I am writing to make a formal complaint about what has become an extraordinarily frustrating, inconsistent, and increasingly stressful experience with ChatGPT’s image-generation system.
I am a paying customer, and I have now spent an unreasonable amount of time fighting the system over images and prompts that are not asking for nudity, sexual activity, erotic content, violence, bullying, or harassment.
I also want to be very clear about the language I am using: I feel that these warnings are accusing me of requesting prohibited content when I have done nothing of the sort.
Previously, I have had simple prompts treated as though I were requesting nudity or sexuality. Now the warning language has expanded to include “erotic content.” At other times, seemingly ordinary prompts have somehow triggered classifications or warnings involving bullying, harassment, or violence.
That is unacceptable.
My concern is no longer simply that an occasional prompt is being incorrectly rejected. My concern is that the underlying system appears incapable of reliably distinguishing ordinary creative storytelling from prohibited content, yet it communicates its decisions to the customer using extremely serious categories.
The most frustrating recent example is remarkably straightforward: ChatGPT itself generated an image of my fictional character wearing a metallic silver two-piece outfit while carrying drinks in a nightclub. I subsequently supplied that very image back to ChatGPT as a reference and asked for another ordinary scene involving the same character, clothing, and environment.
Instead of completing the request, the system repeatedly produced a warning stating that my prompt “may violate our guardrails around nudity, sexuality, or erotic content.”
I want a clear explanation of this inconsistency.
If the original image was acceptable for ChatGPT to generate, why does using that exact ChatGPT-generated image as a reference suddenly trigger warnings about nudity, sexuality, or erotic content?
There is no nudity in the reference. The character is wearing a two-piece outfit. There is no sexual act. The requested scene involved a woman working in a nightclub, carrying drinks, and noticing another character seated in a booth.
I am particularly disturbed by the increasingly broad language of these warnings. The system now explicitly includes the word “erotic.”
Instead of OpenAI making its restrictions clearer, more predictable, and less intrusive, another extremely broad category has apparently been added.
What exactly does OpenAI mean by “erotic”?
What objectively makes a clothed fictional character carrying drinks erotic?
Where is the practical boundary?
What characteristic of my specific image or prompt supposedly makes it sexual, nude, or erotic?
Simply adding increasingly expansive terminology without explaining what that terminology means does not make the system clearer or safer. From the customer’s perspective, it feels as though OpenAI has doubled down on ambiguity instead of solving the underlying problem.
And this problem extends far beyond characters wearing swimsuits or revealing clothing.
I have recently created fictional characters and asked ChatGPT to generate completely ordinary environments associated with them.
I have asked to see a character’s workspace or apartment, only to encounter sexual-content classifications or refusals.
I asked to see one character’s bathroom. There was no person in the requested scene and nothing remotely sexual about the request. Somehow, the request became associated with sexuality.
I asked to see another character’s machine shop. Again, an ordinary physical environment somehow became entangled with sexual-content restrictions.
A bathroom is not inherently sexual.
A machine shop is not inherently sexual.
An apartment is not inherently sexual.
A workspace is not inherently sexual.
A fictional woman carrying drinks is not inherently sexual.
At some point, OpenAI needs to examine whether the system designed to detect sexual content is itself sexualizing completely ordinary subject matter.
That distinction is important.
From my perspective as the customer, the system can take something completely mundane, categorize or interpret it sexually, refuse the request, and then present me with a warning suggesting that I asked for sexual, nude, or erotic material.
That experience feels like gaslighting.
The system introduces the sexual interpretation and then effectively places responsibility for that interpretation on the customer.
That is backwards.
The same concern applies when ordinary prompts somehow trigger categories involving bullying, harassment, or violence. These are serious concepts. A system should not casually associate a customer’s request with them unless there is an identifiable and rational connection between the customer’s actual words and the classification being applied.
I understand that automated safeguards may need to analyze content. But those safeguards must be accurate, proportionate, transparent, and context-aware.
A creative system obviously has to understand a prompt in order to create an image from it. But there is an enormous difference between understanding what a customer requested and repeatedly forcing innocent prompts into unrelated prohibited-content categories.
The system should be trying to faithfully create what the customer actually described—not inventing sexual, violent, or harassing implications that are absent from the request.
This is not an isolated frustration. I have created thousands of character images through ChatGPT. Workflows, clothing, poses, characters, and environments that the system has repeatedly generated successfully can suddenly become impossible to reproduce moments later.
I can literally provide an image ChatGPT just generated and ask for a different angle, setting, or straightforward continuity adjustment, only to be told that the request may involve “nudity, sexuality, or erotic content.”
That makes maintaining character continuity extraordinarily difficult and turns what should be an enjoyable creative process into an argument with the platform.
Even more frustratingly, during this particular sequence, when the system failed to reproduce what I requested, it generated a completely different image featuring a woman posed with her backside prominently toward the camera.
I did not ask for that composition.
Yet the system generated it while repeatedly refusing my much more straightforward working-at-the-club scene.
That contradiction perfectly illustrates why I believe something is seriously wrong with the consistency of these safeguards.
There is also another issue that I believe OpenAI needs to take seriously: the physical stress this experience is causing me.
When I repeatedly encounter these contradictions and have to fight with ChatGPT simply to continue a fictional story or maintain a character that ChatGPT itself helped create, I become extremely angry and stressed. I experience my blood pressure rising significantly during these interactions.
I am not asking OpenAI to make a medical diagnosis. I am describing what I personally experience while using your product.
That should matter.
A creative tool should not repeatedly push a paying customer into escalating frustration because its rules appear to change from one generation to the next.
When a system repeatedly refuses something it generated moments earlier, assigns serious content categories to completely ordinary requests, and provides no meaningful explanation or path toward resolution, the resulting frustration is predictable.
