An AI agent for Google Analytics 4
Tracking Whisper reads a GA4 property through the Google Analytics APIs: its settings, key events, custom definitions and reports. It answers questions about the numbers, explains drops, checks which events arrive against your measurement plan, watches the daily numbers for anomalies, and prepares setting changes that a person approves before they are applied.
GA4 analysis in your own words
Ask in your own words, for example why purchases dropped last week or which channel brought the most key events. The agent runs the GA4 reports it needs and shows what it read. It knows that GA4 is still filling in today and yesterday, and says so when a number may still change.
- Sessions, users, events and key events by day, channel, page or device
- Purchases and revenue over any period
- Questions in any language, answered in the same one
A GA4 audit of the setup
Key events, custom dimensions, data retention and the events that actually arrived are read and compared with what the site and the container send. The measurement plan sets each event the site should have against what GA4, the container and the latest site check show today.
- Events that the container sends but GA4 never receives
- Key events that are not marked, or marked twice
- Ecommerce events without value, currency or items
Explain a drop
The explain-a-drop helper reads twelve weeks of GA4, day by day. Code finds the day a number left its usual level, where on the way to it the loss begins, and in which channels and devices it sits. It then looks at what went live or was raised around that day, and the explanation is written from those facts.
GA4 anomaly detection
Monitoring reads yesterday's numbers every morning at nine on the property's own clock, and sets each one against the same day of the week in the five weeks before. A number that looks wrong is read again a few hours later, because GA4 is still filling in yesterday in the morning. Only what is still wrong then becomes an alert.
- Sessions, key events, purchases and revenue
- Events such as page_view, add_to_cart, begin_checkout and generate_lead
- Read by the hour for a week after a new container version
- About three weeks of history before a number is judged
Reports you can send
The monthly tracking report covers what changed, what broke and what was fixed in a month, with the month's numbers. The performance report puts GA4 beside Google Ads, Search Console and Meta ads, with a mark on every number for how far it can be trusted.
Changes with approval
GA4 has no draft, so the agent prepares a change as a proposal. It is applied only on the Publish level, after a person approves it, and undone on a rollback where Google allows it.
- Create or delete a key event
- Create or archive a custom dimension or metric
- Data retention and enhanced measurement settings
Questions
Does it need BigQuery?
No. It reads GA4 through the Analytics APIs. BigQuery is an extra source when you want the raw events.
Can it explain a drop in conversions?
Yes. The explain-a-drop helper reads the days before and after, the tracking changes in between and the channels, and says what changed.
How does GA4 anomaly detection avoid false alarms?
It compares each day with the same weekday in earlier weeks, allows more room for small or jumpy numbers, and reads a number that looks wrong a second time before it alerts. A number needs about three weeks of history before it is judged at all.
Which GA4 settings can it change?
Key events, custom dimensions and metrics, data retention and enhanced measurement. Each one is a proposal until a person approves it on the Publish level. Archiving a custom dimension or metric cannot be undone through Google's API; the other changes are undone on a rollback.
Which plan includes GA4?
The GA4 package reads GA4 on its own, from €29 per client a month. The Tag Manager and platforms package reads GA4 beside the container, to check that what the tags send arrives.
Does it read the realtime report?
Yes. After a version with a new event goes live, the task watches GA4's realtime report for that event. Monitoring also uses it for a quick look every quarter of an hour in the week after a container change.