The Data Quality Analyst agent uses Autonomous Insights to detect low-quality, fraudulent, or suspicious survey responses. It automatically flags problematic data to reduce manual review and ensure better data integrity.
Access the Data Quality Analyst by navigating to the Survey List, selecting View Responses in the Actions menu, then clicking on the Run Quality Check button at the top right.
Choose which checks you want to perform. All checks will be selected by default, but you can uncheck any that you feel are not essential for your requirements.
A banner displays the progress at the top of the responses list. You are notified when the process is complete.
Survey Response Checks
We provide recommendations based on Fuel Cycle's standard, though we recognize that this standard may not suit everyone. Our goal is for this flag to help you quickly find the responses you need.
Each response is evaluated across two main areas:
- Behavioral Checks—Analyzes response patterns and metadata
- Open-End Checks—Analyzes open-text answers
Behavioral Checks
Through-Liner
Grid responses with the same rating across all items.
Example:
Q: How much do you agree with the following statements?
A: 5, 5, 5, 5 (Flagged)
IP Duplicate
Multiple responses submitted from the same IP address.
Region
Responses from regions known for high fraud rates.
Flagged regions include:
China, India, Bangladesh, Taiwan, Vietnam, Kazakhstan, Russia, Thailand, Venezuela, Indonesia, the Philippines, South Africa, Nepal, Sri Lanka, Egypt, Nigeria, Romania, Ukraine
Speedster
Completed responses were greater than 50 percent faster than the median.
Example:
Median time: 300 seconds
- 100 seconds → Flagged
- 149 seconds → Flagged
- 200 seconds → Not flagged
Open-End Checks
Gibberish
Incoherent or nonsensical responses.
Example:
Q: What is your favorite book and why?
A: asdfghjkl (Flagged)
Off-Topic
Responses that do not answer the question asked.
Example:
Q: How do you approach problem-solving?
A: I love going to the beach on sunny days. (Flagged)
Profanity
Vulgar, explicit, or inappropriate language.
No Response
Empty or non-substantive answers such as:
- None
- NA
- No
- [Blank space]
AI-Generated Response Check
Assesses whether a large language model, such as GPT, likely generated a response.
- More Likely—Indicates AI-generated patterns
- Less Likely—Indicates human-authored content
Evaluation factors:
- Structured or overly verbose phrasing
- Balanced or neutral tone that avoids strong opinions
- Advanced vocabulary that is uncommon for the target audience
- Use of rare or overly formal words
Cross-Duplicate Responses
Near-identical answers from different participants to the same question.
Example:
Q: What drives you professionally?
- Challenge, learning, innovation, making an impact.
- Challenge plus learning plus innovation plus having an impact. (Flagged)
Only applies to responses with 15 or more characters.
Self-Duplicate Responses
A single respondent provides nearly identical answers to multiple open-ended questions.
Example:
Q1: Describe your typical day.
A1: Yoga, healthy eating, focused work, regular breaks, evening reading.
Q2: What does your daily routine entail?
A2: Healthy eating, focused work, regular breaks, and evening reading. (Flagged)
No minimum length is required. Short repeated responses may also be flagged.