The survey data quality crisis is the collapse in reliability of online panel research caused by bots, fraudulent participants and professional respondents. Kantar estimates 30 to 40 percent of online survey responses are compromised, and peer-reviewed research shows LLM-powered bots pass standard quality checks 99.8 percent of the time.
The numbers
Kantar, one of the largest research companies in the world, estimates that between 30 and 40 percent of online survey responses are compromised. Industry analysis finds that roughly 3 percent of devices complete 19 percent of all online surveys, which means a small population of professional respondents is answering a large share of the questions that corporate decisions are built on. Research published in the Proceedings of the National Academy of Sciences found that bots powered by large language models pass the standard attention and consistency checks used to screen survey data 99.8 percent of the time.
GreenBook's GRIT 2025 report, based on surveys of research buyers and suppliers, records that data quality is now the single most cited challenge in the industry, ranked first by 40 percent of respondents, and that concern rose 40 percent in a single year.
Why it is worse than it looks
A concept test where a third of the appeal scores are noise does not produce an obviously wrong result. It produces a confident, precise, wrong result. A pricing study where the willingness-to-pay curve is shaped by bots will be presented with the same decimal places as a clean one. There is no downstream analysis that can recover signal that was never collected. The error is invisible in the output and only shows up in the market, months later, as a product that did not sell or a price that did not hold.
What is driving it
Three things. The economics of online panels reward volume over verification, so incentives attract professionals and bots. The tools built to screen respondents were designed for careless humans, not for language models that can produce plausible open-ended text at scale. And the industry's shift to self-service survey software, which grew 11.5 percent in 2024 according to ESOMAR, has put questionnaire design and sample sourcing in the hands of teams who have no way to audit where their respondents come from.
What works
GRIT 2025 is clear about the direction: tech-enabled full-service research, where a human team designs, executes and interprets the study using modern tools, is "making a comeback", and buyers are moving toward known participants and insight communities. Inex One's analysis of the expert network market notes that researchers are shifting toward qualified B2B respondents recruited through networks precisely because consumer online panels have become unreliable.
The fix is not a better bot detector. It is knowing who you are talking to. That means recruiting named people, confirming their identity and role, recording the conversation, and being able to produce the transcript. It is slower and more expensive per respondent, and it is the only method that produces evidence you can put in front of a board.
What this means for a buyer
- Ask any research supplier what share of their sample is identity-verified. If they cannot answer, assume the Kantar range applies.
- For any decision that is expensive to reverse, insist on named, verified sources, even if the sample is smaller.
- Treat quantitative precision from an unverified panel as a rough indicator, not a measurement.
Theory Intelligence works only with verified human sources and delivers a provenance appendix with every study. Read how the method works, or see the services and prices.
Sources
- User Intuition, The research data quality crisis, February 2026 (citing Kantar, Case4Quality, PNAS)
- GreenBook, GRIT Insights Practice Report 2025
- ResearchWorld, Inside the $153bn insights industry (ESOMAR GMR 2025)
- Inex One, Expert network market size, February 2026