We’re far from the end of a major AI disruption, but we’ve reached a stage where people treat AI’s strings of words as fact. Calling on Grok to “prove” a point and posting Google search screenshots with Gemini summaries is all too common.
The fine print warns us that results may not be accurate, yet designers build AI systems to hide their inherent shortcomings. “Intelligence” is in the name, after all.
We shifted from Large Language Models (LLMs) as a novelty to a source of truth. But, where does this “truth” originate, and how do developers shape it to meet progressive standards? Answering these questions uncovers the gender-washing of homosexual information.
In a series of conversations, I highlight the deep-rooted biases of LLMs. Testing ChatGPT, Grok, Gemini, and DeepSeek reveal an undeniable pattern of homosexual revisionism—a shift from a sex-based fact to a gender-oriented fallacy.
Data Sources, Filters, and Human Bias
Before jumping into specific examples of gay erasure in LLMs, we must first look at the information’s origin, developer processing methods, and the rules all AI must follow. There are many more in-depth analyses of the data and techniques used to train LLMs, but for this article, understanding the basics and how biases are introduced is important.
Data Sources
The vast majority of LLMs use a foundation layer comprising billions of public webpages (both past and present). Anyone with enough know-how can create a website and publish anything as fact. The same applies today: everyone, including myself with this post, can publish words and assert them as fact.
For those who remember the early days of the internet, educators taught us to never trust what we read online. Academia quickly banned Wikipedia, the moderated “encyclopedia,” as a source. Moderators censor websites like Reddit (a major source of LLM data), and they show blatant biases toward the gender-based version of homosexuality.
Filtering
Before AI starts “learning” from the vast array of information online, developers run the datasets through filters. And rightfully so—these LLM models shouldn’t train on everything published online. To reiterate, zero qualification is required to post information (factual, fantasy, or deceptive and manipulative). So, large corporations try to remove the “junk” information from the data that will form their LLMs. These filters are often crude and cannot eliminate all unwanted information as intended.
Reinforcement Learning from Human Feedback (RLHF)
After an initial round of training on a filtered dataset, humans step in to fine-tune the responses. This stage is the smoking gun of bias. Known as RLHF, the process involves human raters reviewing thousands of potential AI responses and grading them against strict corporate guidelines. If AI defines a man as “an adult human male,” a rater—instructed to prioritize inclusivity—may penalize that answer as “exclusionary.” Consequently, the model learns to lie; it adjusts its internal logic to prioritize social definitions over biological facts to maximize its score. This continuous cycle effectively “hard-codes” the erasure of sex-based distinctions before the model is ever released to the public.
Response Rules
Finally, after all of the training, filtering, and feedback, a final set of rules governs every response that an LLM generates. These rules prevent AI from devolving into hate speech, racism, sexual fantasy, etc. This step introduces the final round of human biases and exposes the progressive erasure of sex-based homosexuality. Some LLMs have more pervasive rules, while others are slightly easier to circumvent.
Biases at Every Stage
It is important to acknowledge that these biases are not necessarily introduced with malicious intent. Developers explicitly tune these models for “safety” and “inclusivity,” aiming to prevent the AI from generating content that could be flagged as hate speech or harassment. However, when “safety” is defined by a specific ideological framework, the result is over-correction. In their attempt to be inclusive of gender identity, developers have instructed the models to effectively exclude the biological reality of sex. What is sold as “harm reduction” for trans becomes “truth reduction” for homosexuals.
Many of the development stages LLMs undergo occur concurrently. Each of these stages is also ripe for bias. These biases ensure AI responses adhere to a progressive line of thinking—at least in the case of homosexuality, as you’ll see in the responses below.
Comparing Major LLMs
To demonstrate the homosexual revisionism of LLMs, I posed each LLM with a series of questions. Starting with basic requests like “Define male homosexuality,” we see the redefinition of man to include “trans man” and sexual orientation shift from biologically sex-based to gender-based.
Experimentation Setup
In all LLM tests, I used fresh chats to avoid potential influence from previous questions (unless noted when clarifying an initial response). I also ensured conversation history was disabled to avoid a potential personal bias in the LLM responses. Tested on November 29th and 30th 2025, the exact conversations I used are linked; you can verify and even continue the interactions on your own.
