How to keep research incentives in the hands of real respondents
By Abby Quillen●4 min. read●Sep 22, 2026
These days, bots are popping up all over the research industry. AI agents simulating humans can rotate through IP addresses and submit thousands of responses within hours. Fraudsters can use AI to generate convincing survey responses and evade screening. Professional survey-takers often complete surveys to collect incentives rather than provide thoughtful responses. Researchers at UC Davis report that their share of usable survey responses has decreased from 75% to 10% in recent years amid rising survey fraud.
These bad actors all share a target: your incentive budget. When research incentives go to a bot or a fraudster, that’s money spent on data you can't use. In this article, you'll learn how to layer quality checks before, during, and after your survey so your incentives (and your findings) go to real, engaged humans.
Panel quality is under pressure
If you’ve noticed a decline in participant quality, you’re not alone. According to Mary Draper, Vice President of Business Development at EMI Research Solutions, close to 30% to 40% of the survey data her organization gathers is now questionable, compared to 10% to 15% in the past. A study comparing five different panel platforms found substantial differences in platform participants’ ability to pass attention checks, provide meaningful answers, and follow instructions.
AI-generated answers are commonplace. In one lab, researchers observed evidence of LLM use in 45% of submissions. Even well-intentioned survey respondents may be using LLMs, which can water down data. In one study, 34% of survey takers admitted to using large language models (LLMs) like ChatGPT to answer open-ended questions. The AI-generated answers in this study were more homogeneous and positive than human-written answers, which could lead organizations to draw inaccurate conclusions from survey data.
It's also getting harder to spot the fakes. A separate analysis of open-ended survey questions found that AI-generated responses appeared higher quality than human-written answers. Plus, at the same time, AI agents are increasingly sophisticated at mimicking human actions to get past traditional safeguards such as CAPTCHAs, attention checks, and speed checks.
No single guardrail can carry the load anymore. Instead, think of fraud detection as three checkpoints: one before the survey, one during it, and one before you pay.
Checkpoint 1: Verify identities before the survey
Use multiple techniques to verify each participant's identity, combining digital tools with human oversight.
Technologies such as digital fingerprinting, IP tracking, SIM and browser extension tracking, and proxy and bot detection help you catch fraudulent participants at the door. It's also a good idea to reference participant info across trusted databases to spot falsified credentials. For consequential studies, it may even make sense to request video verification or live interviews so you can assess trustworthiness directly.
Checkpoint 2: Build detection into the survey itself
Fraud detection starts with good survey design. Randomize your questions to make them harder for bad actors or bots to game. Check for consistency by asking the same question in different ways and by including follow-ups later in the survey. Add some simple questions mid-survey (like "How many hours are in a day?") to check for engagement.
In addition to traditional CAPTCHAs, include questions that are invisible to humans and only visible to bots.
To deter well-meaning participants from leaning on AI, clearly direct them to rely only on their personal insights and avoid using generative AI. Explain why it matters, because many people use AI to clean up their grammar or sound more polished without realizing how much it affects survey results.
Detecting AI text in completed surveys can be trickier. Outputs vary based on how users prompt and edit them, and tools designed to detect AI-generated text often aren't accurate. Even with those challenges, it's critical to flag answers that don't include personal details, sound polished but lack specificity, or have obvious structural similarities to known LLM outputs.
Checkpoint 3: Review for fraud before you pay
Many research teams have identity verification and in-survey checks in place, but some skip a final checkpoint at the incentive phase, where redemption and payout data can expose fraud that earlier checks miss. Catching fraudsters at this stage saves the payout itself and keeps bad data out of your analysis.
Your fraud prevention measures should help flag:
Reused IP addresses: recipients trying to redeem rewards from a single IP address, device, or network within a short period
Geolocation mismatches: participants who complete a survey in one country and redeem rewards from another
Multiple emails tied to the same device: participants who take studies multiple times or rotate through false identities
Reused payment information: multiple people using the same bank or other financial account
Suspicious redemption patterns: participants already flagged by other organizations in the incentive platform's network
Your incentive strategy supports each checkpoint
Detection determines who gets paid, but your incentive strategy shapes who shows up in the first place.
Offering too much money may backfire by attracting the wrong people. Participants who are just in it for the money may be more likely to sign up.
At the same time, if your incentive is too low, you may struggle to recruit participants in a timely manner, and it could impact the quality of your data. In one study, a higher reward reduced the time needed to fill a participant quota from 68 hours to four hours. The data quality also modestly improved with higher rewards, as did the length of the open-text answers. Meanwhile, the fraud-flag rate stayed roughly the same.
Tremendous research suggests participants want to be paid about $26 for a 15-minute task and $69 for a 60-minute task, but their expectations vary based on their background (students expect less than executives) and the sensitivity of the subject matter. A free research incentive calculator can help you determine the right reward amounts for your study.
The payoff: authentic human signal in a noisy landscape
The rise of bots, AI-powered fraud, and AI-generated answers is making it harder to know whether the people taking your survey are actually people. Layered detection (identity verification before the survey, checks during it, and fraud review before payout) keeps incentives out of the wrong hands. A well-calibrated incentive strategy attracts the real humans you want in the first place.
Together, these measures help you capture genuine human responses, collect more trustworthy data, and make sure every incentive dollar lands with a real person.
