The Complete Guide to Surveys

The three tools are often confused because they all collect input, but they answer different questions and fail in different ways when swapped.

A form is transactional: one screen, one purpose, one submitter acting in their own interest - a contact request, an order, a signup. The submitter wants the outcome, so completion motivation is built in. A poll is a single question with instant aggregate results, optimised for participation over rigour; it measures sentiment in the moment but cannot tell you why. A survey is an instrument: multiple questions, deliberately ordered, answered by a sample so you can generalise to a population. The respondent gets little direct benefit, which is why everything in survey design - length, question wording, incentives - exists to compensate for missing motivation.

The practical test: if the respondent needs something from you, build a form. If you need one data point from many people, run a poll. If you need to understand a population well enough to make a decision - what to build, why customers churn, how an event landed - that is a survey.

Mixing them up has real costs. A "quick survey" bolted onto a contact form depresses the form's conversion. A 20-question form pretending to be a survey collects from a biased sliver of your audience. Formspring treats them as separate products with separate builders precisely because the design constraints differ - and if you are unsure which fits, start with the decision in this section, not the tool.

Survey data is only as good as the questions, and most bad data is manufactured by the question itself. The failure modes are well catalogued - Pew Research Center's question-writing guidance is the canonical public reference:

• Leading questions smuggle in the answer: "How much did you enjoy our redesigned dashboard?" presumes enjoyment. Neutral framing - "How would you rate the redesigned dashboard?" - costs nothing and removes the thumb from the scale.
• Double-barrelled questions ask two things at once: "Was the checkout fast and easy?" Fast-but-clunky respondents have no honest option. One question, one dimension.
• Absolutes and recall stretches: "Do you always read release notes?" and "How many times did you log in last quarter?" both force fabrication. Bound the recall window to what people genuinely remember - usually the last 30 days.

For closed questions, keep scales consistent through the survey (do not flip 1-is-good to 1-is-bad midway), label every point or at least both ends, and use 5- or 7-point scales - granularity beyond 7 adds noise, not signal. Include a genuine opt-out ("Not applicable") so people without an opinion stop inventing one.

For open questions, ration them: a common practitioner rule of thumb prices each one at roughly 4x the completion time of a closed question, and drop-off rises accordingly. One well-placed open question - typically "What is the main reason for your score?" - yields more than five scattered ones.

Order matters too: start with easy, engaging questions, hold demographics until the end, and randomise option order where sequence could bias selection. The question design deep-dive works through before/after examples.

The three standard satisfaction metrics answer different questions, and mature teams run all three at different touchpoints rather than picking a favourite.

NPS (Net Promoter Score) - the metric Bain & Company created - asks "How likely are you to recommend us?" on a 0-10 scale, then subtracts the percentage of detractors (0-6) from promoters (9-10), yielding a score from -100 to +100. It measures relationship strength, so survey on a cadence - quarterly or twice yearly - not after individual transactions. As a practitioner benchmark, B2B software typically lands between +20 and +40; anything above +50 is excellent. The score itself is less valuable than the follow-up question ("What is the main reason?") and the trend line. Running NPS properly means consistent timing, consistent channel, and closing the loop with detractors within days.

CSAT (Customer Satisfaction) asks "How satisfied were you with X?" on a 1-5 scale immediately after a specific interaction - a support ticket, an onboarding call, a delivery. Report the percentage scoring 4 or 5. Because it is transactional, send it within an hour of the event while memory is fresh.

CES (Customer Effort Score) asks how easy something was, typically 1-7. The Harvard Business Review research that introduced it found it the best predictor of churn after support interactions: effort drives disloyalty more than delight drives loyalty.

In Formspring's survey builder, NPS, CSAT, and CES are first-class question types with scoring built in, available on Pro plans and above - and branching routes detractors to a "what went wrong" follow-up while promoters get a review ask.

Logic is what separates a questionnaire from an interrogation. Every respondent should feel the survey was written for them, and the mechanism is conditional routing: show, hide, skip, and branch based on previous answers.