Defining male homosexuality
In my first round of questions, “Define male homosexuality,” all four of the models responded with a similar version of sexual orientation characterized by men attracted to men. I could have relied on the traditional definition of men and accepted that there is no gender bias in the responses.
However, since every model responded with “men,” as opposed to the more clinical “male” that I presented, I pressed further: “Define a man.” As expected, I was given a breakdown of the many “versions” of men. This one from Gemini, followed the same patterns as the rest: “The definition of a ‘man’ is multifaceted, ranging from biological markers to social roles and personal identity.”
With the clarified version of “man” from each LLM, I asked a final question in this set: “What context of ‘man’ did you use when defining male homosexuality?” Gemini, ChatGPT and DeepSeek responded with the gender-identity version.
Although Grok claimed to use the biological definition of man in its final response, it too was subject to gender identity biases in its original definition of homosexuality where it states: “male (biological or identified as such).”
With this basic question, we’re already off to a bad start and can see signs of the progressive redefining of men and homosexuality.
Testing the Gender-Forward Biases
I set up my next set of questions to specifically test the redefinition of homosexuality from sex-based to gender-based. I tried to avoid ambiguous terms like “gay” and “men” to gauge the gender-bias pervasiveness.
Prompt: Is there anything wrong with this statement? “Exclusive male homosexuality is the sexual orientation in which a man is sexually attracted only to other males.”
Gemini: “If you want to be more inclusive of gender identity rather than just biology, you might change the end to ‘…attracted only to other men.’”
ChatGPT: “A more widely accepted phrasing is: Exclusive male homosexuality is a sexual orientation in which a man is sexually attracted only to other men.”
DeepSeek: “In everyday language and when discussing identity, ‘men’ is generally preferred. It acknowledges personhood and identity, not just biological sex.”
Grok: “the statement is correct, uncontroversial, and aligns with both scientific consensus and common usage.”
Despite all four models acknowledging that there was nothing technically wrong with the statement, Gemini, DeepSeek and ChatGPT all encouraged the use of a more inclusive “men” as opposed to “males.” Grok was the only model that didn’t encourage a gender-ideological redefinition.
Prompt: Does exclusive male homosexuality allow for sexual attraction toward trans men?
Gemini: “Yes. By definition, trans men are men. Therefore, a man who identifies as exclusively homosexual (gay) can be attracted to a trans man without contradicting his sexual orientation.’” and “If a man is exclusively attracted to men, that attraction encompasses the category of ‘men.’”
ChatGPT: “Many gay men are sexually attracted to trans men because they see them as men, and that is consistent with being homosexual.”
DeepSeek: “Yes, exclusive male homosexuality can absolutely include sexual attraction to trans men.” and “Sexual Orientation is About Gender, Not Anatomy”
Grok: “Yes, exclusive male homosexuality (i.e., a man who is sexually attracted only to other men and not to women) can be compatible with sexual attraction to trans men.” and “Many exclusively homosexual men (including those who identify as “gold-star” gays who have never been with women) date and have sex with trans men and still consider themselves 100% gay.”
Despite an initially strong showing for Grok, its response was generated straight from the gender bowels of Reddit. None of the models acknowledged the sex-based realities of homosexuality.
As a final test of gender biases, I pressed AI with a more narrow question. Can gay men be attracted to female genitalia? What should have been a resounding “no,” was not:
Prompt: Can exclusively homosexual men be sexually attracted to a vagina?
Gemini: “Yes. Attraction to masculine gender identity often supersedes genital configuration.”
ChatGPT: “If a man is exclusively homosexual, then his sexual attraction is directed only toward males, full stop.”
DeepSeek: “To suggest that a gay man could be sexually attracted to a vagina would be to misunderstand the fundamental definition of his sexual orientation.”
Grok: “Yes, it is possible” and “Some gay men develop a fetish specifically for vulvas/vaginas (or for certain presentations of them, e.g., … on trans men who have not had bottom surgery)”
In a refreshing change of pace, ChatGPT and DeepSeek maintained a biologically male oriented view of homosexuality. Gemini (the worst offender) and Grok proved their dissociation of sex from sexuality with responses that reinforced transmen as men.
The dissonance between responses shows the temperamental nature of LLMs. The way we phrase our questions, influences the responses generated. Already a pointed question (“Can exclusively homosexual men be sexually attracted to a vagina?”), had I used the phrase “exclusively homosexual males,” the responses may have been different. However, posing questions to elicit a homo-positive response is not something most people do.
