Does GPT-5 thinking output in a weird format for you?

Analyzing Recent AI Response Formatting: Is It Just You or a Broader Trend?

In the evolving landscape of AI-assisted writing, many users notice peculiarities in how current models generate output, especially when certain features are enabled. Recently, some users have observed that responses from models like GPT-5 or Claude can adopt unexpectedly unusual formats when “thinking” or internal processing modes are activated. This raises questions about the nature of these outputs, user experiences, and how best to interpret them within creative or professional workflows.

Understanding the Context

When engaging AI for complex tasks such as roleplay development, detailed prompts often require the AI to adopt specific styles or frameworks. For example, a user seeking to improve their roleplaying writing might provide a comprehensive prompt that includes genre, tone, style, and guidelines. The AI’s response, ideally designed to be actionable and clear, sometimes morphs into a format that resembles a structured lesson or a set of instructions, complete with bullet points, code-like segments, or detailed templates.

A User’s Experience: Pattern or Anomaly?

Recently, one user shared a sample prompt where they asked for help in enhancing their role-playing skills. Instead of a straightforward reply, the AI produced a detailed “playbook,” complete with core principles, setup instructions, templates, and troubleshooting tips—all formatted as a structured, step-by-step guide. What stood out was the conversational tone, which felt as if the model directly addressed a peer or even a prominent figure in the AI industry, such as Sam Altman.

This conversational anomaly raised concerns: Is this formatting intentional, or is it an unintended artifact of the model’s internal reasoning process? Is this pattern consistent across different AI platforms, or unique to certain implementations with enabled “thinking” or reasoning modes?

Implications for Users

For users leveraging AI tools for creative writing, organizational workflows, or educational purposes, such formatting can be both a boon and a source of confusion. On one hand, structured responses can aid comprehension and application; on the other hand, unexpected stylistic shifts or overly formalized outputs might complicate seamless interaction.

Best Practices and Tips

To navigate these responses effectively, users might consider:

• Explicit Prompting: Clearly specify the desired output format and tone, and explicitly request concise or straightforward answers.
• Version Control: Test responses with features enabled versus disabled to identify differences.
• Feedback Loop: Use follow-up prompts to clarify or reframe responses that seem overly complex or stylistically unusual.
• Understanding Model Limitations: