We do not train on customer data

Not "we anonymise it first", not "unless you opt out". Your content is used to answer your questions and for nothing else.

What the answer engine on this site actually is

It is worth being precise, because "AI" is doing a lot of work in most vendor copy. Ask BvLogic performs retrieval, not generation. It finds and quotes passages we have published. There is no language model composing prose in the request path, which is why it cannot hallucinate a figure — and also why it cannot answer a question that needs two documents combined. Both halves of that are true and we publish both.

Where models are used

Where a deployment does use a language model, we name which one, state what is sent to it, and confirm the provider's training commitments in the data processing agreement. A model provider that trains on prompts is a sub-processor decision, not a technical detail.

Governance

• Answers cite their sources, so any claim can be checked against the passage it came from
• The engine declines when we have published nothing on a subject, and we measure how often it fails to
• Model choice is documented per deployment and changeable without rebuilding the product