AI Transparency

Profluence operates a small set of cooperating engines — signal, forecast, narrative, risk, opportunity, and reputation fusion. Each one is intentionally narrow so that its outputs can be inspected, versioned and validated independently.

Outputs are structured: a forecast carries a horizon and a confidence range; an alert carries the signals that contributed to it; a recommendation links back to the underlying observations.

• It does not predict virality.
• It does not profile individual private users.
• It does not make automated decisions about employment, credit, housing, or benefits.
• It does not generate content impersonating real people without explicit, authorized consent.

We use public platform data accessed under each platform's terms, first-party data customers explicitly connect, and aggregated reputation signals. We do not buy data scraped against platform terms.

Customer-connected data is processed under contract and used only for the workspace that connected it. It is never used to train shared models.

Any change that affects how an audience segment or reputation score is computed requires human review before release. Forecasts and recommendations remain advisory — final decisions are made by our customers.

Each model release runs an internal evaluation suite that checks for distributional drift across audience segments. Material regressions block release. We document known limitations rather than hide them.

Some engines use third-party foundation models routed through our AI gateway. Provider, model family and version are recorded with every inference for auditability. We do not allow providers to train on customer data.

Models are versioned alongside code. Every production inference is tagged with the model version that produced it, so we can reproduce, audit and roll back if needed.

If you believe a Profluence output is wrong, harmful, or misused, contact us. We treat AI concerns as priority support and route them to the responsible team within one business day.