Using the AI Chat
Learn how to ask effective questions and get the most out of AI-powered analytics conversations.
The AI chat is where you ask questions about your analytics data in natural language. This guide helps you get the most out of your conversations.
Starting a Conversation
- Click New Chat from your dashboard or property page
- Select which data sources to include (GA4, Search Console, etc.)
- Choose your AI model
- Type your question and press Send
Selecting Data Sources
At the top of the chat, you’ll see colored pills showing your connected data sources. Click any pill to toggle it on or off for that conversation.
Click a source to toggle it. Disabled sources appear faded with a strikethrough.
Why toggle sources?
- Focused responses: If you’re only asking about SEO, disable GA4 so the AI doesn’t pull irrelevant traffic data
- Faster responses: Fewer data sources means less context for the AI to process
- Lower costs: Less context = fewer tokens = lower API costs
- Cleaner analysis: Avoid confusing the AI with unrelated data
Example combinations:
| Sources | Best For |
|---|---|
| GA4 only | Traffic, behavior, conversions |
| Search Console only | SEO, keywords, rankings |
| GA4 + Search Console | Combined traffic and SEO analysis |
| Google Ads only | Campaign performance, ad spend |
| CRM only | Sales pipeline, deal analysis |
| All sources | Comprehensive cross-platform insights |
Your choices are saved per conversation. If you disable Pipedrive for a chat about SEO, it stays disabled when you return to that conversation.
Setting Default Sources
Don’t want a connection included in chat by default? You can change this in your connection settings:
- Go to your property’s Connections page
- Find the connection you want to configure
- Click the chat icon (purple) to toggle the default, or click Edit and use the “Show in AI chat” switch
Connections disabled this way won’t be included when you start new conversations, but you can still enable them manually in any chat.
Choosing Your AI Model
Claude models work best with AI Data Stream as they integrate directly with our data tools. Gemini models are also excellent, especially for cost-conscious usage.
| Model | Speed | Quality | Cost |
|---|---|---|---|
| Claude Sonnet 4.5 | Medium | Excellent | Medium |
| Claude Opus 4.5 | Slower | Best | Higher |
| Claude Haiku 4.5 | Fast | Good | Low |
| Gemini 3 Flash | Very fast | Great | Very low |
| Gemini 3 Pro | Medium | Excellent | Medium |
| DeepSeek Chat | Medium | Good | Very low |
| GPT-4o | Medium | Good | Medium |
Recommendation: Start with Claude Sonnet for the best results. Gemini Flash is great for quick checks when speed matters. DeepSeek offers similar quality to GPT at a fraction of the cost.
Writing Effective Prompts
Be Specific About Time Periods
Vague:
How’s my traffic?
Specific:
How did my traffic perform last week compared to the week before?
Even better:
Compare my sessions and users from January 1-7 to January 8-14, 2024
Mention the Metrics You Care About
Vague:
How’s my SEO?
Specific:
What are my top 10 keywords by clicks in the last 30 days, and how has their average position changed?
Ask About Comparisons
The AI excels at comparisons:
- “Compare mobile vs desktop performance”
- “How does this month compare to last month?”
- “Which traffic sources have grown the most?”
Request Specific Formats
If you want data presented a certain way:
- “Show this as a table”
- “List the top 5 with percentages”
- “Summarize in 3 bullet points”
Request Charts and Visualizations
The AI can generate interactive charts directly in the chat. Just ask for a visual representation:
- “Show this as a chart”
- “Create a bar chart of traffic by source”
- “Visualize this as a line graph”
- “Show a pie chart of device breakdown”
Supported chart types:
| Chart Type | Best For | Example Request |
|---|---|---|
| Line | Trends over time | ”Show sessions over the last 30 days as a line chart” |
| Bar | Comparing categories | ”Create a bar chart of top 10 pages by views” |
| Pie | Showing proportions | ”Show device breakdown as a pie chart” |
| Doughnut | Proportions with emphasis | ”Visualize traffic sources as a doughnut chart” |
| Scatter | Correlations | ”Plot bounce rate vs. session duration as a scatter chart” |
| Polar Area | Comparing values | ”Show channel performance as a polar chart” |
Example: Traffic by Source
Here’s an example of what the AI might generate when you ask “Show me a bar chart of my traffic sources”:
Example: Sessions Over Time
Ask “Show sessions over the last 30 days as a line chart” and you’ll get something like this:
Charts are generated in real-time and support dark mode automatically. Hover over data points to see exact values.
Pro tip: Be specific about what data you want charted. Instead of “show a chart,” try “create a line chart showing weekly sessions for the last 3 months.”
Example Questions by Use Case
Traffic Analysis
What were my top 10 landing pages last month by sessions?
Where is my traffic coming from? Break down by source/medium.
Which pages have the highest bounce rate?
How does weekday traffic compare to weekend traffic?
Show me traffic trends for the last 90 days.
SEO Performance
What are my top 20 keywords by clicks in Search Console?
Which pages have high impressions but low CTR?
What keywords am I ranking on page 2 that could move to page 1?
How has my average position changed for branded keywords?
Which pages lost the most clicks compared to last month?
Campaign Analysis
How did my email campaign traffic perform last week?
Compare Google Ads traffic to organic traffic for conversions.
