Introducing the Property Knowledge Base - Teach the AI About Your Business

Every analytics setup has context that lives in people’s heads. The March traffic spike that’s actually an annual report, not a growth trend. The organic traffic from India that’s 80% bots. The custom channel groupings that don’t match GA4 defaults.

This context is the difference between analysis that’s technically correct and analysis that’s actually useful. An AI that doesn’t know about your March report will confidently tell you that traffic grew 300% - which is true, and also completely misleading.

Today we’re launching the Property Knowledge Base - a way to store this business context directly alongside your analytics properties, where the AI can reference it during every conversation.

---

The Problem

AI models are good at reading data. They know what sessions, bounce rates, and conversion rates are. They can run queries, compare time periods, and spot patterns.

What they don’t know is the story behind your data. Every business has it:

• “Our primary conversion goal is demo_request, not the default purchase event.” Without this, the AI analyses the wrong metric and every recommendation is off-base.
• “Traffic from India is mostly bots - exclude it from organic analysis.” Without this, the AI reports inflated traffic numbers and draws conclusions from noise.
• “We use custom channel groupings: Partner Traffic means affiliate referrals, not partnerships.” Without this, the AI applies default GA4 channel definitions that don’t match how your team thinks about attribution.

You could explain this in every conversation. You could put some of it in a custom system prompt. But system prompts are limited in length and apply at the team level, not per property. And re-explaining context every time defeats the purpose of having an AI assistant.

---

How It Works

The knowledge base is a set of contextual entries attached to each property. Each entry has a title, content written in markdown, optional tags, and an AI-generated summary.

You write entries the same way you’d brief a new analyst joining your team: here’s what you need to know about this property’s data before you start analysing it.

When you chat with the AI, it automatically knows what entries exist. Their titles are included in the system prompt, and the AI searches for relevant entries when your question relates to a topic you’ve documented. You don’t need to say “check the knowledge base” - it happens in the background.

---

Semantic Search

If your team has an OpenAI provider configured, entries are automatically embedded as vectors - numerical representations of their meaning. When the AI searches for relevant context, it matches by meaning rather than exact keywords.

This means a question about “unusual traffic patterns” can surface an entry titled “March Traffic Spike - Annual Industry Report” even though the words don’t overlap. A question about “data quality” can find an entry about bot traffic from India.

If you don’t have an OpenAI provider, search falls back to keyword matching, which still works well for most cases. Both modes can coexist - entries with embeddings use semantic search, and entries without use keywords. The cost of embeddings is negligible (fractions of a cent per entry) and they’re generated automatically when you publish.

---

What Should Go In the Knowledge Base

The most valuable entries are things the AI can’t learn from the data itself:

Data quality issues. Bot traffic sources, misconfigured events, tracking gaps, known discrepancies between platforms.

Business definitions. What your conversion goals actually mean. How your custom channel groupings work. What metrics your team cares about and how you define them internally.

Seasonal patterns. Annual events, holiday effects, budget cycles - anything that causes predictable metric changes that aren’t obvious from the data alone.

Competitor context. Who you compete with, how their activity affects your metrics (e.g., aggressive PPC bidding driving up your branded CPCs), and how you differentiate.

Technical context. Recent migrations, platform changes, tracking limitations, how your setup differs from standard implementations.

---

Compared to Other Features

The knowledge base fills a specific gap between existing features:

Feature	What it does	Time dimension
System prompts	Shape how the AI behaves and reasons	Permanent, team-wide
Annotations	Mark specific dates with events	Tied to dates
Knowledge base	Store business context the AI references	Permanent, per-property

Annotations answer “what happened on this date?” - a campaign launched, a deployment went out, an algorithm update hit. The knowledge base answers “what does this mean?” - how your business defines conversions, why certain traffic patterns are normal, what to watch out for in the data.

System prompts control the AI’s personality and approach. The knowledge base gives it facts about your business.

---

Getting Started

Navigate to any property and click Knowledge Base in the sidebar. Start with the entries that would save you the most time re-explaining:

1. Your conversion definitions. What events actually matter and which ones to ignore.
2. Known data issues. Bot traffic, misconfigured tracking, or platform discrepancies.
3. Seasonal context. Predictable patterns that would otherwise look like anomalies.

Each entry has an AI Summary button that generates a concise summary from your content, making it easier for the AI to quickly assess relevance.

Write entries like you’re briefing a colleague, not documenting a system. The AI reads natural language, so “Our March spike is from the annual report we publish - traffic goes 3-5x for about two weeks” works better than a formal specification.

Full documentation is available in our Knowledge Base guide.