Introduction The increasing presence of counterfeit opioid drugs in the United States can contaminate data collection systems and confound estimates derived from surveillance of the opioid epidemic. Data sources and analyses that can quantify the contribution of counterfeit opioid products are needed to provide accurate and timely data to inform public health responses. We describe a novel approach to identify and quantify intentional abuse and misuse exposures involving suspected counterfeit opioid products in United States poison center data. Methods An ecological study was performed using data, including narrative case notes, reported to participating United States Poison Centers of the Researched Abuse, Diversion and Addiction Related Surveillance System between 2009-Quarter 1 and 2021-Quarter 4. A machine learning natural language processing approach was used to develop a predictive model. Results Sensitivity for detecting suspected non-counterfeit-involved exposures by the predictive model was 92%, specificity was 73%, and the area under the receiver operating characteristic curve was 92%. Overall, only 2.1% of intentional abuse and misuse exposure calls were predicted to be suspected counterfeit-involved during 2009–2021; however, we observed an exponential increase in suspected counterfeit exposures over this time period. There was a 7-fold increase in the estimated number of suspected counterfeit exposures from 2009 to 2021, and 23.7% of all opioid analgesic intentional abuse and misuse exposures were suspected counterfeit-involved in 2021. Discussion We demonstrate the feasibility and reliability of using machine learning natural language processing to identify exposures involving suspected counterfeit opioid products in United States poison center data. Results suggest that suspected counterfeits have had a meaningful influence on rates of intentional abuse exposures to opioid analgesics in more recent years. Conclusions The increasing presence of counterfeit opioid drugs can contaminate data collection systems and compromise the reliability of the data.