How GWI ensures survey data quality at every stage

If you rely on survey data to help make business decisions, chances are you've asked at some point: how do I know these human insights are real?

And it's a question that demands a more rigorous answer than ever. The rushed answers, contradictions and inaccurate responses that have always marked survey fraud remain a live challenge, but AI has introduced a threat of an entirely different order.

Bots don't just complete surveys, they complete them convincingly. They pass attention checks. They vary their response speed to seem human. They can even simulate specific demographics, adjusting tone, spelling, and knowledge level to match a given persona. AI-powered fraud has rapidly become the biggest threat to data integrity in the industry. For buyers making decisions worth millions of dollars, it's not a small problem.

What separates trustworthy data from compromised data isn't a single feature or checkpoint, it's the entire quality control system running before a single response is collected.

At GWI, it's built into every stage of how we collect, structure, and validate the data that our clients use to make business decisions. From how we source respondents to how we catch fraudulent responses, every layer is designed to make sure the data you're working with reflects what real people actually think, feel, and do.

Here's how it works.

It starts with where respondents come from

Not all survey samples are created equal. Some providers rely on what's known as 'river traffic': respondents recruited on the spot from internet ads or app prompts, with no prior verification and no ongoing relationship with the research provider. It's fast and it's cheap, but it's also the most exposed entry point for low-quality or fraudulent responses.

GWI has never used river traffic. Not once in 15+ years of operation. Every respondent who completes a GWI survey comes from a double opted-in, verified research panel that meets recognized industry standards, including ESOMAR and MRS codes. That means the panel knows who they are, has verified their location and email address, and is actively monitoring the quality of their participation over time.

We also don't depend on any single panel. GWI works with over 25 panel partners globally, typically running up to seven in each of our 54 syndicated markets. This multi-panel approach gives us something important: the ability to quickly identify the source of any quality shift, pause a partner if standards slip, and redirect the sample to stronger providers without interrupting fieldwork. A partner only comes back online once we're satisfied with their explanation or remedial action.

Sample design that prevents problems before they start

Good survey data quality doesn't just come from catching bad responses after the fact. A lot of it comes from how you design the research in the first place.

In every market where GWI operates, we maintain extensive original research on population composition, consulting hundreds of government, national statistics, and public body sources each year. This information is then aggregated and analyzed using proprietary modeling techniques to build an up-to-date picture of population size and structure. From this, we build precise demographic quotas (age, gender, education, and country-specific factors like ethnicity and region) that determine exactly who we need to interview. This means our samples are genuinely representative, not just whoever was available.

Fieldwork for GWI Core and GWI USA is continuous. We conduct interviews every day of the year, and respondents declare their demographics at the start of each survey. This creates two important advantages for survey data quality. First, anyone whose declared demographics don't match what the panel or previous surveys have on file is immediately flagged. Second, respondents have no way of knowing which quota groups are open or closed, which introduces a layer of random selection that makes it harder for bad actors (human or synthetic) to game entry.

Research design built for integrity

All of GWI's syndicated studies are designed by expert, tenured researchers with deep experience in constructing neutral, non-leading questions. A respondent should never be able to tell what hypothesis is being tested.

We maximize randomization throughout. Answer scales flip between positive and negative ends. Question blocks are shuffled so the order of themes varies across respondents. Instruction text and answer options are worded to avoid leading the respondent in any direction. They work as structural protections that reduce bias and make it significantly harder for automated systems to produce plausible-looking responses.

A quality control stack that works on two fronts

Once responses start coming in, GWI and its panel partners run a combination of researcher-led and technology-led checks. Neither is sufficient on its own, which is why they operate in parallel.

On the researcher side, responses are scrutinized for behavioral signals that indicate inattentive, inconsistent, or implausible answering. This includes checks on completion speed (both across the full survey and within sub-sections), consistency against answers given in previous studies, and detection of pattern responses. On the technology side, respondents pass through verification systems that assess identity consistency, device signals, and access patterns to detect misrepresentation or duplication. This includes checks for VPN usage, proxy access, and duplicate identities, as well as real-time monitoring of behavioral signals during the survey itself.

These layers work together. Every respondent receives a composite quality score that aggregates behavior across all checks. Cross a specific threshold and you're automatically removed. Fall below it and your responses go through additional human review.

A proprietary advantage built over 15+ years

GWI has been collecting data about people's attitudes and behaviors for over 15 years, conducting over 5 million interviews annually across 54 countries. That depth and consistency of historical data gives us something that newer or less established providers simply don't have: a rich, proprietary understanding of how real people respond to survey questions, including the kinds of questions where there's no right or wrong answer.

This matters because human responses are complex and contradictory in ways that synthetic responses consistently fail to reproduce. We use this historical foundation as an active detection layer, comparing incoming responses against established patterns to identify anomalies that other methods might miss.

We also deploy our recontact methodology across syndicated and bespoke studies. This means re-interviewing individuals who've previously completed GWI Core or GWI USA. They've already passed an intensive set of quality checks, and the additional data they provide gives us another opportunity to verify consistency against their previously declared responses.

Transparent about what we remove

No research vendor should claim to have eliminated fraud entirely. What they can do is show you the work.

Across the last four published quarters of GWI Core, our global removal rates held between 8.2% and 9.3%. With GWI interviewing approximately 250,000 people per quarter for Core, that represents around 22,500 removals per wave. These rates vary by market (a recognized industry-wide pattern), but they stay consistent at the global level because our pre-fieldwork checks stop a large proportion of bad actors before they ever reach the survey.

We also continue to invest in new detection methods. Survey data quality is not a problem you solve once. It's one you commit to solving continuously.

FAQs on survey data quality

How many panel partners does GWI work with?

Over 25 globally, typically three to seven per market across our 54 syndicated countries. Every partner must recruit double opted-in, verified users and pass GWI's stringent onboarding checks before supplying any sample. Partners are monitored continuously and paused if quality drops.

What does "double opted-in" mean?

It means the respondent has actively signed up to participate in research (first opt-in) and then confirmed their identity and details, usually via email verification (second opt-in). This is distinct from river traffic, where respondents are recruited anonymously in the moment with no prior verification.

How does GWI's recontact methodology improve data quality?

Recontact means re-interviewing people who've already completed GWI Core or GWI USA. Because they've already passed a full set of quality checks, the risk of fraudulent responses is lower from the start. Their new answers can also be checked against what they've told us before, giving us an additional layer of verification.

Why do removal rates vary by market?

This is a recognized pattern across the market research industry. Respondent behavior, panel quality, and the prevalence of fraudulent activity differ from country to country. GWI sets scoring thresholds bespoke to each market and demographic to account for this variation.

Does GWI use AI in its own quality control?

Yes. Technology-led checks are used alongside researcher-led review to detect synthetic or fraudulent responding patterns. These methods draw on GWI's proprietary historical data as well as real-time behavioral signals during survey completion.