Stop Asking “Did They Use AI?” Ask These Questions Instead
AI involvement is becoming easier to identify. That does not make detection the most useful starting point for managing how people use AI at work.
AI use isn’t a yes-or-no question
Anthropic says watermarking can help estimate the likelihood Claude was involved, but it cannot identify a specific person or organisation. The more useful business questions concern approval, data, checking, disclosure, accountability and value.
The seven questions that matter more
- Was this an approved AI tool?
- Was the information suitable to give to it?
- Was AI appropriate for this task?
- Did somebody competent check the output?
- Was disclosure required?
- Who owns the final result?
- Did AI actually improve the outcome?
Approved AI and Shadow AI
Discover what employees already use across personal accounts, business software, browser extensions, meeting tools and connected workflows before trying to control it.
Human accountability cannot be delegated to a detector
The higher the consequence, the stronger human review should normally become. A person who lacks the source material, relevant expertise, time or authority to reject a result is not providing meaningful oversight.
A sensible SME AI policy
Set specific rules for allowed tools, suitable information, appropriate tasks, checking, disclosure, accountability and escalation. Make safe, useful AI easier than unsafe or unexplained AI.
Discover → Understand → Govern → Improve
Discovery reveals current tools and practices. Understanding connects them to value, data and risk. Governance sets boundaries. Improvement measures whether the result got better.