Every vendor in this space, us included, will tell you pre-survey screening stops fraud before it costs you money. That is true. It is also the least interesting part of the story. What matters when you decide how to use it is what it misses and what it costs.
We sell screening, so read this as an argument from an interested party, and check it against your own data.
What it catches reliably
Duplicate participation. The same supplier respondent ID entering a project twice is trivial to stop. The same device entering under a new ID is harder but still one of the most dependable checks, because it does not depend on guessing intent. One device, two identities, one study: that is a removal under GDQ code 4.
Careless automation. Headless browsers, scripts with default user agents, HTTP clients pretending to be browsers. A server-side check catches these without any provider at all. It catches only the careless ones.
Infrastructure. Sessions from hosting providers, known proxy networks, Tor and many commercial VPNs. Volume fraud needs cheap infrastructure, and genuine consumer panel members rarely connect from a datacenter.
Location that contradicts the study. A US study receiving a session whose true connection resolves in another country. Providers that look beyond the IP address, at time zone, language and network consistency, catch spoofed locations that a plain IP lookup misses.
What it catches with help from a provider
Agent-driven browsers. Better agents run real browsers with plausible fingerprints. Spotting them needs device intelligence and interaction signals collected in the browser. On ResearchDart that is the job of the connected provider; our own built-in check does not see this.
Multi-accounting across panels. One operator holding many panel accounts shows up as device and identity patterns across sessions and across the provider's customer network. A single study cannot see that on its own.
What it cannot see
Inattentive humans on clean devices. A real person on home broadband who speeds through your grids passes every entry check. In-survey and post-survey checks are still the only defence here, and in the 2025 Insights Association and GDQ benchmark, post-survey cleaning still removed 6.6% to 7.4% of qualified completes globally.
A human using a chatbot to answer. The session is a person on a real device. The fraud is in the open ends. Screening at entry has nothing to measure.
Profile lies. Someone who claims to be an IT decision maker to qualify for a higher incentive is a real human with a real device. Knowledge questions in the survey catch this. Screening does not.
Clean residential proxies used carefully. Residential proxy networks route traffic through real home connections. Good providers flag many of them, but not all.
If a vendor tells you screening removes the need for data cleaning, stop listening.
What it costs
A few seconds for every respondent. Browser-based screening needs the page to load a script, collect signals and wait for a server decision. On ResearchDart the holding page gives up after six seconds. Most sessions finish well inside that, but it is still a step between the panel and the survey, and every step loses some people on slow mobile connections.
Blocked scripts. Ad blockers and privacy extensions stop some third-party scripts. Verisoul's own documentation says that without a first-party custom hostname, 4% to 6% of traffic may be blocked. You then have to decide whether an unscreened session is allowed in (fail open) or stopped (fail closed). We default to fail open and record those sessions as unavailable, because stranding real respondents costs more on most studies than the fraud that slips through a blocked script. Set up the custom hostname and the question mostly goes away.
False positives. The ones we would plan for:
- Corporate VPNs on B2B studies. Many professionals connect through company VPNs. A rule that blocks every VPN will cut into exactly the audience B2B buyers pay most for.
- Shared households and offices. Device deduplication treats two people on one browser as one person. For studies in clinics, schools or shared workplaces, turn duplicate device blocking off.
- Travellers. A genuine US panel member on holiday abroad fails a strict country check.
Supplier friction. A supplier whose traffic is suddenly blocked at 15% will want to know why. Share the reason codes. Screening that looks like an unexplained black box damages relationships with good suppliers along with bad ones.
How we would use it
- Start every new supplier relationship in monitor mode, not enforce.
- Enforce denied sessions once you have compared flags with your own cleaning results.
- Allow suspicious sessions on consumer studies and block them on high-incentive B2B work. In the benchmark, general B2B had the highest removal rate before and during the survey (15.3%) and the highest post-survey cleanout (18.9% of completes).
- Keep attention checks, open-end review and timing checks. Screening narrows the problem; it does not close it.
- Reverse what cleaning still finds, with GDQ codes, so suppliers can act on the source.
The settings behind each of these are on the pre-survey screening page.