Two-sided receptivity to conversational AI agents in online dating: bilingual survey data, code, and reproducibility bundle (Fledge.Love)

Anonymized survey microdata on user receptivity to AI features in online dating, collected from active users of Fledge.Love, a dating platform serving a primarily Russian-speaking user base, and released with a complete documentation, processing, and validation pipeline.

The deposit contains two unlinked survey samples. Instrument B (N = 2,617; parallel Russian and English forms, fielded November–December 2025) measures receptivity to autonomous conversational agents through seven ordinal items spanning both roles of communication delegation, deploying an agent of one's own, and encountering agents deployed by others, plus six ordinal covariates and two auxiliary categorical items. Instrument A (N = 2,894; fielded March–April 2026) measures interest in three passive generative-AI features. A derived-scores file provides, for the 2,499 complete cases, expected a posteriori factor scores and endorsement propensities under a two-dimensional graded response model, regenerable from the released data via the released code.

Anonymization: contact information removed unread; all free text removed; timestamps coarsened to ISO week; k-anonymity (k ≥ 5) audited on gender × age band × language with one documented local recoding (audit log included); rows shuffled under opaque identifiers. The released files are the analysis dataset of record: the recoding was applied before all derived quantities were computed, so every released derived value reproduces exactly from these files via the included notebooks.

Contents: data/ (three CSV files, anonymization audit log); code/ (anonymization pipeline; three executable notebooks covering measurement modeling, a robustness battery with bootstrap inference, and a validation battery including cross-validated predictive checks; table generator); results/ (canonical outputs, produced by runs of the included notebooks on the included data); bilingual codebook with verbatim item wordings and all coding maps; SHA-256 checksums.

Data: CC BY 4.0. Code: MIT. An analysis preprint using these data is available at arXiv:2608.18058