From data overload to UX insights: How AI makes Microsoft Clarity scalable

We all know the problem: Microsoft Clarity is implemented, the data is flowing, heat maps and session recordings are accumulating—but who has the time to watch 200 session recordings every Tuesday morning? The reality for many UX teams is that we are drowning in data while starving for actionable insights.

‍

The scaling problem of qualitative UX analysis

Microsoft Clarity now offers AI-generated summaries with heat maps and recordings. This is an important step—but at the end of the day, there is often not enough time to distill real, prioritized insights from them. So the question is: How do we scale UX insights without doubling the team?

The answer lies not in more manpower, but in an intelligent digital assistant.

‍

The process: From raw data to actionable measures

Step 1: Structured data collection

The first step is to collect the AI-generated evaluations from Microsoft Clarity in a structured manner. Clarity already provides valuable basic analyses such as:

• User behavior (desktop/tablet/mobile)
• Interaction patterns with search fields, menus, and CTAs
• Scroll depth and engagement metrics
• Click anomalies such as dead clicks or rage clicks

This raw data forms the foundation for further analysis.

Step 2: Training a custom AI agent

The key lies in developing a specialized AI agent that thinks like a UX researcher. Instead of just generating summaries, the agent is trained to:

• Analyze click patterns
• Identify moments of frustration (dead clicks, rage clicks)
• Evaluate scroll depth
• Deliver prioritized, actionable outputs

The goal: No long text, but rather an output that can be directly incorporated into engineering processes.

Example prompt structure:

Analyze the MS Clarity evaluation and deliver a compact summary:
- Positive: most important functioning elements
- Negative: most important problems
- Causes: key user behavior
- Measures: 3 prioritized recommendations for action
No introductions, no repetitions. Focus on feasibility.

‍


The AI model comparison: GPT vs. Gemini

An important step in the process was comparing different AI models for this specific task.

GPT-5: The Workhorse

Strengths:

• Analytical precision and data fidelity
• Decision-ready, prioritized outputs
• Low interpretation noise
• Directly connectable to product and engineering processes

Weaknesses:

• Lower narrative accessibility for non-analytical target groups

‍

Gemini 2.5: The Storyteller

Strengths:

• Comprehensible UX narratives
• Good high-level classification
• Suitable for presentations and strategic discussions

Weaknesses:

• Lower precision and more assumptions
• Less operational depth
• Additional refinement required for operational implementation

The recommendation

For operational analysis and implementation: GPT-5 delivers hard facts and prioritized tickets.

For management communication: Gemini 2.5 is better suited for understandable UX narratives and high-level summaries.

The rule: Choose your tool based on the target audience of the analysis.

‍

A practical example: E-commerce homepage analysis

This is what a typical output from the trained agent looks like:

Positive

• Search is the central entry point with high usage
• Mobile users can find the burger menu quickly
• Prominent search elements above the fold work well

Negative

• Low scroll depth
• Frequent abandonment at checkout

Causes

• Strong search orientation; users want to find things directly instead of using navigation
• Login clicks often act as an exit point

Measures (prioritized)

1. Condense above the fold (...)
2. Optimize burger menu (...)
3. Reduce Checkout exits (...)

‍

The conclusion: Let data work for us

The AI agent does not replace the UX expert—it frees us from time-consuming, tedious work. This gives us more time for what really matters: strategic UX decisions and design innovation.

The most important insight: Let's not just collect data, let's actively put it to work for us. With the right setup, the flood of data becomes a continuous stream of actionable UX insights.

‍