Abstract: Objective: To investigate the effects of different thyroid-shielding strategies on radiation dose, artificial intelligence (AI)-based lung nodule detection accuracy, and image quality in chest CT using automatic modulation technology. Methods: An experimental model was constructed using a Lungman PH-1 chest phantom, a CIRS 711-HN head phantom, and two custom-made elliptical phantoms. Three thyroid-shielding protocols were established: Group 1 (no shielding), Group 2 (shielding before scout scan), and Group 3 (shielding after scout scan). Ten chest CT scans were performed for each group. Automatic tube voltage and tube current modulation technologies were utilized, and scanning parameters and total radiation doses were recorded. Doses to the lens of the eye, thyroid, breast, and abdominal organs were measured using thermoluminescent dosimeters (TLDs). An AI-assisted diagnosis system was used to evaluate the detection accuracy of nine types of characteristic nodules (CT values: 100/−630/−800 Hu; diameters: 5/8/10 mm). Subjective and objective image quality evaluations were performed using the contrast-to-noise ratio (CNR) and a 5-point Likert scale. Results: (1) Thyroid shielding before the scout scan (Group 2) resulted in the highest CTDI_vol 5.054 ± 0.276 mGy) and DLP (186.960 ± 10.991 mGy·cm), showing significant statistical differences compared to Groups 1 and 3 (all P < 0.001). (2) Thyroid shielding after the scout scan (Group 3) resulted in the lowest surface doses to the thyroid, lens, breast, and abdomen. Specifically, the thyroid surface dose (3.009 ± 0.626 mGy) was reduced by approximately 73% compared to Group 1 (11.327 ± 1.375 mGy). Surface doses for the thyroid and abdomen were significantly higher than doses measured outside the shielding (P < 0.001). (3) Abnormalities in AI-based nodule recognition were limited to size deviations; notably, −800 Hu/5 mm nodules were not detected in any group. Significant differences in recognition were only observed for −630 Hu/5 mm nodules among the three groups, which disappeared after correction (all P > 0.017). (4) There were no significant differences in CNR across all layers among the three shielding protocols (P > 0.05). (5) Subjective scores showed high consistency (Kappa = 0.706). Significant differences in subjective scores existed among the three groups (P < 0.05), with Group 1 achieving the highest score and showing a significant difference compared to Group 2 (P = 0.005). Conclusion: Thyroid shielding in chest CT can effectively reduce external scatter radiation. However, shielding before the scout scan increases CTDI_vol, while shielding after the scout scan has a slight impact on image quality and AI recognition. Therefore, clinical implementation requires careful selection.