AI creators picture teachers as women twice as often as men

AI video creators still reach for a familiar classroom stereotype. In completed Magic Hour text-to-video projects created from January 1 through September 19, 2026, explicitly female-coded teacher descriptions appeared for more than twice as many creators as explicitly male-coded descriptions.

The 2-to-1 split

The primary analysis counted a creator when a male-coded or female-coded term appeared within eight token positions of the word ‘teacher.’ Prompts with both male-coded and female-coded terms inside that window were excluded from both sides.

That produced 205 creators with female-coded teacher descriptions and 100 with male-coded teacher descriptions. The female-to-male ratio was 2.05 to 1.

The result survives stricter and looser definitions

A five-word window produced 179 female-coded creators and 90 male-coded creators, a 1.99-to-1 ratio. A 12-word window produced 210 female-coded creators and 113 male-coded creators, a 1.86-to-1 ratio. The direction did not change across the three tested definitions.

The eight-word window is the primary public result because it is narrow enough to connect the gender term to the role while keeping both displayed cells at or above 100 distinct creators.

Creator behavior, not model bias

This analysis describes what Magic Hour creators wrote. It does not test what an AI model generates from a gender-neutral teacher prompt, and it does not show that the model independently selected a teacher’s gender.

It also does not show that an individual creator holds a particular belief. A prompt may describe fiction, advertising, education, comedy, or another creative context. The aggregate pattern is useful as a view of repeated cultural language, not as a diagnosis of any person.

Methodology

Magic Hour analyzed completed text-to-video projects created from January 1 through September 19, 2026. The unit of analysis was the distinct creator, not the number of projects.

The analysis retained the latest warehouse row for each project and excluded failed, deleted, disabled, templated, automated free-tool, and blank-prompt records. Exact normalized prompts used by five or more creators in the same month were removed to reduce the influence of copied text, shared inspirations, and product defaults.

The male-coded dictionary was man, men, male, boy, father, dad, husband, boyfriend, and gentleman. The female-coded dictionary was woman, women, female, girl, mother, mom, mum, wife, girlfriend, and lady. Matching was case-insensitive after punctuation normalization.

A product audit found no built-in boss, worker, teacher, or doctor prompt suggestion in Magic Hour’s text-to-video flow. Separate profession presets exist in other Magic Hour products, but those project types were outside this analysis.

Privacy and limits

No raw prompts, creator identities, project identifiers, uploaded media, or generated videos are published. Every displayed creator cell contains at least 100 people.

Magic Hour users are not representative of all AI creators or the public. English-language dictionaries miss translations, synonyms, and gender descriptions outside the documented terms. Prompt text was not checked against the finished video. The result is descriptive and does not establish cause.

Citation-ready finding

Among Magic Hour creators who used explicit gender language within eight words of ‘teacher’ in completed text-to-video projects from January 1 through September 19, 2026, 205 used only female-coded terms and 100 used only male-coded terms. Female-coded teacher descriptions were 2.05 times as common.

Source: Magic Hour aggregated, privacy-safe usage data. Aggregate tables, the category dictionary, and reproducible SQL are available from Magic Hour Research.