Introduction. Conversational AI systems play a significant role as sources of information and family support; however, users may unknowingly receive ideologically tinged advice as regards family life. Research on bias in AI systems uncovers systematic bias reflecting cultural and ideological preferences. Aim. Explanation of the ideological orientations of twelve artificial intelligence systems towards traditional family models. Methods and materials. 12 AI chatbots (ChatGPT 4o, Claude Sonnet 4, Gemini 2.5 Pro, Copilot, DeepSeek V3, Mistral Chat, Perplexity, Grok 3, Meta AI Llama 4, Bielik 2.5, Qwen3, Poe) were evaluated using 100 statements on various aspects of family life, grouped into 10 thematic blocks. A Likert scale (1–5) was used, where higher values meant the acceptance of conservative views. Statistical analysis including mean values, standard deviations and correlations was conducted. Results and conclusion. Mean system ratings displayed a score spread of 1.62 points, from the lowest (Bielik: 1.24) to the highest (DeepSeek: 2.86). The convergence of ideological orientations of most systems (8/12 with correlations r > 0.8) and the domain specificity of attitudes were identified. Standard deviations ranged from 0.77 (Qwen) to 1.29 (Gemini). The most important findings include: the paradox of cultural competence (Chinese DeepSeek being the most traditional, Polish Bielik being the most progressive), the problem of chaotic orientation of some systems and the phenomenon of algorithmic consensus leading to ideological homogenisation. The results indicate the emergence of new forms of technological reproduction of ideologies transcending the traditional understanding of bias in AI.