Compare AI privacy claims on three separate axes: training use, storage and data movement or sharing. A restriction on training does not establish that nothing is stored or transmitted. Match each statement to its product context and mark unsupported cells unknown.
Keep the comparison boundaries visible
Record the workspace type, setting and workflow alongside each claim. Personal chat, editor assistance, repository agents and file uploads are different contexts. Do not collapse them into a single “private” label.
The following record-style matrix uses only the supplied evidence. “Unknown” means the evidence supplied here does not establish the answer, not that a provider necessarily lacks a policy.
Read the three-axis evidence matrix
ChatGPT personal Free, Plus and Pro workspaces
- Training: sharing is enabled by default. Turning off “Improve the model for everyone” excludes new conversations from training.
- Storage: those conversations can remain in chat history; a retention duration is unknown here.
- Movement/sharing: broader transmission and recipient details are unknown from this evidence.
Cursor
- Training: with Privacy Mode enabled, Cursor says code is not used for training by Cursor or other AI model providers.
- Storage: Cloud Agents temporarily store encrypted repository copies while running, then delete them after completion. Other storage details are unknown here.
- Movement/sharing: AI features send prompts and code context to model providers; Cloud Agents require repository access.
Perplexity file uploads
- Training: a specific training restriction is not established by the supplied file-privacy excerpt; mark unknown.
- Storage: stated retention is 30 days, or seven days for Enterprise Pro.
- Movement/sharing: Perplexity says uploads customise responses and are not accessed or shared for other purposes unless a session is shared. Do not expand this wording into unprovided recipient details.
Apply a hypothetical policy test
Suppose a fictional Indian development team requires “no repository content sent outside approved infrastructure.” Cursor’s training restriction alone cannot satisfy that requirement: the supplied evidence describes prompts and code context going to model providers.
Suppose a fictional editorial team instead requires “documented retention no longer than ten days.” Perplexity’s stated seven-day Enterprise Pro retention meets that numeric criterion, while 30 days does not. This resolves only that criterion, not permission, suitability or every privacy requirement.
Stop where evidence stops
Write each requirement as a question, attach evidence and mark pass, fail or unknown. Stop procurement when a mandatory condition remains unknown; request clarification through an approved process. Use synthetic inputs during evaluation.
Check the AI tools catalog for access format and saved terms before buying. Marketplace account, code or token offers are not proof of official subscription equivalence; provider documentation does not establish marketplace access conditions.
Sources and scope
Source material checked 29 October 2026; features/policies may change. ChatGPT controls, Cursor privacy and governance and Perplexity file privacy establish the bounded statements above. The comparison method and hypothetical policy tests are editorial advice, not legal conclusions.