%202.png)
Getting clean, reliable data is the whole point of running a survey. But what happens when results are contaminated by bots, professional fraudsters, or simply inattentive participants? Poor data quality leads to flawed insight and costly decisions. The most effective fix is layering several defences at once: controlling who can access the survey, embedding in-survey attention checks, and using technical tools like CAPTCHA and digital fingerprinting.
No single check catches everyone. The researchers who keep their data clean treat fraud prevention as a stack: control who gets in, watch behaviour once they're in, and let technology quietly flag what a human moderator would miss.
Foundational strategies: controlling who takes your survey
The best way to get quality data is to start with quality respondents. The first line of defence is how you distribute the survey and who you invite.
Avoid open posting on social media
Sharing a public survey link on social media or open forums is an open invitation to bots and organised survey farms. A Science Friday researcher learned this directly: a publicly shared link attracted over 6,800 responses, many clearly AI-generated, including answers that opened with "As an AI language model, I do not have personal preferences." Send unique survey links directly to a pre-screened list instead of an open link, so one person gets exactly one link.
Use random sampling from a verified panel
A panel is a pre-vetted group of people who have agreed to participate in research, with reputable providers continuously monitoring members and removing bad actors. Pew Research Center's work on this is instructive: a rigorously recruited panel produced a bogus-respondent rate of roughly 1%, while an openly recruited online poll ran in the range of 4% to 7% bogus. Pew's more recent research flags that this gap is widening as AI-generated responses get easier to produce at scale, so opt-in polling is under growing pressure even where older benchmarks looked manageable. Not sure what sample size you need to start from? Use Yazi's sample size calculator. Platforms like Yazi also give access to a quality-controlled panel across Africa, making it easier to reach verified, engaged participants in emerging markets. See the Yazi research audience.
Implement stringent screener questions
Screener questions confirm you're surveying the right people, but fraudsters routinely lie on screeners to qualify for incentives. Design behavioural questions instead of self-report ones (tested examples in Yazi's survey question bank). Instead of "do you use Slack at a large company?", ask a specific question about a workflow only a genuine user would recognise. That kind of screener acts as a filter that weeds out imposters before they reach the main survey.
In-survey techniques: catching bad actors in the act
Once a respondent is inside the survey, you need ways to monitor engagement and honesty in real time.
Embed attention checks
An attention check, or trap question, is a simple instruction embedded in a question to see if someone is actually reading, such as "for this question, please select 'Strongly Disagree'." In one academic study, about 7% of participants failed an instructional prompt. But don't rely on these alone: Pew found that 84% of known bogus respondents still passed a simple attention check, meaning it only catches the least sophisticated cheaters. Useful as one layer, not a complete solution on its own.
Use cross-verification and duplicate questions
Ask for the same information in different ways at different points in the survey. Asking "do you have any children?" near the start and "what are the ages of your children?" near the end catches an obvious inconsistency if the two answers don't line up. A repeated question, like income bracket on page 2 and again on page 8, works the same way: an attentive, honest person gives a consistent answer. Use this sparingly, since too many repeated questions cause fatigue and can make even honest respondents give less thoughtful answers.
Include strategic open-ended questions
Open-ended text questions aren't just for rich qualitative insight, they're also a strong fraud signal. Bots and lazy respondents give themselves away here with gibberish, pasted irrelevant text, or nonsensical answers. In the Science Friday case, open-ended answers were the giveaway that revealed AI-generated text, and Pew researchers have identified duplicate takers by finding pairs of "different" respondents whose long-form answers were nearly identical, a pattern that's highly unlikely by chance. Yazi's WhatsApp platform makes collecting open-ended feedback easy and harder to fake: voice notes, photos, or videos are far harder for a basic bot to fabricate than typed text.
Monitor response patterns and timestamps
Completion time is one of the most useful metadata signals. Someone finishing in a fraction of the median completion time for that survey is a "speeder" who almost certainly wasn't reading the questions; a common rule of thumb flags anyone finishing in under one third of the median completion time. Watch too for "straight-lining", where a respondent selects the same option for every row in a grid question, a clear sign of a disengaged participant clicking through to the end.
