This paper presents a low-carbon policy intensity dataset for China's manufacturing industries from 2007 to 2022, covering national-, provincial-, and prefecture-level policies. The dataset includes 7282 policies and quantifies policy intensity using a combination of phrase-oriented NLP algorithms and text-based prompt learning. The low-carbon policy intensity index is calculated by multiplying policy level, objective, and instrument. The dataset is organized in two formats (.dta and .xlsx) for multidisciplinary researchers, providing aggregated intensity data at national-, provincial-, and prefecture-levels, as well as sub-intensity for four objectives and three instruments. The dataset has potential for future studies by merging with macro and micro data related to low-carbon performance. The methodology involves constructing a low-carbon policy inventory, structuring policy texts, and disaggregating texts into policy objectives and instruments. Policy texts are classified based on high-frequency phrases and keywords, and policy intensity is quantified using a formula that incorporates policy level, objective, and instrument. Manual labelling is used to train a prompt learning model for predicting policy intensity. The dataset is validated through technical checks and comparisons with existing studies. The dataset is available on figshare and includes detailed information for data files. The study contributes to the literature by deepening the meaning of low-carbon policy intensity, combining phrase-oriented NLP and text-based prompt learning, and aggregating sub-intensity indices for different policy levels, objectives, and instruments. The dataset is useful for future research on low-carbon policies and their impact on carbon neutrality.This paper presents a low-carbon policy intensity dataset for China's manufacturing industries from 2007 to 2022, covering national-, provincial-, and prefecture-level policies. The dataset includes 7282 policies and quantifies policy intensity using a combination of phrase-oriented NLP algorithms and text-based prompt learning. The low-carbon policy intensity index is calculated by multiplying policy level, objective, and instrument. The dataset is organized in two formats (.dta and .xlsx) for multidisciplinary researchers, providing aggregated intensity data at national-, provincial-, and prefecture-levels, as well as sub-intensity for four objectives and three instruments. The dataset has potential for future studies by merging with macro and micro data related to low-carbon performance. The methodology involves constructing a low-carbon policy inventory, structuring policy texts, and disaggregating texts into policy objectives and instruments. Policy texts are classified based on high-frequency phrases and keywords, and policy intensity is quantified using a formula that incorporates policy level, objective, and instrument. Manual labelling is used to train a prompt learning model for predicting policy intensity. The dataset is validated through technical checks and comparisons with existing studies. The dataset is available on figshare and includes detailed information for data files. The study contributes to the literature by deepening the meaning of low-carbon policy intensity, combining phrase-oriented NLP and text-based prompt learning, and aggregating sub-intensity indices for different policy levels, objectives, and instruments. The dataset is useful for future research on low-carbon policies and their impact on carbon neutrality.