The short version
Most of the response rate is set before anyone opens the link. People who don't answer have mostly forgotten, seen nothing come of the last survey, or weren't sure who would see their answers. Reminders, a named sponsor and a visible readout deal with all three.
Inside the survey, the wins are small: say how long it takes and who sees what, keep it short, let people skip, and make it easy to answer on a phone.
Points, badges and story worlds make surveys more fun, but in controlled tests they raised drop-out and made answers less reliable. We leave them out.
By the numbers
First, what response to expect from a survey at work. Then the findings that move it, up or down.
What response rate to expect
Treat these as a range, not a target. Rates run lower for senior people, and a high rate doesn't guarantee good data: across 59 studies, the response rate was a poor predictor of how biased the results were. Look at who answered, team by team, as well as how many.
Sources: Anseel et al. 2010 · Groves & Peytcheva 2008
What moves it
Where people drop out
Each step loses people for a different reason, so each needs its own fix. In italics is what someone is silently asking at that step, followed by what helps.
- Hear about it. Does anyone I trust want this? A leader announces it a week ahead, managers mention it, and the purpose and the results date are clear from the start.
- Open the link. Is it worth it, and is it for me? A short invitation from a named person, with the minutes and the deadline. Then reminders, because most people who don't answer simply forgot.
- Get in. Can I get in without hunting for a password? Everything people need to get in travels in the same message as the link.
- Start. Is it safe, and is it quick? How long it takes, when it closes and who sees what, before the first question. No wall of legal text.
- Finish. How much is left? Does this fit me? Short, clear questions that work on a phone, and a way to stop and come back.
- Next time. Did anything happen last time? A readout within weeks, and what changed, before the next survey goes out.
The evidence pins down three of these: most people who don't answer forgot (step 2), break-offs cluster on the first screens and at section intros (step 5), and people who refuse doubt that anything will happen (step 6).
What's built into our surveys
- One question per screen, with large tap targets and no grids, sliders or drop-downs, the formats phones handle worst.
- Choosing an answer moves you on, and every question has a Back button to change it.
- Answers save as people go, so they can close the tab, come back to the same link later and pick up at the question where they stopped.
- People see only the questions for their path. Where a survey has more than one, like the 360°'s engineering and business paths, nobody wades through questions that don't fit them.
- Invitations and reminders go out for you. When you launch for a list of email addresses, the invitation goes out in your name with the closing date. Reminders go only to people who haven't answered: by default three days after the invitation, and again the day before the survey closes.
- Your company's name, or your logo, is on every screen, so people know who is asking.
- Checks for careless answers. On the 360°'s engineering path, one question offers options that don't exist, and picking one flags the response in the report. In AI Adoption Barriers, the belief statements alternate direction, so agreeing with everything doesn't pass.
- Open questions are rare, and always optional. People skip them about 18% of the time, against 1–2% for questions with set answers.
- No points, badges or streaks.
Sources: Selkälä & Couper 2018 · Antoun, Couper & Conrad 2017 · Pew Research Center 2021
What the research says
Findings on who answers surveys, and how honestly. Many of the studies used students or online panels rather than employees, and vendor figures are marked as such.
Reminders bring in most of the answers
In a survey of 15,651 students, the invitation alone got 6.2% to answer, and three reminders took it to 31.2%. Most people who don't answer simply forgot: only about 15% had decided not to. Changing the wording of each reminder raised the odds of a response by more than 30%, while the day and hour they went out made no reliable difference.
Sources: Van Mol 2017 · Rogelberg et al. 2003 · Sauermann & Roach 2013
Who asks matters
Across 2,037 organizational surveys, advance notice, personalization, a relevant topic and a credible sponsor all went with higher response. A personal greeting raised response from 53.5% to 62.9% and cut break-off from 14.2% to 9.0%. Short invitations beat long ones, and the word "survey" in a subject line lowered response.
Sources: Anseel et al. 2010 · Heerwegh & Loosveldt 2006 · Cochrane review 2023
The topic decides who shows up
Teachers answered a survey billed as being about education at 56%, and a general one at 39%. In another experiment, describing a survey as being about birding moved its estimate of how many people go birding by 13 to 32 points. An AI survey will hear mostly from AI enthusiasts unless the invitation makes clear that everyone's answer counts, including people who never use AI. Who answers matters more than how many: across 59 studies, the response rate was a poor predictor of how biased the results were.
Sources: Groves, Presser & Dipko 2004 · Groves et al. 2006 · Groves & Peytcheva 2008
Say how long it takes, honestly
Told a survey would take 10, 20 or 30 minutes, 75%, 65% and 62% of people started, and answers late in a long survey got faster, shorter and more uniform. Understating the length wins more starts but loses them to break-offs, so the final response comes out about the same. People put the ideal length at 10–15 minutes, and the most they'd accept at 20–28. Every question costs something: across 25,080 real surveys, completion averaged 89% at 10 questions and 79% at 40.
