genai_conversation_turns

You can use this function to analyze conversation complexity, monitor user engagement, identify outlier conversations, or track conversation metrics for billing and usage analysis.

Usage#

Syntax#

genai_conversation_turns(messages)

Parameters#

Name	Type	Required	Description
messages	dynamic	Yes	An array of message objects from a GenAI conversation. Each message typically contains a `role` and `content` field.

Returns#

Returns a long integer representing the number of conversation turns. A turn is typically counted as a user-assistant exchange pair.

Example#

Count the number of conversation turns in a GenAI chat operation.

Query

['otel-demo-genai']
| extend turns = genai_conversation_turns(['attributes.gen_ai.input.messages'])
| summarize avg_turns = avg(turns), max_turns = max(turns)

Run in Playground

Output

avg_turns	max_turns
4.2	12

This query calculates the average and maximum number of conversation turns, helping you understand conversation complexity and engagement patterns.

genai_message_roles: Extracts all message roles to understand conversation structure. Use this when you need to analyze the role distribution in conversations.
array_length: Returns the total number of messages (not turns). Use this when you need the raw message count instead of turn count.
genai_cost: Calculates the cost of a conversation. Use this in combination with turn count to understand cost per turn.
genai_estimate_tokens: Estimates token usage. Use this with turn count to analyze tokens per turn.

Other query languages#

Splunk SPL users

In Splunk SPL, you would typically use eval with mvcount to count array elements, but there’s no built-in function specifically for counting conversation turns.

Splunk example

| eval turn_count=mvcount(messages)/2

APL equivalent

['ai-logs']
| extend turn_count = genai_conversation_turns(messages)

ANSI SQL users

In ANSI SQL, you would need to unnest the array and count rows, then divide by the number of roles, which is more complex.

SQL example

SELECT
  conversation_id,
  COUNT(*) / 2 as turn_count
FROM conversations
CROSS JOIN UNNEST(messages)
GROUP BY conversation_id