Over the past few years, and accelerating sharply into 2025 and 2026, the cost of generating a fraudulent survey response has effectively hit zero. LLM-powered bots can now pass open-ended sentiment checks, maintain persona consistency across a 20-minute survey, and mimic the kind of considered, imperfect responses that used to be the hallmark of a real human being. They do this at scale, around the clock and from anywhere in the world.
The door we left open
The market research industry spent years optimising for low-friction survey experiences: shorter, simpler and faster to complete. But the easier you make it for a respondent to complete a survey, the easier you make it for a bot to do the same. Friction, it turns out, is a feature. The industry removed it, and that became an invitation.
What’s changed isn’t just the sophistication of the fraud, it’s how it enters the system. Consider what a sophisticated bot looks like in your dataset: it doesn’t speed or straight-line; it passes your attention checks, and it gives open-ended responses that are coherent, on-topic, and appropriately varied. A 50-word open-ended response produced in 4.2 seconds with zero backspaces is a tell, but only if you’re collecting the right data to spot it.
It might even introduce the odd typo for authenticity. The only thing it can’t reliably fake is the messy, hesitant, self-correcting behaviour of an actual person thinking through a question in real time.
The standard response to data quality concerns has been post-hoc cleaning: run the data through your algorithms, remove the obvious outliers, flag the speeders and carry on. The problem is that the bots getting through today are not the obvious ones, they’re the ones your cleaning algorithms are leaving in.
If you’re using the same attention checks and consistency questions study after study, you’re essentially publishing the answers. Good fraud detection has to keep moving: regularly rotating and modernising the checks you embed – impossible claims, expert knowledge questions, image interpretation tasks – so that the survey itself becomes harder to game over time.
The shift required is from passive filtering to active defence
This response has to run through the entire process from procurement, design, recruiting to data cleaning and analysis, rather than sitting as a final quality gate at the end.
Data should be treated as suspect until verified, with integrity built into collection rather than retrofitted afterwards. That means behavioural telemetry as standard – keystroke dynamics, response latency, page-visibility transitions. Humans are messy and inconsistent; bots are efficient – this is where some friction can be reintroduced. Identity verification must go beyond the inbox: SSO or LinkedIn credentialing for B2B work, device-level liveness checks for consumer panels. Persistent device fingerprinting for de-duplication across panels, not just within a single study, and throttling device-level participation over time, so the same device can’t quietly cycle through multiple studies across your supply chain.
It also means treating panel quality as an ongoing discipline. Regularly reviewing sources, tracking discard and archive rates by supplier and being willing to cut off the ones that consistently underperform. Bad neighbourhoods in your supply chain don’t clean themselves up. Being able to show clients a clear chain of custody – where each respondent came from, what was removed and why – should be standard at project close, not an afterthought.
The conversation the industry needs to start having
Research agencies need to have honest conversations about what verified data truly costs, and why it’s worth paying for. There also needs to be awareness and transparency around the conditions that produce fraud – aggressive timelines, rock-bottom cost-per-interview targets, an emphasis on sample size over sample quality – and how these in fact make fraud easier.
We also need to push for shared standards across the industry. Individual action matters, but it’s limited when the broader market continues competing on volume.
Bridging the gap between supplier and stakeholder
Ultimately, protecting the truth isn’t just the agency’s job; it’s a shared mission, moving from a transactional hand-off from supplier to client, to a partnership where both sides agree that rigour matters more than just hitting a sample target. By working together to build verification into the survey design and being transparent about the why behind the data, we move away from just cleaning the results to actually securing them. It’s about ensuring that when you walk into that boardroom, the confidence you’re selling is backed by a process you both trust.
Market research exists because organisations need to make better decisions than their own assumptions allow. A confident but wrong insight is more dangerous than no insight at all.
High-quality research now requires a deliberate re-introduction of verified friction, building the infrastructure to make data verifiable from the start.