Technical and platform-level defences
Modern survey platforms include built-in tools that fight fraud automatically, adding an invisible layer on top of everything above. For more on how Yazi approaches this, read the data security executive summary.
| Defence | What it does | What research shows |
|---|---|---|
| Platform fraud scoring | Scores device, location, and behaviour signals | Flagging rates vary widely by study and threshold, commonly landing anywhere from roughly 30% to 45% of responses as high risk |
| CAPTCHA | Filters automated bot traffic at entry | One large study saw 10.1% of enrollment attempts barred for failing reCAPTCHA screening |
| Digital fingerprinting | Flags one device completing a survey multiple times | One analysis identified 13.1% of completed responses as duplicates |
| IP tracking & geolocation | Blocks submissions from outside the target region | Effective against out-of-country click farms; pair with self-reported location for cross-checks |
On platform fraud scoring specifically, be cautious about treating any single published percentage as universal. Studies comparing fraud-detection systems on the same dataset have found flagging rates ranging from roughly 40% to 45% depending on which system and threshold was used, underscoring that these tools need tuning and validation for each study rather than a single benchmark applied blindly.
Post-survey and panel management strategies
The work doesn't stop once the data is collected. Managing respondents and incentives properly is part of a long-term data-quality strategy.
- 01Identity verification (KYC). For high-stakes research, some panels ask participants to verify identity via a government ID photo or live selfie check. This adds friction but is a real deterrent. Treat vendor "100% catch rate" marketing claims with some scepticism though: independent testing of ID-verification tools shows accuracy varies a great deal by system, and detection rates against AI-generated fake IDs specifically can fall well short of headline accuracy claims built on small test samples.
- 02Incentive control and delayed payouts. Instead of paying instantly, require a waiting period or a minimum earnings threshold before cash-out. This gives the panel time to review data quality and lock a fraudulent account before any money moves, removing the "quick win" that attracts scammers.
- 03Respondent education and awareness. A simple line like "your honest and thoughtful answers are very important to us" primes participants for quality. Avoiding jargon, ambiguity, and needlessly long surveys keeps frustration, and the poor data it produces, to a minimum.
The Yazi approach to data quality
Yazi, a research platform built for WhatsApp, takes a channel-specific approach to fraud. Every participant responds from their own WhatsApp account, tying each person to a unique phone number and naturally limiting duplicates. Controlled distribution avoids the risks of open social media links, and the platform's support for voice notes and photos gives built-in proof that respondents are real, engaged humans, not bots. Combined with backend checks for speeding and gibberish, it's a practical way to keep data clean, especially in emerging markets where WhatsApp is the default channel.
Ready to see how WhatsApp research delivers better data? Book a Yazi demo.
Frequently asked questions
What is the most common type of survey fraud?
Bots automatically filling out surveys, "survey farm" participants from outside the target geography lying to qualify, and individuals creating multiple fake accounts to earn more incentives.
How can I tell if a survey response is fake?
Watch for impossibly fast completion times, the same answer given to every question in a grid, gibberish or nonsensical open-ended answers, and inconsistencies between related questions, like claiming to be childless but later giving children's ages.
Are attention checks enough to stop survey fraud?
No. They catch inattentive or unsophisticated respondents, but Pew's research found the large majority of determined fraudsters still pass them. Use attention checks as one layer in a broader strategy, not the only defence.
How does using a platform like WhatsApp help reduce survey fraud?
Each user is tied to a unique phone number, making duplicate accounts harder. Distribution runs through direct invitations rather than open links, and the ability to collect voice and image responses acts as a strong form of human verification that's hard for a bot to fake. See why WhatsApp works well for research in Africa.
Why does reducing survey fraud matter so much?
Fraudulent and low-quality data can corrupt an entire dataset, leading to inaccurate findings and, in turn, costly decisions based on a false picture of the market or audience. Clean data is the foundation everything else in a study depends on.
See speeder detection, gibberish flags, and unique-device checks running on WhatsApp.
Ready to see how WhatsApp research delivers cleaner data? Book a Yazi demo to see the layered fraud defences in action.
Book a Demo →
