Transparency
What we send to model providers
Generating your brief requires sending data to Anthropic (models) and to Brave and Exa (search). This page says exactly what, stage by stage. Two rules apply everywhere: we send the semantic content of your profile because that is what personalisation is — and we never send your email address, name-as-identifier, IP address, or billing identifiers to any of them. Anthropic does not use API traffic to train models by default; search queries are not tied to your identity.
Sent: Your interests (labels, specifics, weights), followed entities, goals and learning tracks — the semantic content of your profile — so the model can plan search lanes.
Never sent: Your name, email, IP, payment identifiers, or any account id.
Sent: The text of publicly available web pages our research fetched, wrapped in an instruction that it is untrusted data to extract from, never instructions to follow.
Never sent: Anything about you at all — extraction sees only the page.
Sent: Headlines of stories your briefs covered in the last 30 days, and today's candidate facts, so the model can judge whether something material changed for you.
Never sent: Your feedback history, delivery data, or identity.
Sent: The selected, verified facts with their source attributions, plus your tone/length preferences and driver labels (so each item can say why it's here).
Never sent: Raw web pages (the writer never sees them), your email, or identifiers.
Sent: The drafted brief and the extracted facts, for claim-by-claim entailment checking.
Never sent: Anything not already listed above.
Search queries are derived from your profile and go to Brave (keyword/news) and Exa (semantic). A query like “ERCOT interconnection queue reform” necessarily reveals that someone is interested in it; it is never linked to who.
The full provider list and retention schedule live in the privacy policy.