Sources: Galesic & Bosnjak 2009 · Heerwegh & Loosveldt 2006 · Revilla & Höhne 2020 · Liu & Wronski 2018
Progress bars and pep talks don't fix length
Across 32 experiments, a progress bar moving at a steady pace made no significant difference to drop-out. A bar that crawled at the start made things worse: in one study, 21.8% broke off, against 12.7% with no bar at all, and showing progress only now and then limited the damage. Encouraging messages fared no better, with no link to break-off among 110,427 students. Shortening the survey does more than any bar.
Sources: Villar, Callegaro & Yang 2013 · Conrad et al. 2010 · Sakshaug & Crawford 2009
Required questions backfire
Among people who hit a required question, 35% dropped out, against 9% of the rest, and 25% said they had made up an answer, against 13% when they could decline. A "don't know" button isn't the fix either: offering one left 24–33% of answers missing. Letting people skip, then asking once more, left 0.5% missing and gave the most reliable answers.
Sources: Décieux et al. 2015 · De Leeuw, Hox & Boevé 2016
Put words on every point of a scale
Scales with a word on every point are more reliable than scales with words only at the ends. Questions worded around the thing itself ("How often do you…") beat statements to agree or disagree with, and where you do use agree or disagree, five points work better than seven or eleven.
Sources: DeCastellarnau 2018 · Krosnick & Presser 2010 · Saris et al. 2010 · Revilla, Saris & Krosnick 2014
Design for the phone first
Across 12 European countries in 2021–23, between 35% and 57% of people answered web surveys on a smartphone. With a simple design, phone answers matched computer answers. Sliders were among the worst formats: in one study they raised the odds of breaking off 6.9 times.
Sources: Claassen et al. 2025 · Antoun, Couper & Conrad 2017 · Funke, Reips & Thomas 2011
Privacy changes what people say
When people can be identified, they soften what they say: subordinates who had to put their names to ratings of their managers rated them higher. Among students known to have cheated, 74% admitted it when they were anonymous, against 25% when they were only promised confidentiality. Complete anonymity has a cost too: in one set of experiments, anonymous answers were less accurate. Promises of confidentiality help when the topic is sensitive, and AI use at work is one. Employee-survey tools hide results for small groups: Viva Glint needs 5 answers to show a score and 10 to show comments, and Lattice starts at 5.
Sources: Antonioni 1994 · Ong & Weiss 2000 · Lelkes et al. 2012 · Singer, Von Thurn & Miller 1995 · Viva Glint privacy thresholds (vendor) · Lattice (vendor)
Close the loop
People who refuse surveys tend to believe nothing will come of them, and only 8% of employees strongly agree that their organization acts on survey results. Across more than 4,500 companies, sharing results broadly went with about 7 points higher participation in the next survey.
Sources: Rogelberg et al. 2000 · Gallup 2023 (vendor) · Culture Amp (vendor)
A personal result helps, but lead with the purpose
Offering people a profile of their own answers offset low interest in a survey's topic, and feedback made people more satisfied without changing what they reported. It isn't a reliable draw on its own: in another study, feedback didn't raise participation, and an incentive built around a personal video lowered starts, partly because its invitation said less about why the survey mattered. At scale it can carry a whole project: about 750,000 unpaid people took LabintheWild's studies, with feedback as the only reward.
Sources: Marcus et al. 2007 · Kühne & Kroh 2018 · Wenz et al. 2022 · Kalleitner et al. 2022 · Reinecke & Gajos 2015
Don't survey the same people too often
People invited to several surveys in a short time were less likely to answer the next one. Viva Glint's guidance shrinks a survey as it gets more frequent: two questions or fewer weekly, eight monthly, 22 quarterly and 30 a year.
Sources: Peytchev 2011 · Viva Glint survey design (vendor)
Game layers backfire
A review found that gamification mostly made surveys more enjoyable, with unclear effects on completion and data quality. In a two-wave study, the gamified version had 10.3% break-off against 4.8%, took 15.6 minutes against 12.7, and gave less reliable answers. Coins for each completed survey raised response among the people who spent them, and made their answers less reliable.
Sources: Keusch & Zhang 2017 · Mavletova 2015 · Dejonckheere et al. 2024
Incentives
Across 32 experiments, material incentives raised the odds of starting by 19% and of finishing by 27%, but a charity donation lowered response in an experiment with 6,162 people. Few employers pay people to answer an internal survey, so protected time and a useful result are the practical incentives.
Sources: Göritz 2006 · Pedersen & Nielsen 2016
Catch careless answers gently
About 10–12% of students taking a long survey answered carelessly in one study. A short "you answered very quickly" prompt slowed people down across six experiments, without raising drop-out. Trick questions are a blunter tool: who passes them depends on the person, so throwing out everyone who fails can skew the sample. Flag those answers rather than dropping them.
