Diagnose
Who owns what your AI believes? We find out in two weeks.
Most companies shipping AI assistants can't answer a simple question: who is responsible for verifying that what the AI believes about your users and your policies is still true? That gap — corpus accountability — is what this engagement is designed to close. Before a tribunal closes it for you.
We catalogue every source feeding your AI: help docs, personas, research repos, policy documents, support tickets, AI-generated summaries. We date-stamp each: verified this quarter, this year, or unknown.
For companies using retrieval-augmented AI (chatbots, internal assistants, search), we check whether the knowledge your retrieval layer draws from matches current reality. This is the specific gap behind the Air Canada case: the policy in the corpus didn't match the policy that existed, and no one had checked.
We rank stale beliefs by decision exposure. Where is outdated user knowledge most likely to produce a bad product decision, a legal liability, or a confidently wrong answer to a real customer?
We flag AI-feeding-AI loops: meeting notes summarized by a bot fed back into the knowledge base, AI-generated personas treated as research input, synthetic outputs used as ground truth. These loops quietly degrade the corpus over time.
A Flash Findings report: executive summary, prioritized fix list, and a recommended research cadence design for keeping the corpus verified going forward.
Questions first? ian@siblingsystems.limited