Machine learning models show promise for predicting speech-in-noise performance, but until clinical validation benchmarks and standardised input features are established, audiologists should not replace behavioural speech-in-noise testing with AI-predicted scores.