This article is part of weekly updates on new developments in the use of AI methods and tools of surveys (households, individuals, farms…) and administrative data for official statistics

Coverage Period: 27–02 November 2025

Key words: AI, survey research, official statistics, machine learning, data quality, household surveys, statistical methods, data analysis

Executive Summary

This week’s update on the integration of Artificial Intelligence (AI) in survey research highlights significant advancements across the entire survey lifecycle, from data collection and processing to analysis and dissemination. Key developments include the widespread adoption of AI-powered tools for qualitative data analysis, the implementation of AI by national statistical offices for core statistical processes, and a burgeoning, yet critical, debate surrounding the use of synthetic respondents. For researchers and statistical offices, these trends signal a paradigm shift towards more efficient, accurate, and timely data processing, while also presenting new methodological challenges and considerations.

This report synthesizes recent articles, industry reports, and academic papers to provide a comprehensive overview of the state of AI in survey research. We will explore the practical applications of AI in data editing and cleaning, the tangible benefits observed by early adopters like the UK’s Office for National Statistics (ONS), the capabilities of commercially available AI survey platforms, and the critical academic discourse on the future of AI-generated survey data.

The most significant recent trend is the rapid maturation of AI tools designed to automate and enhance the analysis of qualitative and open-ended survey data. Traditionally a time-consuming and manual process, AI is now enabling researchers to extract insights from large volumes of text in a fraction of the time. A recent industry analysis indicates that AI adoption in business functions, including marketing and customer experience, has surged from 50% to 72% in the past year alone, fueling the development of these sophisticated tools [4].

Commercially Available AI Survey Tools

A growing number of commercial platforms now offer advanced AI capabilities for survey research. These tools provide a range of features, from AI-assisted survey creation to real-time feedback intelligence. The table below summarizes some of the leading platforms and their key AI features, as identified in a recent market overview [4].

These platforms are democratizing access to advanced analytical techniques that were previously the domain of data scientists, enabling a wider range of researchers and organizations to leverage AI in their work.

Applications in National Statistical Offices

National statistical offices (NSOs) are beginning to integrate AI into their core processes to improve efficiency and accuracy. A notable example is the UK’s Office for National Statistics (ONS), which has implemented an AI tool called ClassifAI for occupation coding in its Annual Survey of Hours and Earnings (ASHE) [3]. This marks the first direct application of AI into a statistical process at the ONS.

“This is the first time that the ONS has applied AI directly into a statistical process and, in doing so, we have improved the accuracy of occupation coding while saving hundreds of hours of work, which we were able to invest into helping achieve a quality dataset.” [3]

The ASHE survey is one of the largest run by the ONS, with approximately 174,000 employee returns. The successful deployment of ClassifAI has not only improved the accuracy and efficiency of this specific survey but is also paving the way for the tool’s deployment across other ONS surveys. This implementation is part of a broader strategy at the ONS to modernize its economic statistics and pivot data science resources to improve core statistical outputs.

Synthetic Data and Respondents: A Critical Perspective

A more nascent and controversial development is the use of Large Language Models (LLMs) to generate synthetic respondents. The concept involves creating artificial data that mimics the statistical characteristics of real survey data, potentially offering a faster and cheaper alternative to traditional data collection. However, recent empirical evidence suggests that this approach has significant limitations.

At the European Survey Research Association (ESRA) 2025 Congress, a study was presented that tested the ability of three different LLMs to predict the results of the 2024 European elections. The results were described as “disastrous,” with the models predicting an electoral participation rate of 83%, compared to the actual turnout of 49% [2].

The study’s authors highlighted several factors contributing to this failure, including biases in training data and the inherent complexity of social dynamics. While synthetic data may have some utility for descriptive use cases (estimating already existing behaviors), its predictive power for unobserved behaviors or opinions appears to be severely limited. One expert in the field has likened the use of synthetic respondents to “the homeopathy of market research,” stating, “there’s no evidence that they work, but many people still believe in them” [2]. For statistical offices and researchers requiring high levels of accuracy and reliability, this suggests that caution is warranted when considering the use of synthetic respondents.

Academic Research and Methodological Advances

The practical applications of AI in survey research are underpinned by a growing body of academic work. Recent scholarly articles have focused on several key areas:

Automated Survey Coding: Foundational research on automated occupation and industry coding, some dating back over a decade from institutions like the US Census Bureau, has laid the groundwork for the tools now being implemented by NSOs like the ONS [6].

Machine Learning for Data Imputation: A significant body of research is dedicated to using machine learning for missing data imputation. Systematic reviews and comparative studies have shown that deep learning approaches can outperform conventional statistical methods, providing more robust solutions for handling incomplete datasets [6].

This academic work is crucial for validating the use of AI in statistical applications and for developing the next generation of survey methodology.

Conclusion

The developments of the past week underscore a period of rapid transformation in the field of survey research. AI is no longer a futuristic concept but a practical tool that is actively being deployed to enhance data processing, analysis, and reporting. While the potential for efficiency and deeper insights is immense, it is crucial for researchers and statistical offices to remain critical and discerning consumers of these new technologies. The case of synthetic respondents serves as a salient reminder of the importance of empirical validation and methodological rigor.

Moving forward, the successful integration of AI into survey research will require a dual focus on both technological adoption and capacity building within organizations. As the tools and techniques continue to evolve, so too must the skills and expertise of the researchers who use them.

References

[1] Blix.ai. (2025, October 22). How AI Survey Text Analysis Can Save Hours of Manual Work. https://blix.ai/blog/ai-survey-text-analysis

[2] Ochoa, C. (2025, October 21). Synthetic respondents and the future of survey research. Quirk’s Media. https://www.quirks.com/articles/synthetic-respondents-and-the-future-of-survey-research

[3] McKeown, L. (2025, October 23). How the ONS is improving its annual earnings survey. Office for National Statistics / LinkedIn. https://www.linkedin.com/pulse/how-ons-improving-its-annual-earnings-survey-sdz7e

[4] Sharma, S. (2025, October 23). Top 15 AI Survey Tools in 2025 for Smart Feedback Intelligence. Zonka Feedback. https://www.zonkafeedback.com/blog/ai-survey-tools

[5] Sopact. (2025, October 25). How to Automate Qualitative Analysis Using AI. https://www.sopact.com/use-case/qualitative-analysis

[6] Various Academic Sources. (2012-2025). Compiled from search results on AI in official statistics, survey coding, and data imputation.

Contact: bakodramane@gmail.com