Data sent to OpenAI
When you use Sense Chat, Heap sends the following data to OpenAI:
- End-user questions in natural language (for example: "How many users who view a product page add something to the cart?")
- The names of labeled events, segments, properties, and charts, including titles and axis names
- The names of built-in and custom properties such as "domain" or "path," and metadata about events, properties, segments, and charts — including how frequently they appear in other charts and dashboards. This helps the AI select the most relevant events and properties for a given question.
- The top five most common property values. This helps the AI map questions such as "How many signups did we have in France?" to the correct country property value.
Data not sent to OpenAI
Heap does not send the following to OpenAI:
- User identifiers or property values that resemble IP addresses, email addresses, phone numbers, or tax identifiers. Heap uses a regex to identify and exclude these value types.
- Query strings from URLs
- Any values sharing a name with values you are redacting via Heap's data redaction APIs
Data retention
Heap has activated OpenAI's zero data retention policy. Under this policy, request and response bodies are not persisted to any logging mechanism within OpenAI's systems. They exist in memory only for the duration of the chat session, solely to serve the request.
AI model training
Your data is not used to train OpenAI's large language models. OpenAI has committed to not using data submitted via its API to develop or improve its models. For more information, see OpenAI's Enterprise Privacy Policy.
Input and output data from Sense Chat does not train any third-party model. Heap may use this data to improve its own internal systems — for example, its prompting strategy — but such improvements are not customer-specific and do not affect any external model.