Skip to main content

AI Usage > Interactions

Written by Tom Williams

Dataset: AI Usage

Entity: AI Usage Measurement

Field ID: interactions

Type: Number

Description: The number of human-initiated prompts. Not interchangeable with requests or sessions: on agentic surfaces one prompt can trigger dozens of requests. Most providers publish only some of the three, so an empty value is missing coverage rather than a zero.

Source: App

Transformation logic: N/A

App Mapping

GitHub Copilot

user_initiated_interaction_count on surface and CLIENT records; prompt_count on command line and desktop app records; empty on agent app records, where the count is reported as sessions, and empty on LANGUAGE_SURFACE and LANGUAGE_MODEL records, which publish no prompt count at all

Anthropic Claude (coming soon)

N/A

Cursor (coming soon)

the counter matching the record's surface - chatRequests, cmdkUsages, composerRequests, agentRequests or bugbotUsages

OpenAI Platform (coming soon)

N/A - API traffic has no concept of a prompt

OpenAI Codex (coming soon)

turns

Reporting Use Cases

The Interactions field is the closest thing to "how often did a person ask the AI for something", and it is the usual headline measure of AI engagement.

  • Engagement over time: SUM(interactions) grouped by YEAR_WEEK(created_at) is the standard AI adoption chart.

  • Per-developer intensity: Dividing interactions by distinct actors gives an average intensity that is comparable across teams of different sizes.

  • Never mix with requests: On agentic surfaces one interaction can produce dozens of requests. Adding the two together, or charting them on one axis without saying which is which, is misleading.

  • Empty is not zero: Anthropic and OpenAI Platform do not publish an interaction count. Filter to the providers that do before computing an average.

Did this answer your question?