Integrating artificial intelligence into back-translation: Opportunities, constraints, and recommendations for psychological research.

Brislin's (1970, 1986) back-translation procedure is widely used to translate psychological scales, but it depends heavily on researchers' judgments and is often applied uncritically. Across five studies, we examined whether translation tools based on artificial intelligence (AI) can support this process. Two pilot studies compared ChatGPT, Google Translate, and DeepL and identified DeepL as the most suitable tool for Studies 1-3 based on performance and practicality. Studies 1-3 examined English-German scale translations produced through a single back-translation round. AI-generated translations were comparable and, for some scales, superior to human translations in linguistic quality (Study 1) and original-back-translation correspondence (Study 2). The AI- and human-translated scale versions showed broadly similar patterns of measurement invariance (Study 3). Studies 4 and 5 extended the comparison to English-Spanish translations and a multiround design. Study 4 compared a human multiround procedure with an enhanced AI-assisted procedure combining parallel AI-generated translations, back-translation, and structured quantitative and qualitative evaluation with elements of the translation, review, adjudication, pretesting, and documentation approach; Study 5 examined measurement invariance of the resulting versions. Across studies, AI-based tools generally performed comparably to the human procedures examined, supporting their use in psychological scale translation. However, qualified human judgment remains important, and linguistic similarity does not provide evidence of cultural or psychometric equivalence. We provide recommendations for AI-assisted scale translation and introduce the AI Translation Evaluation Assistant for structured qualitative review, used alongside the metric for evaluation of translation with explicit ordering as a quantitative correspondence indicator. (PsycInfo Database Record (c) 2026 APA, all rights reserved).