is reported. This is intentional — you can compare models on the same data — but be careful when reporting to stakeholders.

Standard reports use last non-direct click

The Traffic acquisition and User acquisition reports use last non-direct click attribution regardless of your attribution settings. Your attribution settings affect:

• The Advertising section
• The Data API when you request a specific attribution model
• Looker Studio GA4 connector

Caution

A common confusion: “I configured data-driven attribution but my Traffic acquisition report looks the same.” Yes — standard acquisition reports always use last non-direct click. Data-driven numbers appear in the Advertising section only.

Source/medium attribution logic

GA4 determines traffic source in this priority order:

1. UTM parameters — explicit utm_source, utm_medium, etc.
2. Auto-tagging — Google Ads gclid
3. Document referrer — the referring domain
4. Direct — no referrer, no campaign parameters

Campaign parameters persist for the session. All events in a session are attributed to the source that started it.

Session reconstruction in BigQuery

Count sessions using user_pseudo_id + ga_session_id:

SELECT
  event_date,
  COUNT(DISTINCT CONCAT(
    user_pseudo_id, '.',
    CAST(
      (SELECT value.int_value FROM UNNEST(event_params) WHERE key = 'ga_session_id')
      AS STRING
    )
  )) AS sessions
FROM `project.dataset.events_*`
WHERE _TABLE_SUFFIX BETWEEN '20240101' AND '20240131'
GROUP BY event_date
ORDER BY event_date

Do not count sessions by counting session_start events — some sessions may not have one due to data collection gaps or identity stitching.

Common mistakes

Expecting campaign attribution to persist across sessions

Campaign attribution is session-scoped. A user from an email campaign today who returns directly tomorrow is attributed to Direct on the second visit. Use the Advertising section or BigQuery for multi-touch attribution.

Comparing GA4 and UA session counts for validation

Session counts differ by design. Validate GA4 implementation with specific event counts, not total session volume.

Not adjusting session timeout for long-form content

A site where users read 20-minute articles needs a longer session timeout. The default 30 minutes will split legitimate sessions, inflating session counts and deflating per-session metrics.

Related articles

Attribution Settings Configure default attribution models and lookback windows.

Session Reconstruction Rebuild sessions from raw events in BigQuery.

Attribution Queries Implement multi-touch attribution models in SQL.

Data Discrepancies Why GA4 numbers don't match other tools.