What’s new

If a visitor typed your address and a few minutes later clicked through from a social post, the whole visit used to count as direct and the social channel got no credit. Now, arriving again mid-session from another external site, or through a link with different UTM tags, starts a new session under the new source. Internal navigation, reloads and returns from payment pages are unaffected.

The rule is on by default for new projects and can be switched off under Sessions & channels in website settings. Channels and Referring sites in the Sources report also gain an "every arrival" method that shows how many arrivals each source actually produced. More in How sources and channels are classified.

We reordered how events are stored. Events are written first; identity merging, first-touch attribution and history backfill moved to separate background jobs, and bot and replay data no longer wait behind identity processing. Under load testing, the P95 delay from the collector receiving an event to it appearing in reports fell from about 8 seconds to about 2.

In practice: after installing, your own visit shows up in Real-time almost instantly, and the Sources report reflects a new campaign's traffic sooner.

When a paid plan ends without renewal, is replaced by a lower one or is withdrawn after a refund, there is now a 30-day grace period. Nothing is deleted in those 30 days, you get an in-app and email reminder on day 23, and renewing the same or a higher plan within the period keeps everything. Only after 30 days is project retention reduced to the current plan's limit.

Pro and above can also buy event add-on packs: choose how many extra events per month, for this month only or until the plan ends. This month is priced by the days left, and each month's add-on is valid for that month. Details in Plans, event quotas and what happens when you run out.

The top sources card in Real-time used to count the source of every event in the last 30 minutes. Internal navigation and custom events have no source, so they showed up as direct, and one visitor could appear in two rows. It now reads sessions active in the last 30 minutes, using the same first-source rule as the Sources report, so each visitor appears once.

The same day we fixed two display issues on the ad ROI page: the channel column squeezed to one character per line, and overlapping labels for small channels on the quadrant chart.

Sessions and visitor profiles now show device brand, model, the real browser name and screen resolution, and visitor profiles list the devices a person uses most. Device detection moved to a more complete parsing library.

We also corrected a common misreading: mobile browsers in "desktop site" mode pretend to be Windows, Linux or Mac and used to be recorded as computers. They are now recorded as the real Android or iOS device. The visitor identifier is calculated exactly as before, so no visitor is split in two.

You can now define a set of conversion milestones for a project — sign-up, activation, first payment — and the first time a user is identified we settle which channel they came from and record when they reach each milestone. The new Acquisition page lists people and conversion for each milestone by channel and campaign, as a table, a curve or a matrix.

The server-side events API also accepts the visitor's landing source, so channels still line up when sign-up happens on your server. Configure it under Conversion milestones & attribution in business settings.

On a phone, the session list, visitor list and visitor details now show one card per row instead of a table you have to scroll sideways. The visitor activity calendar and per-region privacy settings were adapted for narrow screens too.

Carrier names are now unified by network AS number. The same carrier sometimes appeared under a local name and sometimes under its English organisation name, which split it into two rows in carrier reports.

Every place in the console where you choose an event — funnel steps, retention, goals, monitors and more — now uses the same picker: searchable, grouped by category. You can manage categories yourself too: create, rename, delete and drag to reorder, and built-in categories can be added back after deletion.

Once you have many events, a round of tidying into categories makes building analyses much quicker.

IP shortening in the privacy center is now set separately: IPv4 by bytes (none, drop one part, drop two parts) and IPv6 by prefix (none, /64, /56, /48), at project level and in regional rules. The default is no shortening.

If you have had EU or UK visitors in the last 30 days, or already have rules for those regions, the privacy center suggests shortening for EU/UK only and can apply it in one click. Existing settings were migrated to behave exactly as before.

The session and visitor lists now show anything with activity in the selected range. They used to filter by session start only, so a visitor who kept a tab open past midnight was missing from the next day's list. Metrics are unchanged: a session crossing midnight still counts on the day it started.

Autocapture gained a project-level switch and page scope settings. To skip a page, add data-autocapture="false" to its script tag; you can also clear click data already collected.

Heatmaps are now drawn over your actual pages, and you can switch the traffic source to see where visitors from different channels click.

Alongside the common international ad click IDs, the click parameters that major Chinese ad platforms add to landing URLs are now recognised, so ad links without UTM tags still land in the right paid channel. 14 ad platforms are supported in total.

Each user's first source is now stored as well, visible on their profile and in attribution analysis.