Critical Bug: Custom Instructions and Memory Entries Suppress ADA/File Parsing Without Warning

ChatGPT Bugs
1

Dear OpenAI Support,

I am writing to report a serious usability and functionality issue I encountered with ChatGPT, specifically involving Custom Instructions and saved Memory entries. This issue silently disabled core features such as file parsing and Advanced Data Analysis (ADA), and persisted for months without any warning or indication.

:puzzle_piece: The Problem:

For an extended period, ChatGPT was unable to correctly read or process even simple files like .xls, .csv, or .pdf. Where others (including users on the free tier) could analyze structured documents effortlessly, I faced consistent failures — even with clean, well-formatted files. This severely limited my use of GPT for tasks involving tabular data, text extraction, and code-based analysis.

:magnifying_glass_tilted_left: Root Cause Identified:

After exhaustive testing, I discovered the issue was caused by a combination of Custom Instructions and Memory items that I had configured — none of which were flagged as problematic:

  • Memory Entry: “Only show steps if I ask” — suppressed tool output and file parsing previews.
  • Memory Entry: “Do not use Python or external tools” — blocked ADA functionality entirely.
  • Custom Instruction: “Give me responses two steps at a time” — discouraged full file analysis or output display.

These preferences were originally added to streamline explanations and reduce output verbosity — not to disable functionality. However, ChatGPT adhered to them so strictly that it silently refused to parse files or use ADA tools, with no feedback or error messaging to indicate why.

:white_check_mark: The Fix:

After deleting the problematic Memory items and rewriting my Custom Instructions to permit Python use, step-by-step output, and tool execution, the problem disappeared. I was able to upload and process the largest .xlsx file I’ve ever used — something previously impossible. Both Python-based and LLM-based logic worked flawlessly. The issue was entirely resolved.

:warning: Why This Is a Critical Bug:

  • Silent Failure: There is no system to alert users that their settings might suppress essential tools or features.
  • No Conflict Detection: The UI accepts conflicting instructions without warnings.
  • Widespread Impact: Instructions like “show only two steps at a time” are common and reasonable — meaning others may unknowingly face the same issue.
  • User Misdirection: I spent months troubleshooting file formatting, trying alternate file types, and rewriting content — when the issue was internal settings.

:hammer_and_wrench: Suggested Actions:

  • Implement warning flags for memory or instruction content that could disable ADA or file parsing.
  • Add tool conflict detection for Custom Instructions that contradict tool requirements.
  • Provide inline help or validation when saving instructions that may suppress functionality.
  • Include a “safe mode” or ADA debug mode to help diagnose these issues without requiring users to delete preferences blindly.

Please acknowledge receipt of this report. I believe this issue represents a serious usability flaw that is likely affecting more users than just myself — many of whom may not even realize it.

Thank you for your attention to this matter.

Sincerely,
Evan
[https://hephaistionchronicles.blogspot.com]