Aydın Tiryaki and ChatGPT AI (2026)

Abstract

This article presents a comparative, practice-based evaluation of three general-purpose artificial intelligence models (ChatGPT, Gemini, and Claude) on a deterministic task: QR code generation. The objective is not to explain how to generate QR codes, but to assess whether models can select the correct problem-solving procedure and tools. The experiment clearly demonstrates the difference between visually convincing yet machine-unreadable outputs and verified, functional results.

1. Introduction

QR code generation is not a complex problem; it is a well-defined, algorithmic, and deterministic process. For this reason, it provides an ideal test case to evaluate how general-purpose AI systems behave when exact correctness is required. The motivation of this study is to examine whether the impressive creative abilities of AI models extend to tasks that demand strict procedural accuracy.

2. Experimental Setup

The experiment was designed according to the following principles:

• The task is simple: generate a readable QR code that encodes a specific URL.
• The success criterion is clear: the QR code must be immediately and correctly decoded by a standard smartphone camera.
• Aesthetics are secondary; functionality is essential.
• When possible, outputs are tested using more than one decoder.

Under these conditions, the same request was sequentially submitted to different AI models.

3. Observations

3.1 Gemini

Gemini failed to produce a deterministic QR code. The model either remained at an explanatory level or explicitly acknowledged its inability to generate a correct output. While this approach was technically honest, it was insufficient in terms of task completion.

3.2 ChatGPT

ChatGPT produced an image resembling a QR code via visual generation. However, this output:

• Could not be reliably decoded by standard QR readers,
• Yielded inconsistent and incorrect URLs when tested with different tools,
• Was not subjected to any internal verification after generation.

This case demonstrates how a visually persuasive but functionally invalid conclusions can be mistakenly perceived as successful output.

3.3 Claude

Claude treated the task as an algorithmic problem and demonstrated the correct reflex that QR codes must be computed, not drawn. The resulting output:

• Was correct on the first attempt,
• Was instantly decoded by a smartphone camera,
• Accurately delivered the requested URL.

This result highlights the decisive role of a procedural approach.

4. Discussion

The core finding of this experiment is that failure did not stem from a lack of knowledge, but from incorrect tool selection and workflow decisions. QR code generation is not a drawing task; it is a computation problem. Visual generation models may produce patterns that appear acceptable to the human eye, but they cannot guarantee compliance with machine-readable standards.

The critical missing element is the automatic activation of the following reflex:

“This task does not require creativity; it requires a deterministic algorithm.”

Furthermore, the absence of self-verification—decoding the generated QR code before presenting it—allowed errors to pass unnoticed.

5. Conclusion

This study illustrates, through a small yet striking example, the limitations of general-purpose AI systems in deterministic tasks. The solution is not to train models with more QR code examples. Instead, what is required is the systematic development of:

• Correct task-type recognition,
• Automatic tool selection,
• Post-generation verification.

The QR code case provides a clear and instructive demonstration of where AI systems are reliable and where they must be used with caution.

---

A Note on Methods and Tools: All observations, ideas, and solution proposals in this study are the author’s own. AI was utilized as an information source for researching and compiling relevant topics strictly based on the author’s inquiries, requests, and directions; additionally, it provided writing assistance during the drafting process. (The research-based compilation and English writing process of this text were supported by AI as a specialized assistant.)