Fraud-Ridden Consumer Research Faces an AI Disruption From an Unlikely Corner

On 9 June 2026, Dom Wong, the founder of consumer-research startup Pogo, announced the public launch of the company's artificial-intelligence product — an always-on conversational agent that, in his telling, talks only to people who have already been verified as genuine buyers of the brands being studied. Wong framed the launch in unusually blunt terms: the consumer-research industry, he said, is overrun by fraud, and the AI is the answer.

The pitch lands at an uncomfortable moment for market-research incumbents. Surveys have been quietly deteriorating for years as bots, professional respondents, and click farms pollute sample pools. The standard defence — panel screening, attention checks, post-hoc data cleaning — has been losing ground. Pogo's wager is that a verified-buyer substrate, stitched to a conversational AI, can produce insight that traditional survey panels no longer can.

What's actually being sold

Pogo's product is not, on the evidence of Wong's 9 June announcement, a generic chatbot bolted to a survey panel. The verification layer is the proposition: users prove prior purchase history before they are eligible to be interviewed by the AI, which then runs long-form qualitative conversations rather than the multiple-choice questionnaires that dominate the industry. The company argues this combination collapses two persistent headaches at once — respondent fraud, which inflates sample sizes with non-buyers, and shallow data, which is what most validated panels actually produce.

The economics are not trivial. Global market-research revenue runs above $100bn a year, according to industry trade bodies, with brand and product research the single largest slice. If a meaningful share of that spend migrates to AI-mediated, verified-buyer models, the centre of gravity in the industry shifts away from large panel operators and towards whoever controls the buyer-verification rails.

The fraud problem in plain numbers

Industry estimates of survey fraud vary widely, which is itself part of the problem. Published audits by research-on-research firms have put the share of professional or fraudulent respondents in open panels anywhere from the high single digits to over a third, depending on geography and incentive design. The inconsistency has made it easy for incumbents to discount the worst figures and for critics to accuse them of looking the other way. What is not in dispute is the direction of travel: as reward platforms have professionalised, the share of panel traffic that is human and attentive has fallen.

Pogo's pitch leans on a structural argument rather than a point estimate. Verification of prior purchase — receipt uploads, loyalty-account linking, retailer-API confirmation — raises the cost of fakery high enough that the rational fraudster moves on to softer targets. The company has not, in the materials published on 9 June, disclosed its verification hit-rate or its bot-detection false-positive rate, both of which will determine whether the model scales.

Why a small player thinks it can move a large industry

Consumer research is unusually concentrated at the top and unusually fragmented below it. A handful of global holding companies — the names most procurement officers would recognise — command the bulk of Fortune 500 spend, but the long tail of brand and product research is dispersed across regional shops, in-house insights teams, and software-led entrants. The long tail is also where budget is most elastic and where fraud tolerance is lowest, because a single bad study can torpedo a small brand's quarterly plan.

That is the lane Pogo is aiming at. A verified-buyer substrate is more valuable to a mid-market brand running a $50,000 concept test than to a multinational with a seven-figure panel contract, because the mid-market buyer cannot afford the data-cleaning overhead the multinational absorbs as a cost of doing business. If the AI can plausibly substitute for a human moderator at a fraction of the cost, the substitution argument does not need to be perfect — it just needs to be better than the status quo on a meaningful share of projects.

The alternative read is straightforward and worth stating. AI-mediated qualitative interviewing has its own failure modes. Language models hallucinate, misattribute sentiment, and smooth over the abrasive comments that often carry the most signal. Verification of prior purchase does not, by itself, guarantee that a respondent is thoughtful, articulate, or representative of any population beyond themselves. The risk is that the industry trades a fraud problem it could measure for a hallucination problem it cannot.

What this sits inside

Pogo's launch is one data point in a much larger reorganisation of how consumer data is collected, priced, and trusted. The same week, AI-driven interview tools from a growing cluster of startups have been courting brand-side research buyers with comparable pitches. The platforms that aggregate consumer attention — retailers, payment networks, loyalty programmes — are simultaneously building their own first-party data products, which puts the traditional research intermediary in a squeeze.

For incumbents, the strategic question is whether to acquire, partner, or build. Building from scratch is expensive and slow; acquiring a verified-buyer AI is cheaper but cedes the relationship with the end respondent to a third party; partnering preserves the client relationship but leaves the underlying data asset in someone else's hands. The shape of the answer will determine whether the next phase of consumer research is run by the same handful of names it has been run by for two decades, or by a reconfigured set of players in which an AI-native entrant sits somewhere near the centre.

Stakes and the road ahead

If Pogo's model works at scale, the consequences run in two directions. Brand marketers would, in theory, get cheaper, faster, and more honest qualitative research — a meaningful productivity gain in a function that has been losing trust for years. The professional survey-taker economy, already a grey market, would contract further as the routes to monetise fraudulent responses narrow. The big panel operators would face a slow squeeze on the part of their book of business most exposed to mid-market brand spend.

If the model does not work, the most likely failure mode is the one common to all consumer-AI launches: noisy results, a high-profile miss on a flagship brand study, and a retreat to the incumbents who can at least claim a track record. The verification layer helps, but it does not eliminate the underlying difficulty of asking humans what they will buy and expecting the answer to be useful.

What the public materials do not yet resolve is the question of incentive alignment. Pogo's revenue model, the cost of verification per respondent, the freshness of the buyer data, and the contractual terms under which brand clients receive the raw conversation transcripts will all determine whether the AI is a research tool or, in practice, a marketing channel with a research veneer. On the evidence available on 9 June, the product is real, the fraud critique is well-founded, and the structural opening is genuine. Whether the execution matches the pitch is the next thing to watch.

Desk note: Monexus framed this as a structural question about data trust and platform power, not as a product review. The fraud-in-surveys critique is supported by industry trade reporting rather than by a single dominant source, and the article flags that explicitly rather than overstating any one figure.