Paying ChatGPT Plus user. Model keeps breaking explicit instructions + feedback channels blocked

Concerns with ChatGPT Plus: Repeated Instruction Failures and Lack of Feedback Channels

As a longstanding paying subscriber to ChatGPT Plus, I find myself increasingly dissatisfied with the service’s reliability and transparency. While occasional errors are understandable in complex AI systems, recurring behavioral failures combined with limited avenues for providing feedback have led me to reconsider my subscription. Below, I outline the specific issues I’ve encountered, which are not isolated incidents but part of a consistent pattern over extended interactions.

Persistent Breakdown of Explicit Instructions

Across multiple sessions and tasks, the AI model frequently fails to adhere to direct commands. Notable problems include:

• Use of Forbidden Phrases: Despite explicit prohibitions, the model continues to use phrases like “if you want,” which I have clearly instructed to avoid.
• Unwanted Follow-up Questions: After being told not to ask follow-ups, the AI often does so anyway, disrupting the workflow.
• Formatting Inconsistencies: The model randomly changes message formatting, affecting readability and professionalism—especially problematic when handling documents, tables, or embedded content.
• Rule Violations: It promises to follow specific instructions but immediately breaches those commitments.
• Contextual Failures: The model struggles to maintain consistent behavior within the same conversation thread, reverting to generic responses or templates unexpectedly.
• Formatting and Layout Issues: Difficulties in managing document formatting details, including tables, Word files, colors, and layout elements, diminish the tool’s utility for professional use.
• Self-Contradictions: The AI sometimes contradicts previous statements or reverts to default templates, undermining trust in its reliability.

This pattern persists across numerous interactions and diverse workflows, indicating a systemic issue rather than isolated glitches.

Feedback System Limitations

More concerning is the apparent restriction on providing user feedback. Despite being part of a “feedback experiment” purportedly aimed at improving the model, my account seems locked into a mode where:

• Binary Feedback Only: Thumbs-up or thumbs-down buttons are the sole means of providing feedback.
• No Textual Comments: Clicking thumbs-down yields no text box or way to elaborate on issues.
• Inaccessible Support Channels: Links to support pages or direct URLs lead to non-existent pages, and the Help Center contains only FAQ content with no option to contact support directly.
• No “Report a Problem” Option: The typical pathways to report specific issues appear to be absent in the current interface.

This severely hampers my