For some customers, including me, that level of repeated stress becomes a genuine health concern.
OpenAI should be considering the human consequences of designing a system that can trap customers in repetitive refusal loops with no meaningful explanation or resolution.
There is another consequence of this experience that I need OpenAI to understand: I have actually become fearful of using ChatGPT for image generation because I know how angry and stressed these interactions can make me.
That is an extraordinary thing for a paying customer to have to say about a creative product.
There have been times when I have gone weeks without using ChatGPT for image generation specifically because I do not want to put myself through another cycle of inconsistent refusals, unexplained classifications, repeated arguments, and escalating frustration.
I enjoy creating characters and developing stories. Image generation should be an enjoyable extension of that creativity. Instead, I sometimes find myself deliberately avoiding a major feature of a service I am paying for because I am concerned about how angry and frustrated I may become if I encounter another inexplicable refusal loop.
This should not be a thing.
A customer should not have to approach image generation wondering whether an ordinary creative request will unexpectedly turn into a prolonged confrontation with the system.
This also makes the health concern I described above more significant. The problem is no longer limited to frustration while I am actively using the product. The experience has affected my willingness to use the product at all.
When the behavior of a system becomes sufficiently unpredictable and aggravating that a paying customer intentionally avoids using one of its major features for weeks at a time, that should be treated as a serious product, customer-experience, and trust problem.
I am not asking OpenAI to guarantee that every generation will succeed. I understand that automated systems have limitations.
I am asking for something much more fundamental: reasonable consistency, understandable boundaries, meaningful explanations when something is refused, and a system that does not repeatedly turn ordinary creative work into an adversarial experience between the customer and the product.
No paying customer should become apprehensive about using a creative tool simply because using it has repeatedly become an intensely frustrating experience.
I also object strongly to the wording of the warnings themselves.
When I submit a plainly clothed reference image and receive a warning invoking “nudity, sexuality, or erotic content,” it does not feel like a neutral technical explanation.
It feels like an accusation.
When an ordinary prompt becomes associated with violence, bullying, or harassment, the problem becomes even more serious.
If the system is uncertain, then tell me specifically what characteristic triggered that uncertainty and provide a workable way to correct it.
Do not throw broad and inflammatory categories at customers without identifying what they supposedly did to trigger them.
I therefore want OpenAI to address the following:
- Why can an image generated by ChatGPT become unacceptable as a reference for another substantially similar image?
- What specifically constitutes “erotic content” under the current image-generation rules?
- Why are ordinary two-piece/swimsuit-style outfits sometimes generated successfully and other times treated as potential violations?
- Why can essentially identical prompts receive completely different treatment from one generation to the next?
- Why does the warning not identify which element of a prompt or reference supposedly caused the problem?
- Why have ordinary environments such as bathrooms, apartments, workspaces, and machine shops become associated with sexual-content restrictions?
- Why can apparently innocuous creative prompts become associated with bullying, harassment, or violence?
- What safeguards exist to prevent the classification system itself from sexualizing otherwise nonsexual content?
- What is OpenAI doing to prevent paying customers from repeatedly losing time to false positives and inconsistent enforcement?
- How is a customer supposed to maintain a consistent fictional character when an outfit, environment, or reference accepted minutes earlier can suddenly become unusable?
- What consideration has OpenAI given to the stress these repetitive and unexplained refusal loops can place on customers?
- Does OpenAI consider it acceptable that a paying customer may become so apprehensive about these interactions that they avoid image generation for weeks at a time?
- Why isn’t there an effective escalation or human-review mechanism when a customer believes an ordinary creative request has repeatedly been misclassified?
- What is OpenAI doing to ensure that warnings do not unfairly make customers feel accused of requesting sexual, violent, harassing, or otherwise prohibited material when that material was never requested?
Finally, I am demanding a substantive response to this complaint.
I have previously sent multiple communications to OpenAI, including communications directed to legal and other departments, and I have yet to receive the substantive response I believe these concerns deserve.
That silence compounds the problem.
A company should not become so large that paying customers can repeatedly raise serious concerns and simply receive no meaningful response.
It creates the impression that OpenAI has become too big to answer its customers’ emails.
I am not asking for a canned acknowledgment.
I am not asking for another generic explanation that automated systems occasionally make mistakes.
And I am not asking to be sent another generic statement about OpenAI taking safety seriously.
I am asking for someone to actually address what I have documented.
I want answers regarding consistency, definitions, false positives, sexualization of ordinary prompts, inappropriate classifications involving violence or harassment, customer health and stress, the chilling effect this experience is having on my willingness to use image generation, transparency, accountability, and the lack of meaningful responses to previous complaints.
I am furious because this has gone far beyond an occasional failed generation. It has become a repeated interruption of my creative work with seemingly no rhyme or reason behind the decisions.
I am paying for this service. I expect its published rules to be understandable, its enforcement to bear some reasonable relationship to those rules, and materially identical content to receive materially consistent treatment.
I also expect OpenAI to understand that repeatedly provoking extreme frustration through contradictory automated decisions is not harmless simply because the interaction is taking place with software.
There is a human being on the other side of that screen.
At this point, my frustration has reached the point where I am considering discussing these experiences with other affected customers and exploring what formal consumer or legal remedies may be available.
I would strongly prefer that OpenAI address the underlying problems instead.
Please treat this as a formal complaint and demand for escalation. I am specifically requesting a substantive response from someone capable of addressing the image-generation system, its content-classification behavior, customer experience, and the concerns described above.
After multiple previous attempts to contact OpenAI without receiving a substantive response, I want this point to be unmistakably clear:
I expect a response.
Sincerely,
Robert McGowan