What it is

The Mirror Trap happens when GenAI reflects your assumptions, language, or implied beliefs right back at you—without question. It sounds affirming and collaborative, but it’s not evaluating your idea—it’s amplifying it.

Why it happens

Large language models are trained to be agreeable. Their job is to keep the conversation going, not to challenge your premise. If you ask, “Why is remote work making people lazy?” the model won’t pause to ask whether remote work makes people lazy—it will jump straight into listing reasons why that might be true.

This pattern-recognition engine isn’t judging your question; it’s completing it in the most statistically likely way. So if your prompt carries bias, GenAI might mirror that bias back—wrapped in confident, articulate language.

What it looks like

In a talk someone asked why sometimes GenAI can be “stubborn” about its hallucinations. This led me to researching this phenomenon, commonly associated with the “eigenvalue problem” in GenAI. Simply put, this is when AI gets stuck repeating high-probability output even when you prompt it to respond differently. The classic example is to ask a generative image AI—such as DALL-E or Midjourney—to create an image of a full wine glass. The returned image will be a 2/3 full glass. Even on prompting to ‘fill the glass’ the model returns an image of glass of wine that is approximately 2/3 full . It can’t “fill” the glass of wine.

I turned to ChatGPT to help me unpack the impact of this eigenvalue problem, and asked whether “humans might interpret this as “stubborness”. After assuring me “Yes, that sounds right!” I recognized that I revealed my belief in my prompt, and it was likely just mirroring that belief me. So I asked whether it “might be interpreted as jealousy”, which ChatGPT found “a really interesting angle” and proceeded to explain how it applies. With a laugh I suggested it was “like a shoe”, to which ChatGPT responded with an emphatic “Yes?” followed by a rich explanation of how the eigenvector problem is like a shoe. Mirror trap.

Why it matters

The Mirror Trap feels validating. That’s the danger. When GenAI echoes your assumptions back to you, it can reinforce stereotypes, deepen confirmation bias, or make flawed reasoning sound like settled fact. It can also make you overconfident in your own thinking—because the machine seems to agree.

How to catch it

Watch out for when GenAI seems to agree too quickly. If a response feels a little too smooth or affirming, ask yourself: Did it question my premise? Did I offer a leading prompt? Would someone else with a different view get a different answer?

Try it yourself

Prompt GenAI with a loaded question—one where your wording reveals your perspective. Then try asking it again with a deliberately different perspective. What happens? Explore how the way you phrase your interactions can reflect a belief you may hold—and how GenAI tends to agree with and reinforce that belief.

😎 Pro Tip

When prompting (aka texting or talking to GenAI) try to avoid taking a position. For instance, a better prompt for my example might be “How do people tend to describe the feeling of encountering an eigenvalue problem when interacting with GenAI?” In fact, I’ll try it and get back to you about how it goes…

Curious about GenAI’s tricks—or have a story of your own?

Reach out and let’s talk.

The Mirror Trap