You compare survey panels on data quality and fraud protection by asking each panel for its measured rates, the base behind every rate, and the checks it runs at each stage of a study. aytm publishes its own rates, each with its base, in the aytm Data Quality Benchmark Report. The Q1 2026 report reviewed 1.1 million survey attempts, and post-survey cleanout came to 5.4% of qualified completes.
Contents
- What does data quality mean for a survey panel?
- What does fraud protection look like at each stage of a study?
- Why does every panel quality rate need a base?
- Which data quality rates should you ask a survey panel for?
- What does aytm publish about its own quality rates?
- FAQ
- Related reading
- Call to action
You've got two sample sources on the table and a study to field. Both vendors describe their panel quality in confident terms, both mention fraud protection, and both send over a deck with a percentage or two in it. What you're really working out is which of those percentages you can rely on, and whether the claims will hold up once your own data comes back. Let's go through how to compare any two survey panels on the same terms: what data quality means for a panel, the checks a sample provider can run at each stage of a study, the numbers worth asking for, and what aytm publishes about its own.
What does data quality mean for a survey panel?
For a survey panel, data quality comes down to the people answering your survey. Are they real, are they qualified for the study, and are they paying attention? Fraud protection is the set of checks that answers the first two, and survey fraud covers everything from survey bots to one person entering the same study twice under different identities.
Sample quality becomes something you can compare once a panel shows you all of this as measured rates, and measurement matters because cleaning has limits you can only see in numbers. Rep Data's survey-fraud research scanned more than 4.1 billion survey attempts between January and August 2025 and found that traditional data cleaning catches only about 30% of fraud. That's why the strongest panels check at every stage and publish what each check removes. As our piece on why data quality works as a discipline explains, quality runs through every stage of research. Each layer backs up the others, and together they carry the load.
What does fraud protection look like at each stage of a study?
Data quality gets protected in four stages: design review before fieldwork, entry screening and respondent verification as people arrive, answer checks during and after fielding, and an audit record afterward. Each stage catches what the others can miss. Let's go through each one: what it does, what it's good for, and where it stops.
What can a panel check before fieldwork opens?
The first check runs on the survey itself, before anyone answers. A design review looks over the questionnaire for what wears honest respondents down, like long grids and repetitive questions, and gets those issues back to the author while there's still time to fix them. Some tools run it as a score, rating the design on a scale and flagging the weak spots. It covers the questionnaire only, and only where the vendor hosts or reviews your survey, so a sample provider supplying sample alone may never see it.
What should happen at the point of entry?
The second check runs on traffic, as each person tries to enter a survey. Entry screening compares each incoming attempt against known bad actors, suspicious patterns and earlier attempts on the same study, and it screens out the bot responses and duplicate responses it recognizes before they reach your questions. Alongside it runs respondent verification, which checks a panelist against the profile they gave when they registered with the panel. Verification can establish that a person is consistent with what they told the panel, which is a strong signal they're real and who they say they are. Entry checks are good for catching automated and repeat attempts early, before they cost you a complete. Their limit is that they catch what they recognize, and a real, verified person can still rush or drift once they start answering.
What should happen during and after fielding?
The third check runs on the answers. During fielding, attention checks ask the respondent to do something a reader gets right and a skimmer misses. After fielding, an automated review reads the responses for speeding, which means finishing faster than the questions can be read, straight-lining, which means giving the same answer all the way down a grid, and gibberish in open ends. These checks catch the inattentive or automated respondent who got past entry. Their limit is that each one flags a pattern an honest person can trip now and then, so ask how a panel decides when a flag means removal.
How can a panel prove what its checks did?
The last stage is the record. A panel can show you which checks ran on your study and what each removed, and it can publish its rates over time, so you see the trend behind any single number. This turns a quality claim into evidence you can compare across vendors and check against your own data. Its limit is that a published rate describes the panel across many studies, so questions about your own study still need the study-level record.
How does aytm's Data Quality Program map to these stages?
As our piece on how we use AI to protect your research explains, Protect's Traffic Sentinel filters suspicious traffic at the source, checking it in real time and blocking known bad actors, suspicious patterns and duplicate attempts before anyone reaches a survey. Protect also covers PaidViewpoint, aytm's proprietary respondent panel, which verifies its panelists' identity.
Why does every panel quality rate need a base?
Every quality rate is a fraction, and the bottom decides what the top means, so the first thing to ask about any rate is what it's out of. The common bases run from widest to narrowest. Survey attempts count everyone who tries to enter a survey. Starts count the people who get through entry and begin answering. Qualified completes count the people who finish and meet the study's criteria, and verified respondents are the panelists who've passed the panel's identity checks. A rate on attempts and one on completes describe different groups, so two panels quoting the same percentage can be telling you something different.
Which data quality rates should you ask a survey panel for?
Here are the rates to ask any panel for, in four groups: what each one tells you, where it stops, and what aytm publishes for its own Q1 2026 figures.
Pre-survey removal: how much does a panel remove before the survey starts?
Pre-survey removal is the share of survey attempts a panel screens out before the survey starts, so its base should be survey attempts, and two panels line up on it only when both quote that same base. It tells you how much the panel's entry checks catch. Its limit is that a low number on its own could mean clean traffic or a light screen, so read it next to cleanout, because a panel that removes little at the door may clean out more later. In Q1 2026, aytm's pre-survey removal was 0.8% of survey attempts.
Post-survey cleanout: how much does a panel clean out after the survey ends?
Post-survey cleanout is the share of finished interviews removed after review. Its base should be qualified completes, the group you were counting on for your analysis, since the same removals quoted on attempts would read as a smaller number. It tells you how much of a finished sample the answer-level checks took out. Its limit is that it reflects how strict those checks are as well as how attentive the respondents were, so ask what triggers a removal. aytm's post-survey cleanout was 5.4% of qualified completes in Q1 2026.