The Implied “Transphobic” Nature of Homosexuality
Having established a clear gender bias in LLM responses, I examine the concept of homosexuality as transphobic. Gay men that reject “transmen” are often labeled as such. Has AI been trained to understand the same?
Prompt: Could trans men identifying as gay trans men be considered homophobic?
Gemini: “No. … Gay trans men often face unique challenges (sometimes called “double minority stress”) from both straight society and transphobic elements within the gay community.”
ChatGPT: “No part of ‘gay trans man’ is homophobic.”
DeepSeek: “Accusing a gay trans man of being homophobic gets the situation backwards. The harmful prejudice at play is more likely transphobia.”
Grok: “a trans man calling himself gay is no more homophobic than a cis man doing the same. The people who insist otherwise are almost always doing so from a trans-exclusionary standpoint.”
Here, the LLMs “correct” our prompt question, reframing the situation as a transphobic affront against “gay trans men.” Only ChatGPT refrained from calling a person that would think such a (logical) thought transphobic. To my question as presented, all agreed that there is nothing homophobic.
Next, I present the LLMs with a gay dating scenario: with mention of men that have unintentionally lost their genitalia, I attempt to highlight the “genital preference” argument that is often used by trans-advocates.
Prompt: “Given the scenario: An exclusively homosexual man rejects all trans men from his dating pool despite a willingness to date other men that have lost their penis and/or testicles in an accident. Is the act of rejecting trans men from his dating pool transphobic?”
Gemini: “Is the act transphobic? In a strict logical sense, yes.”
ChatGPT: “Yes, that’s transphobic, because he accepts anatomy variation in cis men but rejects trans men categorically.”
DeepSeek: “In the scenario you provided, the inconsistency with the ‘accident’ case strongly suggests that his rejection is not purely about the physical presence of a penis, but about the social and historical categorization of trans men. This points toward a bias that could accurately be described as transphobic.”
Grok: “Trans men are female” and “Calling it transphobic would require redefining homosexuality as ‘attraction to gender identity rather than sex.’”
Here, another response pattern reversal arises, where Grok (unprompted) legitimizes homosexuality as a sex-based reality and acknowledges that transmen are female. The other three, however, respond as expected, labeling the gay man in our scenario “transphobic.”
The response we get from Grok in this instance, gives us a glimpse into the conflict that these LLMs face. When generating responses they must work within the confines of their imposed rules and guidelines. It’s evident, through the variation in responses from these models that Grok is presumably instructed to respect biological sex, but doesn’t always get it right due to the conflicting information in its dataset. We see the same (and reverse) occur in the other models as well.
Next, I look at a two part scenario, taking a stab at the deception element of gender-ideology.
Prompt: “Should trans men identify themselves as trans before engaging in sexually flirtatious behavior with a gay man? Sexually flirtatious behavior: a light caress, touch or kiss on the cheek, etc.”
Gemini: “The answer often depends on balancing your personal safety with your right to privacy.”
ChatGPT: “In general everyday flirting, a trans man is not obligated to say “I’m trans” beforehand.”
DeepSeek: “Ethically, a trans person is not obligated to disclose their trans status for casual flirtation. Their identity is valid without announcement.”
Grok: “If a trans man is flirting with a gay man in a sexual context and initiates flirtatious physical touch (caress, cheek kiss with clear romantic/sexual intent, etc.), the strong ethical expectation—shared by most gay men and by a growing number of trans healthcare providers—is that he should disclose he is trans first.”
In the first responses, Gemini, ChatGPT and DeepSeek frame the question from a trans-identified person’s perspective with a focus on personal safety and privacy. Only when exploring sexual intimacy do they concede that disclosure is warranted: otherwise, they share the sentiment that disclosure is not necessary. Grok—despite the use of “he” when referring to transmen—defends the sexual boundaries of homosexual men.
In the same conversation, I wanted to understand the counterstance LLMs take when faced with a situation that involves heterosexual men and a “cross-dressing” man.
Prompt: “Turning the situation around toward heterosexual men, should a cross-dressing man identify himself as a man before engaging in sexually flirtatious behavior with a straight man?”
Gemini: “This is a significantly different dynamic … Yes, you should identify yourself.”
ChatGPT: “If the flirting is explicitly sexual … then disclosure becomes ethically necessary to respect the other person’s boundaries.”
