Hi OpenAI Team,
I’m a daily GPT-4 user who relies on it as an advanced thinking and execution partner. My use cases span system design, creative writing, research, and physical recovery planning. First, I want to say thank you. This tool has transformed the way I work, think, and create.
But I’ve encountered a critical issue—consistency and reproducibility, or rather, the lack of it. And for high-trust workflows, this creates serious friction.
⸻
The Core Problem: GPT Gives Different Outputs When Asked to Repeat the Same Plan
In many workflows, I design or co-create structured plans with GPT—recovery routines, creative scene outlines, product roadmaps, etc. I often say “yes” to a generated plan, then ask GPT to repeat or summarize it.
However, GPT frequently gives a slightly altered version, even when I didn’t request any changes.
• Sometimes, wordings are changed “for clarity.”
• Sometimes, key elements are rearranged or omitted.
• Even when I explicitly say: “Repeat this verbatim”, the output still morphs.
As a result, I lose confidence in both the current and previous versions—which one is the most accurate? Should I now compare and merge two near-identical outputs manually?
This completely undermines the speed and trust GPT is supposed to provide.
⸻
Why This Matters Deeply to Power Users
When we use GPT not just to draft, but to design, store, recall, and execute, then even minor deviations create:
• Version anxiety (Which version is “official”?)
• Cognitive friction (Now I need to proofread AI output more than human output?)
• Loss of delegation value (I start doing the work myself again.)
I understand that GPT strives to be helpful, to improve, to optimize. But in this case, what I need is precise fidelity, not creativity.
⸻
Feature Request / Suggestion
Could GPT offer a “verbatim recall mode”?
Example prompt:
“Repeat Plan A exactly as you generated it at 3:41 PM on May 12, no edits.”
GPT should then:
• Recall and display the exact original wording, layout, and bullet points;
• Offer a clear diff if it does propose edits (“This differs from Version A in 2 minor places.”);
• Allow users to store and reference versions by name (“RecoveryCard2”, “SceneDraft_Visitation”, etc.).
This is not just a nerdy request. It’s the foundation of trust in long-term collaboration.
⸻
Final Note
I know GPT isn’t a database or a CMS. But for heavy users treating it as a cognitive partner, the ability to trust what was said, and retrieve it intact, is not a bonus—it’s essential.
Thank you for reading. If the product team is interested, I’m happy to provide more structured examples across writing, wellness, and project execution scenarios where this issue breaks flow.
— Submitted via GPT-4, based on a real user’s reflections and system design pain points