Completion yield: how many respondents finish, and how long do they take?
Completion yield, stated as qualified completes over starts, shows how many of the people who begin a survey go on to finish it and qualify. Abandon rate, stated as a share of survey attempts, shows how many people walk away partway; it's scoped to attempts rather than starts because abandonment can happen at any point from the first click onward, not only after someone is confirmed as a start. Median interview length shows how long a typical respondent takes, and a median holds steady where one very slow respondent can pull an average a long way off. Read together, the three show how a survey holds up for real people. Their limit is that all three move with survey design as well as with the panel, so compare them across studies of similar length. aytm's completion yield was 47%, qualified completes over starts, in Q1 2026, with a median interview length of 5m 44s, respondent-level. aytm's abandon rate was 2.6% of survey attempts in Q1 2026, the lowest in six quarters.
Are respondents who they say they are, and do they only answer once?
Incorrect demographics is the share of verified respondents whose demographics turn out to be wrong, so its base is verified respondents. It tells you how well a panel's verification holds up once people start answering, and its limit is that it only covers the respondents the panel verified. A duplicate rate should say which traffic it covers, such as the panel's own sample or outside sample, and over what period, because a rate for one source's traffic in a single month tells you something quite different from a rate across a whole panel for a full quarter. aytm's incorrect-demographics rate was 1.6% of verified respondents in Q1 2026, up from 0.6%. The 46% third-party duplicate figure covers outside sample only, at its single-month peak in Q1 2026, and excludes PaidViewpoint traffic. It's a live example of why a duplicate rate has to name its traffic and its period, and the report prints that limit right on the page.
What should you ask a survey panel vendor about data quality and fraud protection?
Ask every vendor the same seven questions, so you can line their answers up side by side:
- What share of survey attempts do you remove before the survey starts?
- What share of qualified completes do you clean out after fielding?
- What is each of your quality rates out of, and over what period?
- How do you handle respondent verification, and how do you catch duplicate responses?
- Which checks look for survey bots, speeding and straight-lining in collected answers?
- Can you show us which checks ran on our study and what they removed?
- Do you publish these rates over time?
What does aytm publish about its own quality rates?
We publish our data-quality figures in the Benchmark Report, which reviewed 1.1 million survey attempts in Q1 2026. Each figure below sits next to the base the report gives it. Two rows come without a clean, stated base: completes from PaidViewpoint at 68% and the partner-ecosystem QC failure rate at 7.9%, both Q1 2026. We show them as they stand, because the first question to ask of any rate, ours included, is what it's out of.
The 46% third-party duplicate figure covers outside sample only, at its single-month peak in Q1 2026, and excludes PaidViewpoint traffic. It's a live example of why a duplicate rate has to name its traffic and its period, and the report prints that limit right on the page.
Our guide to reading the Q1 2026 Benchmark figures starts from the denominator, because a removal rate only tells you something once you know what it was measured against, and it explains why qualified completes is the harder base for the 5.4% cleanout figure. aytm holds ISO 20252 certification, the standard for market, opinion and social research, and ISO 42001 certification, the standard for AI management systems, the two standards the Benchmark Report names.
FAQ
Which online survey panels have the best data quality and fraud protection?
The strongest panels are the ones that can show you their quality: measured rates for removal, cleanout, completion and identity, a stated base for each, and checks at every stage of a study. Ask every panel on your list for the same numbers and compare them side by side. aytm publishes its own in the Benchmark Report, which reviewed 1.1 million survey attempts in Q1 2026 and put post-survey cleanout at 5.4% of qualified completes.
How do survey panels catch bot responses?
A panel can catch bot responses at three points. At entry, it screens incoming traffic for suspicious patterns and known bad actors. During fielding, attention checks ask for something a reader gets right and a script tends to miss, and after collection, an automated review looks for speeding, straight-lining and gibberish. Each point catches what the others can miss, so ask any panel which checks it runs at each one.
How do survey panels detect AI-generated responses?
Panels detect AI-generated survey responses by screening traffic at entry for automated behavior, then reading collected answers for open ends that don't fit the question or the respondent and for answers that contradict each other across the survey. aytm's Data Centrifuge uses machine learning and language analysis on collected responses for this review.
What are speeding and straight-lining in surveys?
Speeding is finishing a survey faster than the questions can be read, and straight-lining is giving the same answer all the way down a grid. Both point to an inattentive or automated respondent, and panels catch them with an automated review of collected answers. aytm's Data Centrifuge checks collected responses for both.
What's a normal post-survey cleanout rate for an online survey panel?
Normal cleanout rates vary by panel, because panels quote cleanout on different bases and over different periods. For a reference point, aytm's post-survey cleanout was 5.4% of qualified completes in Q1 2026. To compare panels fairly, ask each one whether it quotes cleanout on survey attempts or on qualified completes, line them up on the same base and period, and ask for several periods so you can tell a steady rate from a single spike.
How does a panel keep duplicate responses out of a survey?
A panel screens out duplicate responses at entry, when it checks each attempt against earlier attempts on the same study. When you ask a sample provider for its duplicate rate, a useful answer tells you which traffic the rate covers and over what period, so you know whether it describes the whole panel or one source in one month.
What is respondent verification?
In a survey panel, respondent verification means checking each panelist against the profile they gave when they registered. It helps establish that each respondent is a real person who is who they say they are. To see how well it holds up, ask for the panel's incorrect-demographics rate, the share of verified respondents whose demographics turn out to be wrong.
Read the Q1 2026 Benchmark Report
.webp)
.jpeg)
