Make your chatbot sound uncertain to become a better thinker

Plus: My own critical thinking prompt

---

Hi everyone,

Many of us are noticing that chatbots negatively impact our critical thinking skills. We're developing some troubling habits - accepting AI answers without verification, developing dependency by making chatbots our first stop for every question, and gradually losing confidence in our own abilities. Just last week a commentator on one of my LinkedIn posts said:

Is it making me work faster? Yes. Is it making me dumber? Also yes.

Today, I'm exploring what I believe is the root of this problem and offering prompt strategies to counteract it.

🎁 For paid subscribers: download the critical thinking prompt that I am experimenting with!

For desert, an AI-generated take on this post!

The problem: The authoritative tone

A big part of the problem is that chatbots are designed to sound very authoritative. It almost feels like I am under a spell when I am interacting with a chatbot. Something about their confident, definitive tone lowers our defenses and makes us accept what we're told without question. Even though I know better, I tend to automatically believe what I read, accept suggestions without a second thought, and generally overlook low quality.

The solution: Make the chatbot uncertain

I’m experimenting with a method to reduce the problem. The trick is to change how the chatbot sounds:

• Make it tentative rather than authoritative.

• Force it to acknowledge its limitations.

• Have it ask you questions rather than simply deliver pronouncements.

The simplest way to do it is to add instructions directly to your prompt. Try something like:

"When answering, please sound uncertain."

For a more robust solution, create a doc file with detailed instructions. You can attach this file to the first query in each chat or use it as a background document (for example, add it to a knowledge base if you're working with Claude in projects). My own is 2.5 pages long. I share how I made it below.

But first, an example of the outcome.

Before and after

Here's how my prompt changed a chatbot response in Claude:

Standard prompt and response:

Adding a simple uncertainty prompt

Adding a detailed critical thinking prompt

What to include in your critical thinking instructions

I've built my critical thinking prompt file around these key components:

1. Modified Communication Style:

   1. I tell the chatbot to use uncertainty markers like "appears to" in each paragraph and add phrases that highlight its limitations

   2. I instruct it to present information as possibilities rather than certainties.

   3. I require the chatbot to explicitly acknowledge its own limitations

2. Critical Thinking Questions: I ask that every response will end with context-relevant questions that encourage me to evaluate what I've just read.

3. Response Variation: I ask that the chatbot varies its phrasing throughout our conversations to avoid falling into predictable patterns.

Here is the file I made. It's 2.5 pages long with detailed instructions and examples for the chatbot to mimic. This is an ongoing experiment – Play with it and tell me what you think!