AI Transparency Statement

Verified facts are immutable. AI output is a separate, versioned, attributed interpretation of those facts. The two are never blended, and interpretation is always labelled as interpretation wherever it appears.

The model receives only company facts already stored in our database, having been imported from Companies House. It does not browse the web, it is not given data from other customers, and it is not given any source we cannot cite.

Structured output only: an industry classification drawn from a controlled taxonomy, a short plain-language summary, a confidence level, supporting evidence drawn from the input facts, and confidence flags where the record is thin or ambiguous. "I don't know" is a valid and expected output.

It must not invent facts, infer information that is not supported by the record, provide commercial, legal or financial advice, score opportunities, or qualify leads. It is not used to make automated decisions about any individual, and it has no role in credit, insurance, employment or eligibility decisions.

Every interpretation records the input snapshot and its hash, the prompt version and its hash, the model and version used, the timestamp, the confidence level and the supporting evidence. Generation is deterministic, so the same facts and the same prompt produce the same interpretation. Interpretations are versioned rather than overwritten, and previous versions remain visible.

If the underlying facts change at Companies House, the existing interpretation is marked outdated and regenerated against the new facts. If nothing has changed, nothing is regenerated. Interpretations carry a clear status so you always know whether what you are reading is current.

Interpretation is intended to support human judgement, never to replace it. Every output can be traced back to the facts that produced it, and any interpretation can be regenerated or challenged. Where interpretation and the register disagree, the register is correct.

Interpretation is generated through a third party AI provider called from our servers. Provider credentials are never exposed to the browser. The AI layer is deliberately abstracted so that a provider or model can be replaced without changing the guarantees described on this page.