Newsletters

Best Practices for Using Generative AI for Survey Research

11/25/2024

Best Practices for Using Generative AI for Survey Research

Sarah Kelley and Claire Kelley, Child Trends

Generative AI – which can be broadly defined as computer technology that creates content such as text, images or video – surged into public consciousness with the release of ChatGPT in late 2022. Since then, generative AI has been widely adopted in the survey research field, across both industry and academic applications. Survey researchers have used generative AI across all stages of the survey research pipeline including literature review, questionnaire development, interactive chatbot facilitated interviewing, analysis of open ended questions, and support for scientific writing.

Researchers using AI can choose from a wide range of AI models, tools and platforms, many of them designed to allow the use of AI models without the need to write code. And AI tools are increasingly embedded in everyday technologies including browsers, email, and word processing software. As this technology becomes increasingly integrated in the research process, it is essential to consider how to ensure that AI-enabled research meets standards for high quality research. Some considerations on best practices for use of AI in research may include:

Transparency: When conducting research using AI, the use of AI should be disclosed in all reporting. This should include details on the model used, the tasks AI tools were used for, and the procedures undertaken to verify the results.

Efficacy: Choosing the most appropriate AI tool for the task and following best practices for interacting with generative AI models can dramatically improve research output. Selection of the right tool requires comparing model effectiveness for specific tasks (which is often available as published benchmarks) and balancing considerations such as cost, speed and ease of use. Once an AI tool is selected, considerations for improving the results may include prompt engineering, few-shots learning and/or fine-tuning, depending on the application and accuracy required. Simpler options such as prompt engineering should typically be used before more complex and resource intensive options such as model fine-tuning.

Validation: Ultimately the human authors of research are responsible for the validity of results. When using AI tools to produce research, researchers must clearly identify how the accuracy of AI produced conclusions, content or findings will be assessed, and report this validation process along with their results. Validation may include informal processes such as manually editing content written by AI or formal processes such as calculating inter-rater reliability between AI coding and human coding of open-ended questions.

Ethics and Risk: As survey researchers we have ethical obligations to research subjects. When using AI to analyze participant data we must ensure that the data remains secure and that participants are appropriately informed of the potential uses of their data. Researchers should consider if and how to describe the use of AI tools in their survey consent procedures. Researchers should remain aware of the risk of disclosure of sensitive or personally identifying information.

When these best practices are taken into account, generative AI can be a powerful tool to support survey research. Rather than avoiding AI, survey researchers should consciously and reflectively engage this powerful new tool to support their research in the AI era.