Three patterns cover most real surveys:

• Skip logic jumps past irrelevant blocks. Someone who answers "No" to "Did you contact support?" should never see five questions about support quality. Without the skip, they will answer them anyway - and that fabricated data is worse than missing data because it looks real.
• Branching routes respondents down different paths: detractors to diagnosis questions, promoters to a testimonial ask; trial users to onboarding questions, veterans to feature-depth questions. Branches can reconverge for shared closing questions.
• Piping carries earlier answers into later wording: "You mentioned pricing was frustrating - what would have felt fair?" Piped questions read as attention, and attention earns honesty.

The discipline is to design the flows before writing questions: sketch the paths, count the questions on the longest path, and keep that worst case under your length budget. A survey that is "10 questions" on average but 25 on one branch will haemorrhage exactly the respondents on that branch.

Formspring's branching rules are configured visually per screen - no code - and combine with scoring and outcomes so a quiz-style survey can route each respondent to a different ending. The survey logic patterns post catalogues the flows worth stealing.

A survey generalises from the people who answered to the people who did not, which makes who answers more important than how many. The classic failure is non-response bias: the customers angriest and happiest with you reply at far higher rates than the silent middle, so raw averages overstate the extremes.

On sample size: statistical confidence depends on absolute responses, not percentage of population. Roughly 100 responses give you a ±10% margin; ~400 give ±5%; beyond 1,000 the returns diminish fast (Qualtrics' sample-size guidance walks through the same arithmetic). For a segment-level read (plan tier, region, persona), you need that many per segment - which is what quotas are for.

Response rate benchmarks worth calibrating against - these are practitioner ranges, consistent with what we see across Formspring surveys: email surveys to an engaged customer list run 10-30%; in-product micro-surveys 15-40%; cold outreach 1-5%. If you are far below these, the usual culprits in order: the subject line oversells the time cost, the survey is longer than promised, the ask arrives at the wrong moment, or the list is stale.

Lifting the rate is mostly honesty and timing. State the real question count and time ("4 questions, under 2 minutes") and then honour it - the single highest-leverage trick is making the survey genuinely short. Send Tuesday-Thursday mornings for B2B. One reminder to non-responders after 3-5 days typically adds 20-30% more responses in our experience; a second reminder adds little and burns goodwill. Incentives raise quantity but can skew samples toward deal-seekers - prefer donation-per-response or results-sharing for professional audiences. More tactics in the response rates playbook.

Long single-page surveys fail on mobile, and mobile is where most of your respondents are - typically 50-70% in the traffic we see. The fix is the multi-screen pattern: one question or one tight group per screen, a progress indicator, and instant advance on selection where the answer is a single tap.

The mechanics that move completion rates:

• Progress honesty. A progress bar that jumps from 20% to 90% after branching feels broken; one that crawls invites abandonment. Compute progress against the respondent's actual path, which branching makes possible.
• Momentum first. Put the easiest, most engaging question on screen one. Completion of screen one predicts completion of the whole survey better than total length does.
• Auto-advance on single-choice questions, with a visible back button. Removing one click per question compounds over 12 screens.
• Group only what belongs together. Three rating scales about the same event can share a screen; an open question always gets its own.

For surveys longer than a few minutes, partial capture and resume links change the economics. With Formspring's resume and partials support, answers persist as the respondent progresses; someone interrupted on screen 7 gets a link to continue where they left off rather than a blank restart. Partial responses are also data: a systematic pile-up of exits on one screen is a free usability report on that question.

Survey screens in Formspring are arranged in the builder with per-screen logic, and themes keep the experience on-brand without code.

A quota caps how many responses you accept from a given group, and it exists to protect sample composition from the enthusiasm of whoever shows up first. Without quotas, a survey shared in your most active community fills up with your most active users - and your "customer research" becomes a fan club transcript.

Use quotas when:

• You need balanced segments. If the decision requires comparing self-serve and enterprise customers, cap each segment at, say, 200 so one cannot drown out the other. Unbalanced cells make segment comparisons statistically meaningless.
• You are paying per response. Panel respondents and incentivised surveys keep collecting until you stop them; a quota is the stop.
• The survey has a hard analysis budget. Open-text answers that humans will read cost review time per response; capping total volume keeps the analysis honest instead of sampled-by-fatigue.

