AI transparency
How BrandWater AI uses AI, what it can and cannot see, and where people stay in charge.
Last updated: 30 September 2026
Current version. This page describes how BrandWater AI works today. It is updated as the service launches, and the date above shows the latest version.
What BrandWater AI observes
BrandWater AI records the answers that AI platforms give to prompts you define, and the visible sources they show. It measures what any person could see.
What it does not access
- Private model vector databases.
- Hidden reasoning or internal states of any model.
- Private training data.
- Internal retrieval systems of AI platforms.
We never claim otherwise.
Measured versus interpreted
Counts and rates come from stored observations with defined denominators. AI generated explanations, summaries and labels such as sentiment are interpretations and are shown as such, with a link to the evidence behind them.
Evidence and confidence
Facts in Brand Intelligence carry a source, a confidence level and a review status of confirmed, discovered or conflicting. You review them before they are used.
People approve actions
In the current plan the first release only recommends actions and you make any change yourself. When execution is introduced, every action will require human approval. BrandWater AI does not change your website on its own and does not imply that work was done when it was not.
Known limits
AI answers vary between runs, so results are samples, and we show sample sizes. Results describe association, not proof of cause. Platform behaviour changes without notice.