Practical guides ↗

OpenAI Signals · Data preparation

Read the data dictionary before analyzing

Use the OpenAI Signals dictionary to distinguish row units, measures and comparison groups.

4 min readScope: Data dictionary v2.0

Try it in three steps

  1. Open pages 1–3 of the source PDF. Read the scope and observation unit first: do not assume one row represents one person.
  2. Match names to definitions: month is a time label; share_of_messages is the weighted message share within the relevant group; rank orders per-capita usage.
  3. State the denominator before writing a conclusion. Do not interpret message shares as user shares. Record the data version and source.

How to read the output

Independent synthetic exercise — not an actual Signals result: If 50 of 200 messages in one month concern topic X, its message share is 25%. We do not know how many distinct people wrote about X, so cannot compute the user share. A rise to 30% in another month is 5 percentage points.

Test your understanding.

True or false?

In the synthetic example, 25% of messages about X means 25% of users wrote about X.

Related concepts

Official source

OpenAI Signals · Documentation ↗

Documentation checked: . Steps are based on official documentation and have not been tested in the application. Menu names may differ by version and interface language.