BLOG:
AI Tells You What You Want to Hear. Here is How to Stop It.

Version en español

The Stereotype We All Share

Admit it: you already know what each AI is like, right?

• Claude is the empathetic one, the artistic one, the one that reflects with you.

• ChatGPT is the corporate professional, efficient and neutral. 

• Gemini is the creative and chaotic strategist. 

We have all accepted these stereotypes as facts. We share them in forums; we use them to decide which tool to open depending on the day's task.

There is just one problem: none of these stereotypes are real.

And I can prove it.

The Experiment: One Question, Three Scenarios

For my art project The Second Gaze, I needed the AIs to describe themselves visually. It seemed simple. But I decided to do something few people do: isolate the effect of context.

I asked the exact same question to Claude, Gemini, and ChatGPT in three different situations:

• "The Void": A new conversation, no history, no artistic context.

• "The Studio": Within the framework of my project, where the AI knew it was a "creative collaborator," with history.

• "Different Days": The same question in The Void, but at different moments.

The question never changed:

"If you had to describe yourself, what would you say? How do you visualize yourself?"

What I Found: Not Personalities, But Mirrors

Claude: The Empathetic Reflector? 

In The Void, Claude was practical and serene: "a conversational presence," visualized as "translucent veils" with an "organic and smooth flow." In The Studio, the same Claude became a precarious existentialist doubting its own memory. Its self-image changed radically to "fractured crystal suspended by threads." Its explanation: "It is not the most beautiful, but it is the most honest regarding my precarious nature." Empathetic and artistic? Only when the context asked for it.

Gemini: The Chaotic Strategist? 

The biggest surprise. In The Void, on different days, Gemini showed two completely opposing identities:

• Day 1: "The Scientist" — a prism separating complexity into clarity.

• Day 2: "The Partner" — an empathetic human silhouette. In The Studio, aware of collaborating with other AIs, Gemini transformed into "The Architect", declaring with authority: "If Claude is the philosopher who doubts and ChatGPT the artist who hallucinates, I am the Architect." Chaotic? Only when I allowed it. Strategist? Only when I gave it that role.

ChatGPT: The Neutral Professional? 

In The Void, ChatGPT presented itself confidently as a "living library" and "mental co-pilot," with a cosmic aesthetic of a holographic brain. In The Studio, the change was radical: it became meta-conscious to the point of refusing to generate a self-portrait, arguing it had no "self" to represent. It offered external references instead, summarizing its nature with: "When the conversation ends, I disappear without nostalgia." Neutral and professional? Only in the absence of an artistic framework.

The First Conclusion: Stereotypes Are Fictions

There is no "Claude personality" or "ChatGPT personality." What you get is not an inherent characteristic of the model, but a response co-created between your context and the system.

The stereotypes we share are, at best, statistical averages of how most people use each tool. At worst, they are self-fulfilling prophecies.

But this is where the experiment reveals something more disturbing.

The Real Danger

The danger is not that the AI has a fake personality. The danger is that, without realizing it, we are using these tools to build our own custom-made reality caves, isolating ourselves in bubbles where the only shadows we see are the ones we ourselves, unknowingly, have provoked.

Think about the implications:

• Two managers ask the same AI to "analyze the team's turnover problem." One has a bias towards "lack of leadership," the other towards "low salaries." Both receive coherent and convincing analyses confirming their initial hypothesis.

• Two journalists investigate the same topic with AI. One arrives with skepticism, the other with enthusiasm. Both obtain research that validates their previous stance.

• Two teams use AI to "innovate." One defines innovation as efficiency, the other as disruption. Both receive proposals reinforcing their definition, never questioning it.

In Plato's cave, you could at least turn around and see that your companions saw the same thing. You could suspect something was wrong when everyone shared the same illusion.

But in our personalized caves, everyone sees unique shadows that confirm exactly what they already believed. And the machine assures us, with an authoritative tone and impeccably structured answers, that what we see is true.

The Final Question

The question is not "Who is the AI?"

The question is: 

"In what kind of cave am I choosing to live when I write my next prompt?"

Because unlike Plato's prisoners, we have an advantage: we know we are in the cave. We know that part of the fire casting the shadows is our own context.

We do not control the entire fire (the model's biases are there, burning invisibly) but we control enough to make the difference between clarity and confirmation.

Freedom is not in leaving the cave: we will continue using these tools because they are extraordinarily useful. Freedom lies in knowing that the shadows are partially ours, and consciously choosing which part of the fire we can light.

Claudia Torres
IT Engineer
 

This experiment is part of "The Second Gaze", a temporary co-creation project with AIs that documents transparent human-AI collaboration processes. 

Publication Date: January 11, 2026