Organisations have decades of practice identifying and countering groupthink. Devil's advocates. Red teams. Pre-mortems. Structured dissent protocols. The premise behind all of them is the same: when a group's thinking converges too quickly, challenge it before the decision hardens.

These mechanisms work by detecting the social signals of premature convergence — a dominant voice going unchallenged, a room that stops producing genuine objections, a process that moves from divergence to conclusion too fast. They are entirely blind to what is now the most pervasive source of convergence in most organisations.

What invisible groupthink is

When teams simultaneously consult the same AI systems — querying the same foundation models, using the same enterprise AI tools, summarising the same documents through the same algorithms — they produce convergent thinking without any of the social friction that conventional groupthink detection relies on. Nobody is being dominant. No one is being silenced. The process feels open and analytical. And the outputs are systematically more similar than the team's independent thinking would have been.

A May 2026 paper from SSRN named this mechanism explicitly: Invisible Groupthink. AI-mediated conformity that operates without directive leadership and without social pressure. The safeguards designed to catch groupthink look for a person causing the convergence. Invisible groupthink has no such person.

When teams simultaneously consult homogeneous AI systems, they unknowingly undergo a process of AI-mediated conformity that is structurally invisible to the organisational safeguards designed to prevent it.

Why AI homogenises thinking

Large language models are trained on overlapping corpora. When team members query a model with a strategic question, the model's response is shaped by the distribution of positions in its training data — not by the full range of perspectives relevant to this organisation in this context. Ask the same question to the same model through different interfaces, different phrasings, different team members — and the outputs will converge on what was statistically central in the training distribution. The problem is that 'most plausible given training data' and 'most useful for your organisation's specific strategic challenge' are not the same thing.

The epistemic diversity that disappears

Epistemic diversity — the range of ways of knowing that a group brings to a question — is the source of the unexpected connections, the productive anomalies, and the stress-tested conclusions that genuine deliberation produces. When teams reach for AI at the point where they would previously have needed to consult each other, they are replacing the most epistemically diverse part of their process with the part that is most homogeneous.

How collective sensemaking operates differently

The protection against invisible groupthink is not avoiding AI. It is ensuring that the part of the process where epistemic diversity is built happens before and alongside AI analysis, not after it. A SparkMap builds a living structure in which every contribution carries its epistemic ground — expert fact, lived experience, belief, values, gut feel, speculative inference — and where the connections between perspectives with different grounds are as visible as the perspectives themselves.

This is not a rejection of AI's analytical power. It is the layer that ensures AI operates on genuine collective intelligence rather than its simulation.