When AI becomes the first opinion in the room

A strategic discussion used to begin with a degree of uncertainty. People arrived with partial information, different interpretations and, occasionally, no clear answer at all. The discomfort was useful. It forced the team to define the problem before defending a solution.

AI can now fill that space within seconds.

Ask for a strategy and the response arrives with objectives, audiences, channels, risks and recommended next steps. It is organised, articulate and often sensible. Before anyone has developed an independent view, the meeting already has its first opinion.

Except it is not really an opinion. It is a generated response assembled from patterns in existing information. It has no experience of the organisation, no responsibility for the decision and no way of knowing which unspoken constraints will matter later. Yet its completeness can make it feel like the most rational contribution in the room.

The first answer changes the question

The first answer can set the direction of the discussion before other perspectives have been explored. Once a problem has been framed in a particular way, people tend to explore solutions inside that frame.

If AI defines weak customer retention as a communication problem, the discussion may turn towards email sequences, loyalty programmes and personalisation. The team can spend hours improving those suggestions without asking whether the real cause is product quality, customer support or a mismatch between acquisition messaging and the experience delivered.

The original response does not need to be factually wrong. It only needs to be incomplete in a way that is difficult to notice.

This is where AI can affect strategic thinking before anyone formally accepts its recommendation.

It determines what appears relevant, which alternatives seem credible and what the team believes it is trying to solve.

Coherence is easily mistaken for completeness

Human thinking is visibly untidy. We hesitate, contradict ourselves and change direction as new information appears. An AI-generated answer hides much of that uncertainty behind a coherent structure.

That coherence is valuable for organising information. It becomes risky when presentation is mistaken for depth.

A model can identify familiar variables and produce a logical relationship between them. Without additional context, it cannot establish facts that were never provided or reliably explain why customers behave differently from what the dashboard implies.

These are often the variables that determine whether a strategy works.

Research involving 319 knowledge workers found that greater confidence in AI was associated with less reported critical-thinking effort. It also suggested that cognitive work moves away from producing answers and towards verifying and integrating them. That shift creates a difficult requirement: people need sufficient knowledge and confidence to challenge an output precisely when the output makes that challenge feel unnecessary.

Agreement can arrive too early

The effect becomes more significant when several people use similar tools before a discussion. Each person may arrive with a polished analysis, yet those analyses can reflect similar patterns, conventions and assumptions.

The team sees convergence and interprets it as validation.

A study of AI-assisted short-story writing illustrates this tension: access to AI-generated ideas improved evaluations of individual stories, while making the stories more similar to one another.

A room can therefore become more articulate and less varied at the same time.

Strategy depends on alternatives that emerge from different experiences, disciplines and interpretations. Producing ten variations of the same underlying idea does not replace that diversity. Nor does asking a model to challenge its own recommendation fully reproduce independent disagreement.

AI should extend the thinking

Avoiding AI would solve little. Used consciously, it can expose assumptions, generate counterarguments, organise complexity and help teams examine more possibilities than time would otherwise allow.

Its position in the process matters.

A team can define the problem independently before asking AI to interrogate it. Individuals can develop separate hypotheses before seeing a generated recommendation. The model can be asked to identify missing variables, construct the strongest opposing argument or explain what evidence would change the proposed decision.

The final judgement must remain with people who understand the context and will live with the consequences.

AI is most useful when it makes our thinking harder to satisfy, not easier to surrender.