AI Models Rank Accented English Speakers Unequally
A University of Washington study found major AI models scored speakers of different English accents differently in interviews and tests.
Updated on Nov. 2, 2026 in Artificial Intelligence
Live Poll
Would you trust an AI tool to fairly judge your job interview answers?
Researchers at the University of Washington tested AI models from Google, OpenAI and Alibaba using English read with American, British, Chinese, Indian and Nigerian accents. The models generally ranked American and British speakers highest and Chinese and Nigerian speakers lowest, though they transcribed the accented speech accurately.
Why it matters
The findings raise concerns about the fairness of AI tools that evaluate spoken language, such as automated scoring in language tests. Because the models scored speakers unequally on the same script, the bias could affect how people are assessed based on how they speak.
The study compared five English accents across general-purpose AI models from Google, OpenAI and Alibaba using identical scripted readings. It remains unresolved whether the same scoring gaps occur in commercial products used for hiring or testing.
The players
University of Washington
A public research university in Seattle whose researchers conducted the study on accent bias in AI models.
Pearson PTE Academic
An English-language proficiency test by Pearson that uses AI to score some spoken answers and was administered about 1.2 million times in 2023.
The details
The researchers found the models transcribed accented speech accurately but still scored content differently by accent, sometimes lowering scores for Chinese and Nigerian speakers. The researchers suggest the bias may stem from training data that describes accented speakers as unclear or from data sets that underrepresent some accents.
Timeline
Oct. 25, 2025: Research presented at the EMNLP conference in Budapest.
2023: Pearson's PTE Academic test, which uses AI to score some spoken answers, was given about 1.2 million times.
Big Picture
The study places accent bias within the expanding use of AI to score English-proficiency exams such as Pearson PTE Academic, which was given about 1.2 million times in 2023. It shows that general-purpose models, not only dedicated scoring products, can produce unequal results across accents.
Test-takers and job applicants whose spoken English is accented may face scoring that depends on accent rather than content. Those relying on AI-scored language tests should be aware that results may vary by speaker accent.
The takeaway
Automated scoring tools should be checked for accent-based differences before they are used to judge people's language ability. Speakers should understand that AI-assessed results may not reflect content alone.
Further reading
Explore more reporting in our Artificial Intelligence section.
Source note: This article includes information reported by GeekWire.
Live Poll
Would you trust an AI tool to fairly judge your job interview answers?