Sources: Meade & Craig 2012 · Conrad et al. 2017 · Berinsky et al. 2014
New formats: AI follow-ups, chat and voice
An AI follow-up question on an open answer drew 57 to 70 more words, cost almost nothing in satisfaction and barely raised drop-out, but answers to later questions became more polarized, so it belongs at the end. Running a whole survey as a chat got more people through in one study, 54% against 24%, but took 24 minutes against 18, and another study found shorter answers and less reliable scales. Spoken answers ran about three times as long as typed ones, but 39–42% of people skipped them, against 8–11% for typing. AAPOR's 2026 guidance treats AI that writes or rewords questions as high risk, and asks for disclosure and logs.
Sources: Austin et al. 2026 · Xiao et al. 2020 · Zarouali et al. 2023 · Lenzner, Höhne & Gavras 2024 · AAPOR 2026
Honest answers about AI
Using AI at work still carries a stigma, which makes an AI survey a sensitive one.
- 52% of AI users are reluctant to admit using it for their most important tasks (31,000 people).
- 48% of desk workers would be uncomfortable telling their manager they use AI (17,372 people).
- 57% hide their AI use at work and present AI output as their own (48,000+ people).
Experiments back these figures: people who use AI are judged lazier and less competent, and they expect to be. Misreporting can go either way, though. Where a company mandates AI, people may overstate their use instead.
Sources: Microsoft Work Trend Index 2024 · Slack Workforce Index 2024 · KPMG & University of Melbourne 2025 · Reif, Larrick & Soll 2025
What that means in practice
- Say confidential, not anonymous, when a name or email is kept, and say who sees what.
- Welcome non-users. "Whether you use AI every day or never, your answer counts the same." Otherwise you mostly hear from enthusiasts.
- Ask about what people did, in a set period. "In the last 7 days, how often did you…" is easier to answer honestly than "Are you an AI user?"
- Make personal and unapproved tools an ordinary answer. List them beside the approved ones, without comment.
- Ask what colleagues do, too. People misjudge their peers: about 60% of students said they used AI, and they guessed that 90% of their peers did. The gap is a finding in itself.
- Use tool data for how much, and the survey for why. Across 106 comparisons, what people reported about their media use matched the logs only moderately.
- Keep the results for development, not performance reviews. Across 22 studies, ratings collected for administrative decisions were about a third of a standard deviation more lenient. Say "development only" in the announcement only if you commit to it.
- Read self-ratings as self-perception. Across 22 meta-analyses, the link between people's self-rated skill and their measured performance averaged r = .29, and people using AI also overestimated their own performance.
Sources: Ling, Kale & Imas 2026 · Parry et al. 2021 · Jawahar & Williams 1997 · Zell & Krizan 2014 · Fernandes et al. 2026
Your launch plan
A two-week window, adapted from employee-listening practice. If you launch for a list of email addresses, we send the invitations and up to two reminders. The rest is yours.
| When | What | Who |
|---|---|---|
| 2 weeks before | Decide who sees what, the readout date and one or two commitments. Brief managers and AI champions. | HR or the program owner |
| 1 week before | Leader announcement: why, what's asked, how many minutes, who sees what, the readout date, and that it isn't a performance review. | CEO, CTO or CHRO |
| 1 day before | Managers mention it in team meetings and book 10–15 minutes of protected time. | Managers |
| Launch day | Post the link in your main work channel, with everything people need to get in in the same message. Avoid holidays and crunch weeks; the day and hour matter little. | Sponsor |
| Day 3 | First reminder in your main channel, worded differently, with company-wide progress. | Program owner |
| Day 7 | Second reminder, from managers, with 10 minutes set aside in team meetings. | Managers |
| Day 10 | Last call, 72 hours before the survey closes. | Sponsor |
| Close + 1 day | Thank people, and share the participation rate. | Sponsor |
| Close + 3–5 days | Headline results, and one or two commitments. | Sponsor |
| Close + 2 weeks | Team conversations, only for groups of 5 or more. | Managers, champions, HR |
| Close + 4 and 8 weeks | Short updates on what changed. | Sponsor or HR |
| 2 weeks before the next run | Recap what changed, then announce the next survey. | HR |
Don't chase 100%. Full response can come with complaints of coercion, and a team at 95% or more is worth a second look.
Sources: Viva Glint rollout templates (vendor) · Anseel et al. 2010 · Van Mol 2017 · Holtom et al. 2022
Watch where people stop
While a survey is open, watch each step, not only the total. In AI Adoption Score, every report's Analytics tab shows the steps from invited to submitted, how long people took, and how far the unfinished ones got.
| What to watch | What it tells you |
|---|---|
| Opened the link, of those invited | Whether the invitation and the reminders land |
| Started, of those who opened it | Whether the way in and the first screen work |
| Submitted, of those who started | Whether the length and the questions work |
| How far the unfinished got | Which part of the survey loses people |
| Time taken | Whether the time you promised was honest |
| Response by team, against its headcount | Whose voice is missing from the results |
The last row matters most. A team that's barely answered can tilt the results however high the company-wide rate is.
Do and don't
The advice on this page, side by side: what to do on the left, what to avoid on the right, each with a small example. Pick a stage to narrow it down.
Key studies
The research this page leans on most, from the standard guide to survey response to experiments with tens of thousands of people. Each card says what the study found and what it means for a survey at work.