What's my cost per conversion across all campaigns?
Which ad campaigns have the best ROI?
Break down Google Ads performance by campaign.
Content Performance
Which blog posts get the most organic traffic?
What's the average time on page for my top 10 articles?
Which content categories have the highest engagement?
What pages have improved the most in search rankings?
Ecommerce
What was my revenue this month vs last month?
Which products have the highest conversion rate?
What's my cart abandonment rate?
Break down revenue by traffic source.
What's the average order value by device type?
Using Prompt Presets
Prompt presets configure the AI’s analysis style before you ask your question. They’re like giving the AI a specific role.
Built-in Presets
Click the preset selector before typing your question:
| Preset | Focus |
|---|---|
| Data Analyst | Objective, fact-based analysis |
| SEO Specialist | Organic search and rankings |
| Marketing Manager | Campaigns and ROI |
| Content Strategist | Content performance |
| Executive Summary | High-level KPIs |
| Ecommerce Analyst | Sales and revenue |
When to Use Presets
- Regular analysis: Use Data Analyst for objective insights
- Reporting to stakeholders: Use Executive Summary for concise highlights
- SEO audits: Use SEO Specialist for search-focused analysis
- Content reviews: Use Content Strategist for engagement insights
See Prompt Presets for detailed descriptions.
Conversation Features
Context Memory
The AI remembers your conversation. You can ask follow-up questions:
You: What were my top 5 landing pages last month?
AI: [Lists pages]
You: Now show me the traffic sources for just the first one.
AI: [Shows sources for that specific page]
Annotations in Context
When you ask time-based questions, the AI checks your annotations:
You: Why did traffic drop on March 15?
AI: Looking at your annotations, I see you noted “Server migration - 4 hours downtime” on March 15. This likely caused the traffic drop during that period…
This is why keeping annotations up to date is valuable.
Multi-Source Queries
When you have multiple sources connected, you can ask combined questions:
Show me pages with high search impressions but low GA4 sessions.
Compare Search Console clicks to GA4 organic sessions.
Which keywords from Search Console are driving conversions in GA4?
Message Options Menu
Each message in a conversation has a menu with helpful options. Hover over any message and click the ⋯ button to access these features:
Excluding Messages from Context
The AI remembers all messages in a conversation when generating responses. Sometimes you might want to exclude certain messages from this context - perhaps an early question led to an irrelevant tangent, or you want the AI to “forget” a specific exchange.
To exclude a message:
- Hover over the message you want to exclude
- Click the ⋯ menu button
- Select Exclude from context
Excluded messages appear with a dashed border and an “Excluded” badge:
To re-include a message: Click the menu and select Include in context.
When to use this:
- Remove off-topic exchanges that might confuse the AI
- Focus the conversation on specific topics
- Reduce token usage by excluding lengthy responses you no longer need
- Start fresh without losing the full conversation history
Forking Conversations
Sometimes you want to explore a different direction from a specific point in your conversation without losing your current thread. Fork conversation creates a new conversation that branches from any assistant response.
To fork a conversation:
- Find the assistant message where you want to branch
- Click the ⋯ menu button
- Select Fork conversation from here
A new conversation is created containing all messages up to and including that point. You can then take the analysis in a completely different direction while keeping your original conversation intact.
When to use this:
- Explore “what if” scenarios without losing your main analysis
- Branch into a detailed sub-topic while keeping the overview conversation
- Test different approaches to the same question
- Share a conversation with a teammate while continuing your own analysis
Tips for Better Results
Start Broad, Then Narrow
- Ask a broad question first: “How did last month perform?”
- Follow up on interesting findings: “Why did conversions drop on the 15th?”
- Dig deeper: “What happened to mobile conversions specifically?”
Use Date Ranges Strategically
- Last 7/30/90 days: Good for trend analysis
- Month-over-month: Good for reporting
- Year-over-year: Good for seasonal context
- Specific dates: Good for event analysis
Ask “Why” Questions
The AI can help identify causes:
- “Why did bounce rate increase last week?”
- “What might explain the traffic drop on Tuesday?”
- “Why is mobile conversion rate lower than desktop?”
Request Actionable Insights
End questions with what you want to do:
- “…and what should I do about it?”
- “…what are the quick wins here?”
- “…which should I prioritize?”
Managing Conversations
Viewing History
All conversations are saved. Access them from:
- Your property’s Conversations tab
- The Recent Chats section on your dashboard
Exporting
To save a conversation:
- Open the conversation
- Click the Export button
- Choose format (Markdown, PDF, or copy to clipboard)
Deleting Conversations
- Go to Conversations
- Find the conversation to delete
- Click the delete icon
- Confirm deletion
Deleted conversations cannot be recovered.
Troubleshooting
”No data found for this period”
- Check that your connections are active
- Verify the date range has data
- Try a broader date range
AI gives generic responses
- Be more specific in your question
- Include the metrics and dimensions you want
- Use a more capable model (Claude Sonnet, GPT-4o)
Responses are slow
- Use a faster model (Gemini Flash, Claude Haiku)
- Ask for less data (shorter date ranges)
- Break complex questions into parts
AI seems confused
- Start a new conversation
- Rephrase your question more clearly
- Check that the right data sources are selected