There’s a lot of folklore about survey completion rates — “keep it under 10 questions,” “mobile users won’t finish anything long,” “shorter always wins.” Some of it holds up. Some of it doesn’t, or only holds up once you control for the right variable.
We pulled real numbers instead of repeating the folklore. Starting from 459,682 raw respondent sessions across SurveyRock’s history, we built a clean three-year analysis set — 2023-07-05 to 2026-07-05, 85,900 responses after filtering out test accounts, bot-speed completions, and broken timestamps — because recent behavior is a better guide to 2026 than an older blend. (The full 12-year set, 400,050 responses, shows the same pattern.) Full methodology at the bottom. This is a fixed, point-in-time pull, not an ongoing count — see the methodology for what that means.
The headline number: completion falls off a cliff between 15 and 16 questions
| Questions | Completion rate |
|---|---|
| 1–5 | 99.4% |
| 6–10 | 97.4% |
| 11–15 | 98.0% |
| 16–25 | 77.1% |
| 26+ | 80.3% |
Below 15 questions, completion is essentially flat — you’re trading almost nothing for a slightly longer survey. Cross 15 questions and completion drops 21 points in one step, from 98% to 77%. That’s not a gradual slope. It’s closer to a threshold effect.
(One honest wrinkle in the data: completion ticks back up slightly at 26+ questions versus 16–25 — 80.3% vs. 77.1%. We don’t have a confirmed explanation. One plausible read: surveys that long are more likely to be deliberate, high-stakes instruments — annual engagement surveys, structured research — completed by a more motivated audience than a 20-question survey someone half-heartedly extended. We’re reporting the number, not the theory.)
Most abandonment happens before the halfway point
Where do people who don’t finish actually stop?
| When they quit | Share of all abandonment |
|---|---|
| First 25% of the survey | 32.5% |
| 25–50% | 34.4% |
| 50–75% | 14.6% |
| Last 25% | 18.5% |
Two-thirds of all abandonment (66.9%) happens in the first half. Once a respondent passes the midpoint, they’re considerably more likely to finish than to quit.
This matters for where you spend editing effort. If your completion rate is bad, the instinct is to trim the end of the survey. The data says look at the front half instead — that’s where you’re actually losing people.
Mobile completion holds up on short surveys, then falls behind on long ones
The uncontrolled comparison is misleading here, so we’re only publishing the controlled version. Split by question-count bucket:
| Questions | Desktop completion | Mobile completion |
|---|---|---|
| 1–5 | 98.3% | 99.6% |
| 6–10 | 98.9% | 97.1% |
| 11–15 | 97.9% | 98.1% |
| 16–25 | 82.0% | 75.7% |
| 26+ | 84.0% | 76.2% |
On surveys of 15 questions or fewer, mobile and desktop are close — mobile even edges ahead on the shortest surveys. Past 15 questions, mobile falls 6–8 points behind desktop. If your audience is mostly on phones (ours is — see below), the 15-question line is a harder ceiling for you than it is for a desktop-heavy audience.
A separate, related number: across the same three years, 72.3% of all responses came from mobile devices, versus 27.7% from desktop — and mobile’s share is growing (it was closer to 69% three years ago). We can’t cleanly separate tablets from that split; more in the methodology note below.
Open-ended questions cost about 4 points of completion
| Survey type | Completion rate |
|---|---|
| Has at least one open-ended question | 90.8% |
| Closed-ended questions only | 95.1% |
Not a reason to avoid open-ended questions — they’re often the most valuable data you collect, and this isn’t a large enough gap to justify cutting your best question. It’s a real, specific cost, worth weighing against what the open-ended answer is actually worth to you.
What this data can and can’t tell you
Being direct about the limits, because a benchmark that hides its limits isn’t trustworthy:
- This is SurveyRock’s own respondent mix, not a universal sample. Our accounts skew toward people-ops, CX, and product-research use cases. If your audience or use case is very different, treat these as a strong prior, not a guarantee.
- “Completion rate” here means the respondent finished the survey, not that they answered thoughtfully. We didn’t measure straight-lining or answer quality — only whether they reached the end.
- The recent-3-years window was a deliberate choice. We also have data back to 2014; the last three years is more representative of 2026 respondent behavior (mobile web, attention patterns, distribution channels) than a 12-year blended average would be.
- Correlation, not designed experiment. We didn’t randomly assign question counts to otherwise-identical surveys — a real, longer survey might differ from a shortened version of itself in ways beyond length (more topics, different audience, different intent). Treat the question-count relationship as a strong pattern, not a guaranteed causal lever.
The bottom line
If you take one number from this piece: 15 questions is the practical ceiling before completion rate drops sharply, and that ceiling is tighter if your audience is mostly on mobile. If you take two: most of the fight for completion happens in the first half of the survey, not the last few questions.
Neither of those replace measuring your own surveys — but they’re a better starting assumption than “keep it under 10” with no data behind it.
Want to see this data for your own surveys? Start free with SurveyRock — the completion funnel and per-question timing are built into every survey, from the free plan up.
Methodology
Source: 459,682 raw respondent sessions, queried directly (not sampled) from a complete copy of SurveyRock’s production database, current through 2026-07-05 — the date we cut over to our current platform. That makes this a fixed, point-in-time dataset rather than a live feed: it doesn’t grow on its own. If we rerun this analysis against data from the current platform, we’ll publish it as a separate, dated piece rather than editing the numbers below. Window: last 3 years, 2023-07-05 to 2026-07-05. Final N after exclusions: 85,900 for the last-3-years figures above (400,050 for the full 12-year history, used only as a secondary reference — see the companion pieces below).
Exclusions applied: sessions with invalid/zero timestamps; sessions where the recorded finish time preceded the start time (a data artifact, not a real response); responses on accounts with test/demo-pattern names; and likely bot or junk completions, defined as averaging under 1 second per question among finished responses (a threshold sized relative to survey length, not a flat cutoff — a flat cutoff would have wrongly excluded thousands of legitimate 1–2 question surveys). Quiz-mode surveys were checked separately and found negligible (effectively unused on this platform) — no filter was needed.
On the zero-tablet result: our device tracking recorded literally zero tablet sessions across the entire dataset. We don’t believe that means no one ever took a SurveyRock survey on a tablet — it’s much more likely a device-detection gap where tablets were classified as desktop or mobile. We’re reporting the Desktop/Mobile split we can stand behind and flagging the tablet gap rather than guessing at a number.
This snapshot reflects data through 2026-07-05, pulled 2026-07-11. It won’t change on its own. If we publish an updated analysis using data from the current platform, it’ll be a new, dated piece — not a silent edit to the numbers above.
Companion pieces using this same data pull: How Long Should a Survey Be? and Mobile-First Survey Design.