Speaking in the webinar, Jo Bygrave, director of data quality at NielsenIQ and member of the MRS data quality working group, said that clients can “have huge influence over survey quality”, such as in areas including procurement, methodology, timelines and validation expectations.

One area where clients could have impact was in supplier selection. “Too often quality only becomes part of the conversation later if an issue has arisen, but really, quality expectations should be built into procurement discussions right from the very beginning of your project or working period with someone,” Bygrave said.

“Clients should understand how participants are recruited, how quality will be monitored and what validation checks will be done, as well as what will happen if things go wrong and your study might need to course correct. These quality conversations should be had up front and not after problems have occurred.”

Sample sizes could also make a huge difference to data quality, said Bygrave, and clients need to be prepared to have detailed discussions about the subject to make sure they get the best possible data from the project.

“It is really important to have an honest conversation at the beginning about the trade-offs between quality, representation, speed, cost and what data or sample size is actually needed to answer your research objectives,” she added.

“Too large a sample size and your study will become really difficult; too small, and you might not be able to cut the data in ways that you need to.”

Bygrave said that well-designed surveys encourage good-quality data and responses from genuine participants. “The better your survey, the better the participant experience and, generally, the better your data will be,” she said.

“If you design the research well, you are giving respondents the best chance to give you the data you need.”

Clients should ask about validation checks, Bygrave said, and ask for sample sources, quotas and weighting of the data, among other areas. Researchers should aim to use as many signals as possible to understand a “complete picture of that respondent”, rather than a few indicators that might not give the full story.

Also speaking in the webinar, Chris Stevens, consultant and member of the MRS data quality working group, said that some valid participants have been misidentified as being fraudulent.

For example, Stevens said that straight lining responses – where someone picks the same option for a series of questions in a survey – was not always a sign that the respondent is fraudulent or giving poor quality responses.

“Straight lining response can be a valid response,” he said. “Practitioners need to stop flagging people based on straight lining alone. Additional checks and validation need to be considered.”

He added: “When you are checking data, there is always a level of subjectivity; it is not necessarily a black and white decision.”

Stevens said that additional data points or context could help identify whether someone is a genuine participant answering questions, or whether there are data quality issues at play.

“There will be some people who put gibberish for whatever reason, but that doesn’t mean they’re a poor participant,” Stevens added. “You should make sure you are being inclusive of how people can behave when they see open-ended questions.

“The main takeaway from this is that one flag does not make a poor participant. You need to be able to build a rounded picture of that data from the participant and have evidence-based checks in place.”

He concluded that over-removal can be problematic for getting a representative sample. “There are some good participants out there who are being flagged. We can’t afford that, as the participant pool is getting smaller.”

Debrah Harding, managing director at MRS, added: “Disengagement, technical fraud, bot activity and AI-generated responses all create challenges for survey quality. But a single flag is not the whole story.

“Overly aggressive removal can exclude genuine participants, reduce representativeness and introduce bias.”

Greg Matheson, co-chief executive at Quest Mindshare and chair of the Canadian Research Insights Council (CRIC) Online Research Committee, said AI was having a profound effect on data quality issues.

“It is kind of guerrilla warfare,” he said. “It is not necessarily the sophistication of AI – it is the sophistication of the bad actors getting better in how they use technology and how they use AI.”

Matheson said that machine versus machine is “the only one way to go about it” in tackling AI-led data quality issues: “The bad actors in this space are looking for quick hits, quick money, a quick way to have transactions. Their sophistication has increased, so it is worth thinking about how they are using AI and how we get in front of that.”

He added: “They are becoming better at prompting and using AI, and using AI to circumvent different things you are doing to affect datasets.”

John Tabone, chief executive at CRIC, added that surveys should be short and engaging by design, and said that “the emphasis on participant experience is critical, and is a key thing”.

Clients need to ask for information on areas such as panel management and sample quality, Tabone said, as it will encourage researchers to “have good answers” in anticipation, and therefore support data quality. 

The webinar followed the release of an updated  ISO 20252:2026 international standard for market, opinion and social research last month that included new data quality guidance, including on participant experience.