Gender Differences in Conditional Willingness and Trust Expectations for AI Legal Services in Bangladesh

Abstract Artificial intelligence may widen access to legal information while reproducing inequalities in who can use, evaluate and challenge its outputs. Research on gender and legal AI therefore needs to distinguish conditional willingness to use a service from confidence that its design supports trustworthy operation. This study examines that distinction in the supplied Group 10 Law-SDG database, linking gender and technology, procedural justice and digital inclusion perspectives. All 385 records inform descriptive analysis; comparisons between respondents recorded as female and male use 364 records, with the remaining gender responses retained descriptively. Item-level analysis combines distributional comparisons, stratified bootstrap intervals, paired contrasts and covariate-adjusted logistic models. Conditional willingness was endorsed by 307 records (79.7%), whereas 198 (51.4%) endorsed the proposition that inclusive design can improve trust. Their paired difference was 28.3 percentage points (95% bootstrap interval 21.8–34.8). The female-minus-male willingness difference was − 0.7 points (-9.0-7.5); the adjusted difference was 1.0 point (-7.4-9.4). Inclusive-design trust expectations and discrimination salience likewise showed small estimated gender contrasts with substantial uncertainty. Low block reliability precluded latent-scale modelling. The findings support neither a gender-deficit explanation nor a claim of established equality: similar reported willingness can coexist with unequal access and unmeasured exposure to harm. The contribution is a measurement-conscious distinction between conditional acceptance, trust expectations and substantive inclusion. Interpretation remains bounded by undocumented recruitment and by codebook labels identifying respondent sequences and comments as synthetic; numeric response provenance requires verification before these records can support claims about a human population.