When all AI end up recommending the same thing

For a long time, we imagined that each artificial-intelligence engine developed its own vision of the world.

After all, ChatGPT, Gemini and Claude are built by competing companies, trained with different approaches and based on distinct architectures. It would therefore seem logical for their recommendations to diverge on a regular basis.

Yet some measurements tell a different story.

On a question about the leading repair balms or multi-purpose creams for sensitive skin, the three providers studied converge on exactly the same primary recommendation. The product in question appears in 100% of each engine's answers, both when the AI answers on its own and when it relies on a web search.

This observation does not prove that this product is the best. Nor does it prove that the models use the same data, or that they copy one another. What it does highlight is a measurable phenomenon: competing AI engines can produce a near-perfect consensus on the same question.

An unexpected convergence

The differences between the major AI models are well documented.

• Their training methods differ.
• Their architectures evolve independently.
• Their product strategies are competing.

We might therefore expect several dominant recommendations to appear depending on which provider is queried. Yet that is not what the measurements taken here show.

The same player is cited by Anthropic, Google and OpenAI across all the answers analyzed. Competitors do appear, but only occasionally. None of them shows up systematically in the answers.

In other words, the engines keep enriching their answers with other references, but they appear to share a common anchor point.

Consensus does not mean uniformity

It would be tempting to conclude that all AIs answer in exactly the same way. The report does not allow us to claim that.

In reality, the engines do not always emphasize the same arguments. Depending on the provider, certain criteria are used more heavily: composition, performance, convenience, price or versatility occupy a different place in the reasoning presented to the user.

The narrative varies. The primary recommendation, however, remains stable. This distinction matters.

Two AIs can build different lines of reasoning while arriving at the same conclusion.

With or without web search, the result stays the same

One of the most interesting observations concerns the use of web search. You might think that access to the web profoundly changes the recommendations. In this particular case, that is not what is observed.

The presence of the primary product remains identical with or without web search: 100% in both cases.

This does not mean web search is useless. The report actually shows that several categories of arguments shift when this feature is enabled. The primary recommendation, on the other hand, does not change.

When a market seems to organize around a consensus

This measurement raises a broader question. Are there markets on which AI engines end up sharing a common view?

The report does not allow us to answer that. It documents only one case where this convergence is observed. But this observation opens up a strategic line of thought.

If several independent engines regularly converge on the same references, competition between brands could gradually shift. It would no longer be only about being visible to consumers. It would also mean understanding how consensus emerges in AI answers.

For now, this idea remains a hypothesis. It would require analyzing many other markets before it could be confirmed.

What this measurement really shows

This study measures neither the actual quality of a product nor its market share. It does not seek to determine whether the AIs' recommendations are accurate. It observes only what the engines display at a given moment.

In this particular case, the measurements show three facts:

• three competing providers converge on the same primary recommendation;
• this convergence persists with or without web search;
• the other players remain present but never become systematic.

A question that goes far beyond this market

Ultimately, the subject may not be this product category. The real question lies elsewhere.

At what point do several AI engines stop having different recommendations and end up building a shared consensus?

Understanding this transition could become a major challenge for every company seeking to be recommended by artificial intelligence.

Because if the models sometimes end up telling the same story, then the battle may no longer be only about appearing in AI answers. It may be about understanding how a consensus is born within their discourse.