Is your organisation's approach to AI the one that works, or the one that backfires? Five questions. Two minutes. Grounded in a field-experiment finding that rigid procedural protocols for AI use measurably reduce productivity, while cognitive reframing of AI as a 'thought partner' raises top-end quality.
A recent Microsoft Research field experiment (Farach, Cambon, Tankelevitch et al., April 2026) gave 388 Fortune-500 retailer employees the same AI tool and randomly assigned them one of three conditions: free-form use, behavioural scaffolding (a structured protocol telling them when and how to use AI), or cognitive scaffolding (short training that reframed AI as a thought partner rather than a tool to operate).
The headline finding: the behavioural protocol reduced document quality and substantially reduced document production. The cognitive reframing raised top-end quality. Full paper: arxiv.org/abs/2604.08678.
This self-check maps your organisation's current pattern onto the paper's two arms. Score < 0 means you're leaning toward the shape that hurt the study; score > 0 means you're leaning toward the shape that helped. It's a framing diagnostic, not a prediction — but it tells you which direction to push.
Five questions. About two minutes. We score each answer against the behavioural-vs-cognitive axis tested in the April 2026 Microsoft Research field experiment. At the end you get a score between −10 (your pattern matches the intervention that measurably hurt productivity) and +10 (your pattern matches the intervention that helped), plus a plain-English explanation of what to do next. Answers are not stored anywhere; the calculation happens entirely in your browser.
We're asking about the rule of engagement, not the tool.
Think about the dominant mode, not every training session you ran.
The honest organisational pattern, not the stated policy.
What actually gets reported to leadership?
Pick the closest mental model, not the politest one.
Honest caveats: one study, one firm, one industry (retail), short-run outcome. The study's effect sizes are modest at the average and concentrated at the top of the distribution. We cite it as useful direction-of-travel evidence, not as proof. See Transition Era for the broader evidence base we're reading against.
Privacy: answers are held in your browser only — we don't log, store or submit your selections. Refreshing the page loses them. Take it as many times as you like.
Where to go next: Full piece on the study · Why UCM is the right home for AI skills · The AI Transition Era · 20-year workforce projections
