In short
Bad survey responses cost you three times: you pay for them like good ones, you spend hours cleaning them out, and you claim refunds one by one. Kantar found that up to 38 in 100 collected responses get thrown away, and a 2026 benchmark put business surveys at about 60 in 100. This calculator works out that bill for your survey in money and hours, the usual way and with Dot, which turns bad responses away before they count.
Your bad data bill. Change the numbers.
What bad responses cost you in money and hours, the usual way and with Dot.
The usual way
$0
With Dot
$0
Dot turns most away before they count. You read its reasons for the rest.
| Where it goes | The usual way | With Dot |
|---|---|---|
| Paying for bad responses | — | — |
| Cleaning them outSpeed, attention checks, duplicates, every written answer | — | — |
| Claiming refundsA reason and evidence per response, before the deadline | — | — |
| Ordering replacementsBriefing the panel again | — | — |
| Protect | — | — |
| Total |
A few fake answers flip your result. That's the costliest line on the bill.
0fake "no" answers
10 to 52 fake answers
That's all it would have taken to flip the predicted winner in seven national polls before the 2024 US election. The AI that wrote them passed almost every attention check (Westwood, PNAS, 2025).3
20 in 100 instead of 3 in 100
Under-30s who agreed the Holocaust is a myth in an online opt-in poll, against 3 in 100 in a carefully drawn sample (Pew Research Center, March 2024).4
$10 million of made-up answers
Sold by two panel companies from 2014 on, US prosecutors say. They reached published university studies (US Attorney's Office, New Hampshire, 2025).9
Everything you'd clean by hand, Dot checks while people answer.
-
Bots and scripts
Mouse, clicks and typing no hand makes, scripts driving a real browser, cloud servers.
By hand you can't: nothing in the answers shows it.
-
AI-written answers
Every written answer scored from 0 to 100, and marked if it was pasted.
By hand: reading every answer and guessing.
-
Attention checks
Dot's own questions, placed for you and scored, with "2 of 3" next to each person.
By hand: writing them, scoring them, picking the cut.
-
Rushing
Too fast for the survey's length, or far quicker than its usual pace.
By hand: sort by time, pick a cut-off, argue about it.
-
Nonsense answers
"asdf", "good", off-topic text, and answers shorter than you asked for.
By hand: reading every open answer.
-
The same person twice
The same browser or address again, or one device answering from three countries in a day.
By hand: matching IP addresses in a spreadsheet, if you have them.
-
Wrong place, wrong hour
Outside your countries, a clock hours off from the city, answering at 4am local time.
By hand: you rarely have the data.
-
No refund claims
Flagged people are turned away before they count, so there's nothing to claim back.
By hand: a reason and evidence per response, before the deadline.6
-
A reason for every flag
"This device answered 33 times in 9 hours, from 8 countries." You see why, in plain words.
By hand: a gut feeling you can't show the panel.
Where the numbers come from.
The hours per step are our own guesses, and you can change every one in the calculator.
- Kantar, The State of Online Research Panels, 2022. Up to 38 in 100 collected responses are thrown away. kantar.com
- Global Data Quality benchmark, first half of 2026. About 60 in 100 business (B2B) completes removed on review. research-live.com
- Westwood, PNAS, 2025. 10 to 52 AI-written answers would have flipped seven national polls. pnas.org
- Pew Research Center, March 2024. 20 in 100 against 3 in 100 of under-30s on the Holocaust question. pewresearch.org
- Pew Research Center, August 2026. Attention checks miss most fake respondents. pewresearch.org
- Cint, reconciliation policy. Refund claims need a reason and evidence for each respondent. legal.cint.com
- Rep Data, reconciliation guide. Completes nobody claims back are billed after a month. docs.repdata.com
- US Bureau of Labor Statistics, 2024. Market research analysts earn about $37 an hour. bls.gov
- US Attorney's Office, New Hampshire, 2025. $10 million of made-up survey data. justice.gov
Questions people ask.
How much of my survey data is likely to be bad?
Kantar's study of online panels found that up to 38 in 100 collected responses get thrown away. A 2026 industry benchmark found about 60 in 100 business (B2B) completes removed on review. It depends on who you ask and where the people come from.
What counts as a bad survey response?
Bots and scripts, AI-written answers, people rushing through without reading, nonsense open answers, the same person answering twice, and people outside the country or audience you asked for.
Do I pay for bad responses?
Usually, yes. Panels bill every complete, and you claim refunds afterwards with a reason and evidence for each response, before a deadline. Completes nobody claims back are billed, and not every claim is accepted.
How long does it take to clean survey data by hand?
Nobody publishes hours for it, so the calculator uses our own guesses, which you can change: a quick look at every response, reading each open answer, a couple of minutes per refund claim and time to order replacements. For a few hundred responses that adds up to days.
Can a few fake answers really change my results?
Yes. A 2025 study in PNAS found that 10 to 52 AI-written answers would have flipped the predicted winner in seven national polls before the 2024 US election. The calculator shows how many it takes to turn your own result around.
Don't attention checks catch bad responses?
Some, not most. Pew Research Center found in 2026 that attention checks miss most fake respondents, and AI passes them easily. Dot uses them as one signal among many and looks at how the survey was answered too. How we check for AI explains each check.
How does Dot cut the bill?
Dot checks every response while people answer and turns flagged people away before they count, so you don't pay for them, clean them out or claim them back. Every flag comes with its reason in plain words. It's part of Protect.
What does Protect cost?
It's included in every Build plan for surveys built in SampleLabs. For a survey built in another tool you pay per person checked, from 5 cents each and less at higher volumes, and the first 10 checks each month are free.
Does it work with my survey tool?
Yes. Paste your survey's link into Protect and people go through Dot's check page before they reach your survey, in Google Forms, Qualtrics, Typeform or any other tool. Protect your survey now.
Stop paying for answers you'll throw away.
Dot checks every response while people answer, in SampleLabs or in the survey tool you already use.
Works with Google Forms, Qualtrics and any other tool.
Rather talk it through? Book a call