A three-month rolling average of 59% of records removed before analysis. That's not a typo, it's where a lot of sample providers are at right now, according to

At the same time, GRIT's practice report found data quality concerns up 40% year over year, with synthetic respondents and Gen Z survey fatigue named as the main drivers. The fraud got better at its job too: fluent, on-topic, polite. The lazy tells (speeding, gibberish, blank fields) that used to flag a bad respondent don't work the same way anymore.

So here's the real question for anyone running studies right now: where's your line?

• Do you allow any synthetic or AI-assisted responses into a study at all, even for pretesting?
• If yes, how do you validate it before it touches a real deliverable?
• Has this changed your sampling budget, more validation steps, smaller samples, extra vendor checks?

We hear wildly different answers to this depending on who you ask, agency, client-side team or independent consultant. Some won't touch synthetic data for anything that reaches a client deck. Others blend it in quietly and never mention it.

How do you handle this today, and would you tell a client if synthetic data was part of their results? If you're building the survey itself, this is exactly the kind of validation logic a tool like Helix is built to help catch early, before bad data ever reaches analysis.

Sources

• Rival Technologies: 2025 GRIT Insights Practice Report: Notable Trends for Brand-Side Researchers (2025)
• GroupSolver: Synthetic respondents vs. real people: What still counts as evidence in market research (2026)

#MarketResearch #DataQuality #SurveyDesign #MarketIntelligence