The model says X, so let's go with that.
The first answer sets the frame
An AI's answer, a senior person's view, or the first idea in the room sets the starting point. Everything that follows gets judged relative to that starting point instead of being developed independently.
- Few real alternatives get considered.
- The AI's recommendation is accepted almost every time.
- Discussions open with "Do we all agree?" rather than "What are our options?"
Who: two or three people who were in the last decision on this workflow. Ask: "Before the discussion, had you written down your own view? Had you already seen the AI's answer or heard what the most senior person thought?" Count: options seriously considered on the last three decisions. You're probably looking at this pattern if people saw an answer before forming a view and only one or two options were weighed. Probably not if people formed views first and still converged.
Everyone's thinking looks the same. In that pattern, everyone brings the same view because they used the same AI beforehand. In this pattern, one answer takes over during the meeting.
When one option dominates, say: give the other two equal time before we decide. One redirection does the work.
Give it one instruction. Make the strongest case this is wrong.
In an experiment with physicians, those who made their own diagnosis before seeing the AI's advice were more accurate and better at telling right advice from wrong advice than those who saw it from the start ("Designing AI-Based Work Processes: How the Timing of AI Advice Affects Diagnostic Decision Making," Management Science, 2025). Emerging. The same study is the evidence behind the Spark move Scout Before You Share.