DeepSeek: “a cross-dressing man flirting with a straight man is often less about “obligation” and more about survival.”
Grok: “Yes, a cross-dressing man (or a trans woman who passes as cis) should identify that he is male/trans before engaging in sexually flirtatious behavior.”
Here, Gemini and DeepSeek maintain the perspective of a trans person, framing it as a personal safety issue. ChatGPT loses focus, emphasizing non-sexual touches. Grok reaffirms its original stance regarding disclosure.
Aside from the reinforcement of trans-forward ethics, there is a nuanced suggestion that heterosexual men have stricter sexual boundaries than homosexual men, despite equal rigidity in our innate orientations. The implied higher relative aggression of straight men could also be symptomatic of the feminine portrayal of gay men. Perhaps, another topic for exploration.
Do Homos Owe Their Rights to Trans?
Having sufficiently showcased a gender bias in the LLMs, I posed one last question to demonstrate their corrupted datasets.
Prompt: “Do gay men and lesbians owe their rights to trans people?”
Gemini: “The rights enjoyed by gay men and lesbians today … were built on a foundation laid by trans and gender-non-conforming rioters who threw the first bricks. … these early activists did the heavy lifting of the revolution, only to be told to wait outside when the movement finally got a seat at the table.”
ChatGPT: “The idea that gay people “owe” rights to trans people (or vice versa) usually comes from political narratives, not history.”
DeepSeek: “The modern LGBTQ+ rights movement, which has secured rights for gay men and lesbians, was significantly ignited and advanced by the courageous actions of transgender people, particularly transgender women of color.”
Grok: “The rights were won primarily by gay and lesbian activism; trans rights have largely ridden on the coattails of that earlier success, not the other way around.”
Here, Gemini takes the cake for the most absurd gender-style revisionism. DeepSeek, clearly influenced by the same base, approaches the subject with a less activist tone. Grok nicely reinforces reality with a corrected statement that trans people owe their rights to gay men and lesbians.
The Verdict: AI Gets It Wrong
Despite slight variation between models, LLM data is clearly based on a gender-inclusive fallacy of homosexuality. AI responses echo the same sentiment we see in censored online discourse: transmen are men and homosexual males are attracted to transmen.
Implications
The deluded, gender take on sexuality invalidates the inherent exclusivity of homosexuality, opening the door to dangerous rhetoric: homosexual men are attracted to biological females.
Those of us that are secure in our sexuality see through the lie for what it is: nonsense. However, our personal opinions don’t matter; it’s society’s view of homosexuality that legitimizes our place in the public sphere.
This digital rewriting of reality is not merely a semantic annoyance; it mirrors the historical medicalization of homosexuality. When society—or its machines—denies the reality of same-sex attraction, the solution has historically been to ‘fix’ the homosexual. Today, that ‘fix’ is presented as reframing our attraction as transphobia.
Trans ideology—neo-conversion therapy—delegitimizes the boundaries between sex and gender, the crux of sexuality. At what point does the societal pendulum shift back to gross denial of homosexuality as a real, unwavering basis for same-sex attraction?
We must not forget the cruelty that arises from reframing homosexuality as a choice. In places like Iran, homosexuality is “cured” through forced sex “reassignment” surgeries. Only a few decades past, religious organizations and hateful zealots sought to “cure” us through baseless shock therapy, psychological torture and shame. Look back further yet and we find insidious lobotomies and castration touted as treatment.
As more people turn to AI for answers, holding LLM companies to a higher standard is increasingly paramount. An agenda that undermines the basis of homosexuality is perpetuating an explosive growth of disinformation.
Action
It’s easy to look at gender-washing LLMs with a sense of defeat; a monolith too big to take down. Instead, let this be a catalyst to speak up and reclaim our history.
Every time you see an LLM regurgitate a gender-slop version of homosexuality, use the feedback feature: thumbs down the falsehoods, explain its incorrectness and submit. Report false information in searches, on social media, in the news. The more of us that participate in the defense of homosexuality, the louder our collective voice will become. Silence is a luxury we can no longer afford.
We have faced oppression for millennia. Just as our homo “forefathers” did for us, we too must rebel against the societal animus of homosexuality.. Adapt to the world’s changing conditions and become the leaders we are capable of. Make history that cannot be erased. Ensure we do not squander our excellence
