Google Analytics' built-in AI vs an agent: insights you're shown against questions you ask
Google Analytics will tell you that a metric moved unusually and generate narrative around it. That's useful and free. But it decides what's worth surfacing, it explains only in terms of Google Analytics dimensions, and it stops at the property boundary — which is exactly where the questions that change decisions begin.
Two different modes
| Google Analytics built-in insights | An agent reading Google Analytics | |
|---|---|---|
| Who picks the question? | Google Analytics decides what to surface | You ask what you want |
| Scope of the answer | Google Analytics data and dimensions | Google Analytics plus ads, Search Console, warehouse, store |
| When it arrives | On Google Analytics' schedule, in the UI | On yours, wherever you want it delivered |
| Follow-up questions | Not a conversation | Ask again, narrower |
| Output | A card in the interface | A report, a dashboard, an email |
The property boundary
Google Analytics anomaly detection is real and worth leaving on. What it can't do is attribute a change to something outside Google Analytics. A traffic drop that coincides with a paused campaign, a conversion dip that follows a Search Console ranking slide, a channel that looks strong in Google Analytics and unprofitable once ad cost is included — each needs a second dataset Google Analytics doesn't hold.
The dimension trap that produces wrong answers
Google Analytics' own model is a common source of confidently wrong analysis. "First user" dimensions are user-scoped and attributed to a first touch; the unprefixed versions are session-scoped. Both are correct, they answer different questions, and mixing them in one report produces numbers that refuse to reconcile. A tool reasoning over Google Analytics should know that rather than faithfully returning whichever you happened to name — see Google Analytics' CM360 dimensions explained.
What to do with each
- Keep Google Analytics' insights on. Free anomaly detection with no configuration is worth having.
- Ask your own questions elsewhere, especially the ones spanning Google Analytics and something else.
- Put the recurring analysis on a schedule so it arrives rather than waiting to be opened.
- Render it where stakeholders will actually look — a link beats a login.
Frequently asked questions
- Does Google Analytics have built-in AI?
- Yes — it surfaces anomalies and generates narrative insights about your property automatically. It decides what's worth surfacing, explains in Google Analytics' own dimensions, and stops at the property boundary.
- Why isn't Google Analytics' anomaly detection enough?
- It can flag that a metric moved unusually but can't attribute the change to anything outside Google Analytics — a paused campaign, a ranking slide, or ad cost that makes a strong-looking channel unprofitable all live in other systems.
- What's the most common Google Analytics analysis mistake?
- Mixing user-scoped "First user" dimensions with session-scoped unprefixed ones in a single report. Both are correct and they answer different questions, so combining them produces numbers that never reconcile.
- Should I turn off Google Analytics insights if I use an agent?
- No. Free automatic anomaly detection is worth keeping. The agent covers the questions you choose to ask, especially those needing data Google Analytics doesn't hold.