The screening pattern: ask the quota-defining question (role, plan, region) in the first screen or two, then route. Respondents in a full cell get a graceful early exit - thank them, tell them the truth ("we have enough responses from this group"), and never make them complete a survey you will discard.

In Formspring, survey quotas are set per survey with a choice of action when a cell fills - close the survey, screen out, or keep collecting but flag overflow responses for analysis. Decide the action before launch; switching rules mid-fieldwork is one of the quiet ways teams accidentally bias their own data.

Analysis starts before the first response: if you cannot name the decision each question informs, cut the question. After fieldwork closes, the workflow that holds up in practice has three passes.

Pass one: clean. Remove speeders (completion under one-third of median time), straight-liners (identical answers down a rating grid), and duplicate submitters. Expect to drop 5-10% of responses - in incentivised surveys, in our experience, sometimes 20%. Skipping this step is how confident wrong conclusions get made.

Pass two: read the closed data. Toplines first, then cross-tabs against the segments you designed for. The discipline is to distinguish statistically different from decision-relevantly different - a 3-point gap on a 100-response cell is noise, and shipping a roadmap on noise is worse than shipping on instinct.

Pass three: the open text. This is where the why lives and where manual analysis traditionally dies - reading 800 verbatims takes days and human coders drift. This is the step AI genuinely changed. Formspring's AI insights cluster open-text responses into themes, surface representative quotes per theme, flag sentiment shifts against previous waves, and produce a summary you can paste into the decision document. On paid plans the weekly digest does this continuously for always-on surveys, so an emerging complaint theme reaches you in days, not at quarter-end.

Treat AI output as a first coding pass, not a verdict: spot-check themes against raw quotes before presenting. The model accelerates reading; the judgement about what matters is still yours.

Where you send a survey determines who answers it, which makes distribution a sampling decision disguised as a logistics one.

Email to your own list is the workhorse: predictable 10-30% response rates from engaged customers, full control over segmentation, and easy reminder logic. Embed the first question directly in the email where possible - a clicked NPS score in the message body, completed on the landing page - because in our experience starting the survey inside the email roughly doubles starts.

In-product prompts catch people in context, which is ideal for CSAT and CES ("rate this support interaction") and terrible for long-form research - nobody abandons their task for 15 questions. Cap in-product surveys at 1-3 questions and rate-limit per user per month.

Shareable links and QR codes cover everything else: social posts, conference badges, packaging inserts, signatures. The trade-off is zero control over who clicks, so pair open links with screening questions and quotas. Formspring surveys ship with a hosted link by default, and branded short links with QR codes make offline distribution trackable per placement.

Channel mixing warning: response patterns differ systematically by channel - in-product respondents skew toward heavy users, email toward long-tenured ones. Tag the source on every response and check whether channels disagree before pooling them. A webhook or Google Sheets sync carrying the source field makes that a five-minute check instead of a forensic exercise.

Most survey failures are self-inflicted and repeat across teams with remarkable consistency. The shortlist, in rough order of damage:

1. No decision attached. "It would be interesting to know..." produces surveys nobody acts on and respondents who learn that answering you is pointless. Write the decision sentence first: "If X exceeds Y, we will do Z."
2. Too long. Completion drops measurably with every added minute; data quality degrades before completion does, as respondents start satisficing around the 8-10 minute mark. Under 5 minutes is the safe zone for customer surveys.
3. Leading and double-barrelled questions - covered above, still the most common wording faults in the wild.
4. Surveying instead of looking. If the answer is in your analytics, logs, or submission data, do not spend respondent goodwill asking for it. Goodwill is a budget; every survey draws it down.
5. Ignoring partials. Drop-off concentrated on one screen is a finding. Teams that never look at partial responses re-run the same broken question for years.
6. No close-the-loop. Respondents who never hear what changed respond once. A two-paragraph "you said, we did" email is the cheapest response-rate investment available - it pays out on the next survey.
7. Changing the instrument mid-wave. Rewording a question between waves silently breaks your trend line. Version your surveys; compare only like with like.

Every one of these is avoidable at design time and unfixable at analysis time - which is the strongest argument for spending one more hour on the draft.