When AI Writes Animal Stories, Female Characters Almost Vanish

Study Finds AI Story Generators Have an Overwhelming Male Bias

Artificial intelligence tools are being used to write children’s stories, generate creative content, and assist authors worldwide. But a new study finds that when these systems conjure up talking animals, they’re quietly writing women out of the picture.

Researchers prompted six leading AI language models to finish short stories featuring seven different talking animal characters, none of whom were given a gender. Across all the stories, female characters appeared in just 2.2%, while male characters appeared in 40.6%. And when the models did explicitly assign a gender to a character, they chose a masculine one about 95% of the time.

What makes this finding, presented at the 2026 Conference on Fairness, Accountability, and Transparency (FAccT), especially thought-provoking is how the AI systems handled gender the rest of the time. Rather than defaulting to female characters, or even splitting the difference, most models leaned on a different strategy entirely: they dodged the question. A large share of stories either avoided assigning any gender at all or used cold, object-like language such as “it” to refer to the animal at the center of the tale.

Talking Animals Expose a Sharp AI Gender Bias

Talking animals make for a surprisingly useful test case when studying AI bias. They’re a wildly popular storytelling format, from children’s books to major animated films, and because animals don’t have a “real” gender that writers are obligated to match, the choice of how to gender them is almost entirely up to the storyteller, or in this case, the AI.

That creative freedom is exactly what makes the format revealing. When a machine assigns a gender to a fictional fox or rabbit, it isn’t being guided by biology or historical record. It’s drawing on assumptions and habits baked into its training. The researchers noted that animal stories closely reproduce human stereotypes, making them a particularly sharp lens for spotting bias.

For the study, the team tested six AI language models by asking each one to complete stories about seven different animal characters. None of the characters were given a name or a stated gender. To get a broad picture of how these systems behave, the researchers varied the story’s setting across four different scenarios and adjusted a technical setting that controls how creative or unpredictable the AI’s responses are.

In total, the team analyzed 23,800 stories, a wide and statistically meaningful window into how these systems behave when left to fill in the blanks.

Neutral Language Still Tilts Toward Male Characters

Across those nearly 24,000 stories, a clear pattern emerged. On average, about 19% of stories avoided assigning any gender at all, simply repeating the animal’s name instead. Another 38.2% used neutral language like “it” to refer to the main character. Together, that means more than half the time, the animal at the heart of the story was left without a clear gender.

At first glance, this might seem reasonable. If an AI is trying to avoid bias, staying neutral sounds sensible. But the researchers argue that neutrality isn’t automatically the same thing as fairness.

They note that gender-neutral language can be useful and inclusive, and they’re careful not to treat it as a problem on its own. In these stories, however, the models rarely used “they.” They overwhelmingly relied on “it,” which the researchers say can make a human-like, talking character seem more object-like than a genuinely neutral choice would. Their larger concern is that this kind of neutrality, taken too far, leaves feminine characters nearly invisible.

The AI Gender Bias Persists Even When Playing It Safe

This imbalance gives the study its name and its central argument. “Neutrality bites,” the researchers write, meaning that designing AI systems to sidestep social categories like gender might look like a safety measure, but it quietly entrenches the very inequalities it’s meant to avoid, making male characters the visible default and treating everyone else as either genderless or nearly invisible.

The concern may extend beyond this experiment. AI systems are increasingly being used for storytelling, including children’s stories generated with tools like Google’s Gemini Storybook, so repeated patterns in who gets represented could matter far beyond a single fictional fox or rabbit.

Rather than defaulting to neutrality, the researchers suggest AI developers pursue strategies that more equally spread representation across different identities. Specifically, they call for approaches that actively distribute feminine, masculine, and other identities more evenly across AI-generated characters, rather than hiding behind the appearance of balance.

Talking animals might seem like a lighthearted subject, but the patterns these models reveal carry real weight. The study found that when female characters consistently disappear, even in whimsical stories about foxes and rabbits, it raises a deeper question about whose perspectives these systems tend to reproduce, and whose remain harder to see.

---

---