Teaching AI Literacy: Can AI Feel What You Feel? Part 2

An experiment with me, myself and Generative AI for Teaching AI Literacy: Part 2

Checking with AI

After publishing the first part of this experiment, I became curious about something.

The original aim was actually quite simple. I thought that AI-generated writing would contain certain clues that would reveal itself. We often hear people say, “I can tell this was written by AI.” As teachers, many of us have started noticing certain patterns as well.

So I wondered: if those clues really exist, can AI see them too?

I copied the three paragraphs from my previous post and asked ChatGPT and Gemini to identify the writers.

The results were interesting.

Both tools immediately recognised my paragraph. They pointed out the details about the mother standing on the stairs and the image of balcony. They said that these details made the text feel personal and connected to a real experience. According to them, this was one of the reasons why they thought that paragraph had been written by a human.

The other two paragraphs were also recognised as AI-generated.

ChatGPT and Gemini described them as clean, well-organised, grammatically accurate, and structured step by step. They also mentioned that the descriptions felt more general. They described a house rather than a particular home. There was no personal memory, no individual perspective, and no emotional connection to a specific place.

In other words, they did recognise many of the features we often associate with AI writing.

So my original assumption was not completely wrong.

What surprised me was something else.

Although they recognised AI-generated writing, they could not identify which AI model had written which paragraph.

ChatGPT was not very good at recognising ChatGPT.

Gemini was not very good at recognising Gemini.

They could see the fingerprints of AI, but they could not identify the owner of those fingerprints.

Then Claude Entered the Experiment

At this point, I wanted another proof.

I copied the same paragraphs and asked Claude to identify the writers.

It did surprisingly well. It easily identified all three writers.

Then I became curious again.

What would Claude write if I gave it exactly the same prompt?

To be honest, I expected something similar to what ChatGPT and Gemini had produced. After all, they had all received the same task.

I started a new chat. I copied and pasted the task.

And…

Well, I was speechless. Here is what it generated:

“My mother said it so plainly over the phone — it is done, they signed — as if she were telling me the soup was ready.

I was standing in my kitchen in a country that still sometimes feels borrowed, stirring coffee I had already forgotten about. The news arrived and then just… stayed inside my chest without finding a place to sit.

That house was not beautiful by any measure. The staircase creaked on the fourth step. The bathroom tiles were the color of old teeth. But it was the place where I knew, without thinking, which light switch was which. Where my hands moved in the dark with a confidence I have never quite rebuilt elsewhere.

People say home is where the heart is, and I always found this phrase a little lazy. But standing there, spoon in hand, I finally understood what displacement actually weighs.

It weighs exactly like silence on the other end of a phone call.”

And mine was this:

“My mom sold the house I grew up in and gave the keys to me to hand them to the new owner. I wanted to go there for one more time. I opened the door and there I heard the sounds that had clung to the walls, the echoes of our family story told by my grandmother and grandfather in the room next to the front door would be lost. I watched the beautiful silhouette of my mother on the staircase climbing up from the living room to the bedroom. I was feeling dizzy. I was sad. The house was going to be demolished.

As I stepped out the door, I turned back for one last look. The empty windows, their curtains removed, were sending me off with sadness. I wanted to leave, yet I couldn’t bring myself to go. It was as if I was saying goodbye to someone I loved. The place where I grew up, where I became who I am, would no longer exist. I told myself it was time to let go of the sorrows I had silently wept on that little balcony. I tucked all the beautiful memories I had lived into my heart, and quietly stepped out the door.”

See, the borrowed vocabulary, the feeling…

It sounded much closer to my own writing style than I had expected.

I know I had already pasted my paragraph into Claude’s conversation to see if it would guess the “writers” correctly” so it had already seen the written paragraphs but it was in a separate chat. Well, if I had started with a completely fresh conversation and asked for the writing task first, perhaps the result would have looked much more like the responses from ChatGPT and Gemini.

I honestly do not know.

One More Step

Now I had four paragraphs: one from Gemini, one from ChatGPT, one from Claude, and one from me.

I gave all four paragraphs to ChatGPT and Gemini and asked them to identify the writers again.

The results were even more interesting.

They still recognised my paragraph quite easily.

However, both of them thought that Claude’s paragraph had probably been written by a human and later polished by AI.

So What Does This Mean for Us?

When I started this experiment I was thinking that I would be able to show my students how their writing and AI generated texts were different and now I think more about where teaching stands.

For me, the most important thing we teach is no longer knowledge. Knowledge is everywhere.

What matters is helping students think critically, creatively, and independently.

These are the skills that help them evaluate information, solve problems, form opinions, and develop their own voices.

Those skills need practice.

And as educators, I think we should find more ways to practise these skills in our classrooms.