Methodology
How Gauss counts
Every metric in Gauss follows a rule you can read. Where a number has limits, we say so.
Visitors
A visitor is a unique daily identifier on one site. For periods longer than a day, the report shows visitors per day, summed over the period, because Gauss does not recognise a visitor from one day to the next. Days are counted in UTC for the identifier and in the site's time zone for reports, so sites far from UTC may see slightly higher daily counts.
Visits
A visit is a series of events from one visitor with gaps of less than 30 minutes. A visit that crosses midnight UTC counts as two.
Pageviews
The number of page_view events, including page changes in single-page apps.
Bounce rate
We use the GA4 definition. A visit is engaged if it lasted at least 10 seconds of visible time, had at least two pageviews, or included a conversion. Bounce rate is the share of visits that were not engaged.
Time on page
Only time while the tab is visible counts. A tab in the background adds nothing. Mobile browsers sometimes don't send the last event when a tab is closed, so time on the last page of a visit can be missing.
Channels
Each visit gets one channel, checked in order: paid, email, search, social, referral, direct. An ad click is detected from parameters such as gclid or yclid, and only the fact of the click is stored.
Paths, funnels and attribution
Paths and funnels follow a visitor within one day. A lead left three days after an ad click is credited to the channel of the visit where it happened. Deals and payments from your CRM are linked to the original lead or order by its ID for up to 90 days.
Bots
Known bots are filtered by user agent, data center IP ranges and signs of headless browsers. Events from localhost and site builder previews are not counted.
Accuracy
We are measuring the error against server logs on beta sites. The measured figure and the method will be published here.
AI recommendations Coming soon
Code finds the facts in aggregated reports: significant changes, funnel bottlenecks, channels whose share of value differs from their share of traffic. Segments with fewer than 100 visits in the period are skipped. The language model receives only these aggregated facts and segment names, and writes them up. An insight with a number that is not in the facts is discarded.