Screen out bots and AI agents before the first question
A respondent who never reaches the questionnaire costs nothing to remove. ResearchDart checks the browser, the network and the respondent's history at the entry link, then lets real people through in a few seconds.
| Signal | Sessions | Action |
|---|---|---|
| Datacenter or VPN network | 68 | Blocked |
| Automated browser | 44 | Blocked |
| Same device, new respondent ID | 61 | Blocked |
| Multi-accounting risk | 37 | Flagged |
| Provider unavailable | 9 | Allowed |
| No risk signals | 1,623 | Passed |
What happens between the entry link and the survey
Screening adds one step to the respondent journey. Suppliers keep their entry links and buyers keep their survey URLs.
Entry checks
Project status, quotas, repeat respondent IDs and duplicate devices are checked on the server as the link opens.
Holding page
A short page loads the screening script, which collects device, network and interaction signals from the browser.
Server decision
ResearchDart asks the provider for a decision on that browser session and compares its location with the project country.
Project policy
The project's mode decides whether a risky session is blocked, flagged or allowed, and records why.
Survey or return
Passed respondents open the survey. Blocked respondents return to their panel as quality terminates with a GDQ code.
Signals that still separate people from software
Language models can answer attention checks. They cannot easily hide the automation around them, the infrastructure they run on, or the fact that the same operator is behind many panel accounts.
- Automation
- Headless and scripted browsers, agent-driven sessions and browsers with no human interaction events.
- Anonymised networks
- VPN, proxy, Tor, spoofed addresses and datacenter hosting, which are rare among genuine panel members.
- Location
- A connection whose true country contradicts the project country, including location spoofing and impossible travel.
- Uniqueness
- The same device entering under a new respondent ID, and one person operating many accounts across panels.
- History
- Addresses and devices recently associated with fraud, from the screening provider's network.
Roll out without guessing
Every project chooses how strict to be. Change the mode at any time; the history of every decision stays in the session record.
- Off
- No screening. Entry checks for duplicates and quotas still run.
- Monitor
- Every session is scored. Risky sessions are flagged in reports but still enter the survey, so you can compare flags with your own cleaning.
- Enforce
- Denied sessions are stopped before the survey. Suspicious sessions are blocked too if you choose.
- Outage policy
- Fail open to protect fieldwork speed, or fail closed when a study cannot accept unscreened traffic.
- Sandbox sessions
- Test sessions skip quotas and duplicates, so you can check the screening page on real devices.
- Supplier visibility
- Suppliers see the screening status and removal reason for every respondent they sent, by API and in exports.
Built-in Verisoul integration
ResearchDart connects to Verisoul with your own project ID and API key. The browser SDK runs on the holding page, and the decision comes from Verisoul's session authentication API.
Each respondent is sent to Verisoul as a stable, one-way hash of the supplier and respondent ID, which lets multi-accounting be detected across studies without sharing the supplier's own identifier.
Verisoul is a trademark of its owner. ResearchDart requires a Verisoul account of your own; the provider layer is designed so other screening services can be connected.
| Provider result | ResearchDart reason | GDQ code |
|---|---|---|
| Fake, high bot score | automation | 1 |
| Fake or suspicious, anonymised network | anonymized_network | 2 |
| Fake or suspicious, other risk | fraud_tool | 2 |
| Real, country contradicts project | geo_mismatch | 3 |
| Duplicate device or respondent ID | duplicate_device | 4 |
Screening questions
What does pre-survey screening check?
Does screening slow respondents down?
What is the difference between monitor and enforce?
Which screening provider does ResearchDart use?
What happens if an ad blocker stops the screening script?
Is screening data shared with buyers?
Further reading
AI agents now pass attention checks. Plan your fieldwork accordingly
A research team built an AI respondent that passed 99.8% of attention checks. That does not mean online sample is finished. It means the checks most s...
Data QualityPre-survey fraud screening: what it catches, what it misses and what it costs
Screening respondents before the survey opens is the cheapest place to remove fraud. It is not free, it is not complete, and it creates false positive...
Data QualityMonitor before you enforce: turning on fraud screening without wrecking fieldwork
The fastest way to lose a good supplier is to switch on blocking and watch their traffic disappear without an explanation. Run screening in monitor mo...
Measure your fraud rate before you enforce
Turn on monitor mode for one project, compare the flags with your own data cleaning, and decide with evidence.