Screening vendors like to show a dashboard full of blocked sessions. Blocked sessions are not the goal. Clean completes at a sustainable cost are, and a screening rule that blocks 20% of a supplier's traffic might be saving you money or might be throwing away real respondents. You cannot tell which from the block count.
That is why ResearchDart has a monitor mode, and why we tell buyers to use it before enforcing anything. It is also, candidly, the setting that makes our product look least dramatic in a demo.
What monitor mode does
Every session is screened exactly as it would be in enforce mode. The decision and the reason are recorded on the session. Then everyone continues into the survey anyway. Sessions the provider judged risky are marked flagged; the rest are marked passed.
You get the full picture of what enforcement would have done, at the price of fielding some traffic you might later reject.
A rollout that holds up
1. Pick one live study per supplier
Choose studies you will clean properly anyway. Monitor mode is only useful if you have a post-survey verdict to compare against.
2. Let enough traffic through
Wait until each supplier has sent a few hundred entries. That is a judgment call, not a statistical rule, but with a few dozen sessions a single bad afternoon from one sub-source will dominate the result.
3. Join flags to your cleaning outcomes
Export sessions with their screening status and join them to your cleaned survey data on the session ID. You want four groups:
| Removed in your cleaning | Kept in your cleaning | |
|---|---|---|
| Flagged by screening | Agreement: screening would have saved this complete | Possible false positive |
| Passed screening | Screening missed it | Agreement |
4. Read the table honestly
Flagged and removed is the money you would have saved, plus the analyst time.
Flagged and kept is the interesting cell. Some of these are fraud your cleaning missed. Some are real people on corporate VPNs, shared devices or travelling. Look at a sample of them by hand before assuming either.
Passed and removed is what screening cannot see: inattentive humans, profile liars, chatbot-assisted open ends. It tells you how much cleaning you still need, not whether screening works.
5. Decide per study type, not per company
- Consumer studies: enforce denied sessions; flag suspicious ones.
- B2B and other high-incentive work: enforce denied sessions and block suspicious ones, unless your monitoring shows corporate VPN users being flagged in numbers. In the 2025 Insights Association and GDQ benchmark, general B2B had the highest post-survey cleanout of any study type at 18.9% of qualified completes.
- Shared locations such as clinics or schools: turn duplicate device blocking off before you enforce anything.
6. Tell suppliers before you switch
Send each supplier their own numbers from the monitoring period, with the reason codes. A supplier who sees that 70% of their flagged sessions came from datacenter addresses can fix a sub-source. A supplier whose traffic simply stops converting will assume you broke something.
The outage decision
Screening depends on a script running in the respondent's browser. Some browsers block it. Verisoul's documentation estimates 4% to 6% of traffic may be blocked without a first-party custom hostname. You need a policy for those sessions:
- Fail open (our default): let them in and record them as unavailable. Right for most studies, because stranding real respondents costs more than the fraud that slips through a blocked script.
- Fail closed: stop them. Right when a study genuinely cannot accept unscreened traffic.
Set up the custom hostname either way. It makes the question much smaller.
When to go back to monitor
Switch back when you add a new supplier, change the screening provider's rules, or start fielding in a new country. Anything that changes the traffic changes the false positive rate.
The settings for each of these are on the pre-survey screening page. What the screening itself can and cannot catch is in this piece.