What AI Is Doing to How Teams Make Decisions
Teams now bring AI tools into their decisions, and most of them have never agreed on how far to trust what those tools say. An AI answer usually comes first and sounds confident. People can start agreeing with it before anyone has fully thought through the question. A team that brings AI into its decisions without ground rules can end up with less disagreement and weaker judgment. Leaders who set a few rules early can keep the speed AI offers. They can also keep the decision, and the responsibility for it, with people.
A Confident First Answer
When the most senior person at a meeting speaks first, everyone else learns which answer is safe. An AI tool can do the same thing to a group. Someone pastes a summary or a recommendation into the chat before the discussion begins. From then on, the group edits that answer and stops forming its own.
Editing a draft feels productive, so people fix a phrase and check a number while nobody questions how the tool defined the problem in the first place. A member who saw the problem differently now has to argue against a polished draft. That takes more nerve than offering an idea when nothing has been written yet. Many people will not bother, especially if they are new or junior.
Researchers who study pilots and others working with automated systems have a term for the pull toward trusting the machine: automation bias. People with a reliable tool tend to accept its output and check it less often. The more reliable the tool, the less closely they look. In flight simulator studies, crews missed problems when the automation failed to flag them, and they followed automated directions that were wrong. Training and clear procedures reduced both kinds of error.
Where AI Helps and Where It Misleads
A 2023 field experiment gives a useful picture of both sides. Researchers from Harvard Business School, Wharton, MIT, and other schools worked with 758 consultants at Boston Consulting Group. On tasks the AI handled well, consultants who used it finished more tasks and worked faster, and the quality of their results went up. On a task chosen because the AI was likely to get it wrong, consultants with AI were 19 percentage points less likely to reach the correct answer than consultants who worked without it.
The researchers called the line between those tasks a jagged frontier. AI is strong on some tasks and weak on others that look similar, and users often cannot tell which side of the line a task falls on. The wrong answers on the hard task were plausible and well written, which helps explain why capable people accepted them.
For a team, the finding means that the same tool can make one decision better and the next one worse, and the output will look equally confident both times. Trusting every AI answer will mislead a team on some of its hardest questions. The useful skill is telling which kind of question the team is facing before deciding how much weight the tool's answer gets.
When Team Members Trust It Differently
Few teams have settled this question, so each member usually decides on their own. One person treats an AI answer as a fact and brings it to the meeting as settled. If a colleague asks a different tool and gets a different answer, both may defend their answers with the same line: the AI said so. Another member distrusts anything a tool produced and dismisses it without reading it.
The split often follows each person's history with the tools. A member who uses AI every day has seen it be right many times and checks it less. A member who once watched it invent a source may reject even its correct answers. Each person is reacting to what they have seen. The team has never agreed on a rule, so two people at the same meeting may handle the same answer in opposite ways.
The discussion then turns into an argument about the tools. The question the team came to answer gets less attention. Nobody has checked the evidence yet, so the team ends up arguing about whose source to believe.
Ground Rules That Keep People in Charge
Positive Leadership's approach to communication, called Communication4Alignment (C4A), starts with a principle we call Advocate Reality. Reality should decide the question, and an AI answer is a claim about reality. Treating it as a claim leads to a practical rule. Before a team acts on anything an AI tool produced, someone on the team checks it against the source it depends on, whether that is the budget, the patient record, the contract, or the data itself.
A shared Purpose, Outcomes, and Methods (POM) points to another rule. The Purpose and the Outcomes are for the team to decide, because they depend on what the people involved care about and what they are committed to reaching. AI is most useful within the Methods, where it can draft tactic options or flag risks in a plan. When a team keeps AI in that role, the tool adds ideas and leaves the direction to people.
The discussion order needs its own rule. Each member forms a view of the problem before the AI output comes out, even if that view is two sentences written on a notepad. The tool's answer then becomes one more input, and the team can compare it with what people already thought. Members who notice a problem the tool missed write it down before the polished draft appears, and they are more likely to raise it.
Two smaller practices also help a team keep track of where its information came from. When someone brings AI output to a meeting, they can say so and say which tool produced it, so the team knows what it is looking at. Every decision also needs a named owner. A tool cannot be accountable for a result, and a person can.
These rules take little time to set, and they give members permission to question a confident answer without seeming to resist new tools. With them in place, a team can use AI heavily and still know who decided what and why.
The next time a decision is on a team's agenda, a leader can start by asking one question before opening any tool: what does each person already believe the answer is?
