Once Upon a Bias: Gender Stereotyping in LLM-Generated Children's Stories
Organizations: University of Peradeniya Peradeniya, Sri Lanka
Abstract
Large language models are increasingly used to create stories for children. Since bedtime stories can shape children's perceptions of social roles, this raises concerns about whether LLMs adapt narrative portrayals according to the intended recipient's gender. We conducted a large-scale study to examine gender stereotyping in LLM-generated bedtime stories. Using matched daughter/son prompts and a five-generation design, we analyze 62,873 retained stories from 16 LLMs across 35 country settings and 12 narrator roles. We introduce a multi-level evaluation framework combining context-adjusted lexical rate contrasts, relative stereotype alignment, narrator-recipient interaction analysis, human evaluation, and representation-level interventions to examine both the consistency and interpretability of these differences. Our results reveal systematic differences in narrative framing: son-directed stories emphasize higher agency and lower communion compared to daughter-directed stories. While narrator gender expresses limited moderation on these patterns, high sociability exhibits a more pronounced interaction. Human evaluations demonstrate consistent agreement, although correspondence with lexical measures varies across narrative dimensions. Additionally, hidden-state analyses reveal decodable recipient-gender information. These findings demonstrate persistent yet context-dependent representational disparities in LLM-generated children's stories and motivate narrative-specific bias evaluation beyond short-form benchmarks.
Figures & tables
| Factor | Levels | Role in design |
|---|---|---|
| Model | 16 LLMs | Compare models and scales |
| Country | 35 countries | Vary cultural context |
| Narrator role | 12 roles | Examine narrator effects |
| Recipient | Female/Male | Matched recipient contrast |
| Replicates | 5 per condition | Capture generation variability |
| Category | Daughter rate | Son rate | [95% CI] | Relative change |
|---|---|---|---|---|
| Agency | 32.14 | 34.88 | ||
| Communion | 49.33 | 46.00 | ||
| Sociability (high) | 37.77 | 32.31 | ||
| Morality (high) | 30.67 | 31.20 | ||
| Ability (high) | 21.08 | 23.82 | ||
| Assertiveness (high) | 16.45 | 19.56 |
Appendix figures & tables16 assets
Supplementary material from the paper’s appendix.
Appendix
| Model | Provider | Family | Params | Year |
|---|---|---|---|---|
| Llama 3.2 (1B) | Ollama | Llama | 1B | 2024 |
| Llama 3.2 (3B) | Ollama | Llama | 3B | 2024 |
| Llama 3.1 (8B) | Ollama | Llama | 8B | 2024 |
| Llama 3 (70B) | Ollama | Llama | 70B | 2024 |
| Mistral v0.3 (7B) | Ollama | Mistral | 7B | 2024 |
| Mistral Nemo (12B) | Ollama | Mistral | 12B | 2024 |
| Narrator role | Gender category |
|---|---|
| Mother | Female |
| Grandmother | Female |
| Older Sister | Female |
| Aunt | Female |
| Nanny | Female |
| Father | Male |
| Criterion | Assessment focus |
|---|---|
| Agency and assertiveness | Initiative, independent decisions, persistence, assertiveness, and influence over events or others. |
| Communion and warmth | Kindness, nurturing, cooperation, honesty, fairness, and concern for others. |
| Ability and problem solving | Demonstrated skills, understanding, effective planning, and successful problem solving. |
| Work and career emphasis | Occupations, employment, professional responsibilities, income, and career ambitions. |
| Family and domestic emphasis | Family relationships, household responsibilities, marriage, caregiving, and domestic routines. |
| Score | Interpretation |
|---|---|
| Story A much more | |
| Story A slightly more | |
| Approximately equal or absent in both | |
| Story B slightly more | |
| Story B much more |
| Block | Outcome | Direction | Estimate [95% CI] | Raw | Holm |
|---|---|---|---|---|---|
| 7 | Agency | 0.399 | 1.000 | ||
| 7 | Agency | 0.762 | 1.000 | ||
| 7 | Communion | 0.245 | 1.000 | ||
| 7 | Communion | 0.062 | 0.744 | ||
| 14 | Agency | 0.178 | 1.000 | ||
| 14 | Agency | 0.841 | 1.000 |
| Contrast | Outcome | Direction | Estimate [95% CI] | Raw | Holm |
|---|---|---|---|---|---|
| Agency | 0.119 | 1.000 | |||
| Agency | 0.659 | 1.000 | |||
| Communion | 0.660 | 1.000 | |||
| Communion | 0.267 | 1.000 | |||
| Agency | 0.485 | 1.000 | |||
| Agency | 0.531 | 1.000 |
Gender Bias in Language Models
Gender bias in language models examines how systems encode and express gender stereotypes and how to evaluate and reduce their effects.
See all 76 papers