A deep investigation into how AI fraud, synthetic data and platform gaming have reshaped the consumer insights industry — and what forward-thinking brands are building in its place. The category-defining report for everyone who runs, buys or relies on consumer research.
40%
of panel responses may come from non-human sources
£0
cost to generate a convincing AI survey response in 2026
3×
growth in synthetic respondent deployment since 2023
100%
of CHOOSI respondents verified by voice before participation
The consumer research industry is facing a trust crisis that most of its participants have been slow to name. This report examines how we arrived at a moment where the provenance of research respondents is genuinely uncertain, what the practical consequences are for the brands and organisations that depend on consumer data, and what the architecture of trusted consumer intelligence looks like in 2026 and beyond.
The research panel model was built for a world that no longer exists. In its original form, it was a pragmatic solution to a real problem: how do you access a large, willing sample of consumers quickly and affordably? The answer was incentivised online panels — pools of pre-recruited respondents who agreed to complete surveys in exchange for cash or points rewards.
For two decades, this model worked well enough. It was imperfect — there were always quality concerns — but the scale of fraud was manageable, and the checks designed to catch it were broadly effective. Then three things changed simultaneously: incentive panels became commercially dominant, LLMs made synthetic response generation trivially cheap, and the industry's quality control infrastructure failed to keep pace with either development.
Key findings
Between 20–40% of responses on major online panels are estimated to come from non-human sources, based on independent audit methodologies including honeypot traps, semantic analysis and IP clustering.
Standard quality checks fail against sophisticated AI. Attention traps, minimum time filters and straight-lining detection were designed to catch disengaged humans — not language models that pass them consistently.
The selection bias problem predates AI fraud. Incentivised panels systematically over-recruit professional survey takers whose profiles do not match real consumer populations, regardless of bot presence.
The cost of generating a synthetic respondent approaches zero. As LLM API costs have fallen, the economics of large-scale survey fraud have become viable for a widening range of operators.
The consequence of these converging trends is a research ecosystem in which the baseline assumption — that survey data comes from real people with real opinions — can no longer be taken for granted. Sophisticated research buyers know this. Many have already adjusted their internal standards. The organisations that haven't yet are making decisions on data of unknown quality against competitors who aren't.
Parallel to the bot fraud problem is a growing willingness among some organisations to use AI-generated synthetic data as a stand-in for genuine consumer research. The logic is superficially attractive: synthetic data is fast, cheap, always available and can be generated to any demographic specification. The problem is that it cannot deliver what market research is actually for.
"A language model doesn't know what your specific consumer thinks right now. It knows what consumers have typically said. That gap is the difference between insight and pattern-matching."
Consumer opinion is contextual and volatile. It responds to news cycles, competitor actions, personal circumstances and social pressures that no training dataset can fully capture in real time. An AI generating synthetic consumer responses is drawing on historical patterns of what people have said in similar contexts. It is not accessing current reality. For directional guidance, this limitation may be acceptable. For any research that informs significant investment decisions, it is not.
The structural risk compounds over time. Organisations that use synthetic data to train internal AI systems — recommendation engines, brand health dashboards, segmentation models — are building those systems on fabricated foundations. The feedback loop between synthetic research inputs and AI-driven business outputs is a problem the industry has barely begun to discuss.
What does genuinely trusted consumer intelligence look like? This report argues that it has four components: a verified source, a transparent chain of custody, a design that captures authentic rather than constructed responses, and an analysis framework that can distinguish signal from noise.
Verification happens before data collection begins — not as a post-hoc quality check, but as a prerequisite for participation. The most robust current mechanism is voice-based screening: a live spoken interaction that confirms human presence through signals text-based AI cannot reliably replicate. Voice AI research conducted through platforms like CHOOSI produces voice-screened consumers — a categorically different pool from unverified online panel participants.
Every response in a trusted dataset has a provenance record. How was the respondent recruited? How were they verified? What was their verification score? This chain of custody travels with the data and can be audited. Research without chain of custody is like financial data without an audit trail — it may be accurate, but you have no way to know.
AI-native research platforms are developing survey and interview instruments designed to elicit authentic rather than socially desirable responses. Conversational research, conducted through voice or structured chat, captures responses that are less susceptible to the framing effects of traditional survey design. Voice-based open-ended questions, in particular, produce richer and more authentic data than text boxes.
Verified, authentic data still requires intelligent analysis. AI tools applied to clean, verified datasets produce dramatically better outputs than the same tools applied to contaminated data. The analysis layer is where the investment in data quality pays compound dividends over time.
The transition to verified, AI-native consumer intelligence is not a future possibility — it is happening now. The leading FMCG brands, financial services organisations and technology companies have already begun rebuilding their research infrastructure around verification as a baseline standard. The competitive dynamics of this transition will accelerate: as verified data becomes table stakes for sophisticated buyers, the pressure on research suppliers to demonstrate verification capability will intensify.
For insight teams and research buyers, the practical implication is clear. Procurement standards need to include verification requirements. Research proposals should specify how respondent identity was confirmed. Quality metrics should include provenance documentation alongside traditional data quality indicators.
The brands that make this transition earliest will build research operations that compound in value — not just because their data is better, but because the decisions made on that data will be more often right. In a competitive landscape where every brand has access to the same AI tools, the brands with better data will win. The scarcest resource in consumer intelligence is no longer computation. It's authentic, verified human insight.
CHOOSI verifies every respondent by voice. Talk to us about how AI-native research can transform your insight operation.
Request a demo →
