The Homogenisation of Thought, Not Text

Most conversations about AI homogenisation stop at the sentence. People notice that the copy reads the same across competitors and treat that as the whole disease. It is the symptom. The convergence that costs money happens one level up, in the strategy room, because that room now runs on the same three or four models as every other strategy room in your category. When three teams reason with the same machine, they do not just write alike. They choose alike.

The short version: the worry about prose sounding identical is real but small. The larger problem is that judgment converges when everyone reasons with the same models, so teams land on the same segments, the same positioning, the same plan. Each strategist feels sharper because the model lifts any individual. The field of strategies collapses toward one at the same time.

This is the sibling of why every brand sounds the same, moved from text to judgment. Same mechanism, higher stakes: a converged sentence is embarrassing, a converged decision is expensive.

The convergence moves from the page to the plan

A model returns the most probable continuation. Point Claude or its rivals at a blank page and you get the average sentence. Point the same model at a strategy question and you get the average answer, drawn from the same training distribution that every competitor's model draws from. Kirk and colleagues named this reduction mode collapse in preference-tuned models, and it does not stay inside the prose. It reaches the reasoning, because the reasoning is generated the same way the prose is.

So when your team asks Claude which segment to target, how to position against the category leader, what the wedge should be, the machine hands back the consensus move. It reads as insight. It is the middle of the middle, wearing your brief. Three teams asking the same well-formed question get three versions of one answer, because every model draws from one center.

Each strategist sharpens while the field flattens

Here is why the room never notices. The convergence does not arrive as a worse plan. It arrives as a better one, one desk at a time.

In a controlled study, Doshi and Hauser found that generative AI raised the quality of individual work while reducing the diversity of the work taken together: assisted output scored higher on its own and resembled other assisted output far more than unassisted work did. Every strategist gets a lift. The set of strategists homogenises. Both are true, and only the first is visible from your chair. Doshi and Hauser put it plainly: generative AI "enhances individual creativity but reduces the collective diversity of novel content."

Read that Doshi and Hauser finding as a claim about judgment, not prose, and it is the whole problem in one line. The lift you feel is the convergence you cannot see, because every desk feels the lift and no desk feels the merge. You experience a smarter plan; the category experiences one plan.

Your decisions converge before your sentences do

The order matters, and it runs against intuition. Copy is downstream, because every sentence is the last step of a decision made earlier. By the time two brands publish sentences that read alike, the deeper convergence already happened upstream, when both teams reasoned their way to the same segment and the same angle. The identical prose is not the cause; it is the receipt for an identical decision made weeks earlier.

That is why switching tools or prompting for a distinct tone changes nothing. You would be editing the receipt. The decision that produced it came from the same averaging machine, and rewording the output does not un-converge the choice underneath it. A distinctive plan has to be supplied from outside the model, by a person who knows something the training data does not.

The prose problem (small)	The judgment problem (large)
Your copy resembles a competitor's	Your segment choice matches a competitor's
Fixable by editing the sentence	Set weeks earlier, in the strategy room
Embarrassing when a reader notices	Expensive when the market cannot tell you apart
A model averages your words	A model averages your reasoning

Where distinctiveness has to come from

If the machine defaults to the average plan, distinctiveness is the one plan it cannot default into. It has to be engineered on purpose. At Ivanooo, Firoz Azees built the instrument to hold that distinctiveness in three places a model cannot average away:

1. Voice. A structure a machine can measure as distant from the category center and hold as recognisably yours. AI can average your voice; it cannot originate one. That distance is a property no competitor reaches by prompting.
2. Entity. An identity the machine resolves by name, so the distinctive judgment attaches to your brand and not to the category. Without it, even a sharp decision credits the field instead of you.
3. Topic Authority. Proof only you own: your data, your named methods, the decisions you made and can defend. Material no other brand's model could have produced, because it was never in anyone's training set.

The reason a model returns the consensus strategy is the same reason it gives generic answers to any question: the exception is the one thing it cannot generate for you. So the brands that stay distinguishable treat convergence as an operating condition. They watch the distance from the category average as a live number and defend it, because the pull to the middle never rests, and it now pulls on the decision, not only the draft.

The metrics most teams track will miss all of this. Share of voice and mention counts tell you whether the machine knows you exist. They say nothing about whether your strategy is still your own. See where your voice sits against your category's average: paste your URL, get the distance measured, with the evidence. No call required.

FAQ

What is AI homogenisation? It is the collapse of variety that happens when many people produce work with the same models. Most treatments cover only the prose: copy across a category starts to read alike. The deeper form is homogenisation of judgment, where teams reasoning with the same models converge on the same strategies and positioning.

Why is converged judgment worse than converged copy? Copy is downstream and editable. A converged decision was set weeks earlier in the strategy room and shapes everything after it. Two brands publishing similar sentences already made similar choices about segment and angle. The prose is the receipt for a decision that had already converged.

If AI makes my strategist better, how is that a problem? Because individual quality and collective diversity move in opposite directions. Each strategist produces a stronger plan while all the plans grow more alike. The improvement is real and visible from your desk. The convergence is real and invisible until the whole category lands on the same move.

Can switching models or prompting harder fix it? No. The averaging is a shared property of how these systems are trained, so changing tools moves you between near-identical centers. Prompting for distinctiveness targets the average of everything already tagged distinctive. The exception has to come from outside the model.

How do I know if my team's thinking has converged? Run the plan through the same test as the copy. If a competitor could have reasoned their way to your segment, your positioning, and your wedge using the same models, you have converged in judgment, and the identical sentences downstream are only the visible proof.