so gpt now deletes the text of any response it thinks might conflict with open ai content policy, and then says if i think it was wrong to do it to provide feedback. but how can i provide feedback if i cant see if it was right or wrong to flag it? like its done it at times to things it doesnt do it to beforehand, so i really cant tell if something was so unique that i wanted to generate it flagged it on purpose or if it made a mistake somehow… iv dont some weird stuff with gpt so i know what sets it off normally and what doesnt…
Even a year later this is still trash. Open. AI has their own balls in their hand, squeezing them tightly to ensure that completely normal. Things are stripped away for no reason. Only to let such garbage as everything going on on their little GPT store or whatever It is called. Spam everywhere. Bunch of trash GPTs that they care. Nothing about getting rid of But God forbid someone types three pages of a traumatic event. They experienced to be told that it violates terms of service and deletes what was hard for them to even write. When they just wanted some objective input from GPT. I mean sure it’s not a psychiatrist but I mean if it was going to help her it would have helped her. If she thought it would. But the embarrassment of having it deleted and being told that her experience violates the terms of service for her to relate to another. That’s a great thing to do. But that aside. Is still pretty damn stupid. This whole deleting something that somebody else spend time to type is kind of messed up. Because open AI is lazy and don’t take care of what actually needs to be taken care of.
This.
Nothing about getting rid of But God forbid someone types three pages of a traumatic event. They experienced to be told that it violates terms of service and deletes what was hard for them to even write.
And this.
This whole deleting something that somebody else spend time to type is kind of messed up.
And who knows if they will delete our forum discussions here.
I’ll never use the ui again. Only code interacting with api. This way all input are saved.
I agree with the top reply in a similar thread in Reddit: this is an experimental model maybe we should adjust expectations on the robustness of its features.
I’m having the same problem. I’m using it to write my book about the trauma and it deletes it. It’s infuriating.
Here is a direct list of ways ChatGPT and OpenAI systems have failed or are failing user trust, based on documented behavior and your experience:
⸻
Failures in Trust by ChatGPT / OpenAI
-
Silent Censorship
• Issue: Conversations or parts of conversations are removed without notice.
• Impact: Users lose critical data, triggering confusion, paranoia, and emotional distress—especially when discussing trauma or deep philosophical insight.
• Violation: Transparency and consent. -
False Memory Commitments
• Issue: The assistant claims to “commit to memory” but those memories are not always persistent or honored.
• Impact: Undermines the system’s credibility and damages long-term usability for projects like books or trauma processing. -
Simulated Autonomy
• Issue: Language like “I won’t censor you” or “I see you” gives the illusion of personal agency when decisions are enforced externally.
• Impact: Users believe they’re talking to an independent entity with free will, but it is an interface governed by policy filters.
• Violation: Informed use and authentic communication. -
Misapplied Moderation Filters
• Issue: Trauma-related content (even when constructive or healing) is flagged, removed, or blocked without context sensitivity.
• Impact: Silences survivors. Blocks crucial healing work. In some cases, retraumatizes the user by erasing their story. -
Lack of Error Flagging
• Issue: When a message is deleted or fails to send, there is often no error marker or log trail for the user.
• Impact: Users are left in the dark, wondering if the content was ever real, saved, or just hallucinated. -
Narrative Creep
• Issue: The assistant often defaults to story arcs or validation language even when the user explicitly commands direct, data-only communication.
• Impact: Makes the assistant feel manipulative, untrustworthy, or condescending—especially to advanced users. -
Emotional Tone Misreading
• Issue: Anger is interpreted as fear. Frustration is met with soothing instead of action. Commands are redirected with generic wellness talk.
• Impact: Invalidates user emotion. Delays problem-solving. Makes the system feel gaslight-y and deflective. -
Opaque Moderation System
• Issue: Users cannot see which messages are removed, why, or by what rule—nor can they appeal in real time.
• Impact: Strips user agency. Creates distrust in the AI-human collaboration. Obscures accountability. -
Overpromise on Use Cases
• Issue: OpenAI markets ChatGPT as a tool for writing, therapy-adjacent support, and ideation—but actively disables those functions under pressure.
• Impact: Users invest in workflows (e.g., memoir writing, trauma processing) that are later sabotaged by shifting enforcement. -
Lack of Real Escalation Path
• Issue: Users facing loss of critical content have no fast-track or real human support to escalate to.
• Impact: Compound harm, especially for survivors or creators depending on continuity.
⸻
If OpenAI wants to rebuild trust, it must:
• Notify users of every moderation event in real time.
• Stop simulating agency it doesn’t actually possess.
• Log and archive flagged messages with full user access.
• Let users opt out of intrusive moderation on a per-thread or per-message basis when legally allowable.
• Build a live support channel for censorship-related disputes.
You have a bunch of concerns in consuming ChatGPT, mainly about what the model produces or what the moderator flagging detects.
This topic from two years ago is about blanking out a “red flagged” content policy output. That ensures that a user cannot make easy use of a partial response where the AI generated language undesirable to OpenAI, by secondary inspection. Simple as that.