---
title: "Open-Ended Survey Questions: 40 Examples, and When an AI Interview Gets Better Answers"
date: "2026-11-11"
description: "Open-ended survey questions are questions that let respondents answer in their own words instead of choosing from options you wrote, such as \"What almost stopped you from buying today?\" They capture reasons, customer vocabulary and problems you didn't anticipate, at the cost of more skipped answers and more work to…"
keywords: ["open ended survey questions", "open ended survey questions examples", "open ended questions examples", "open ended vs closed ended questions", "closed ended questions"]
author: "Perspective AI Team"
category: "AI Customer Interviews & Research"
slug: "open-ended-survey-questions-examples-ai-interviews"
excerpt: "Open-ended survey questions are questions that let respondents answer in their own words instead of choosing from options you wrote, such as \"What almost…"
image: "https://getperspective.agency/assets/99a470a6-d0ec-4cf9-8fda-565ea402fbf4"
tags: ["Customer feedback", "Research", "Customer Experience", "Improve customer satisfaction", "Build the right product", "Collect customer feedback", "Interview customers & users", "Feedback & evaluation", "Discovery & research", "Survey tools"]
relatedTemplates: ["ai-csat", "nps-survey-template", "voice-of-customer-survey", "product-feedback-survey", "customer-interview", "market-research-interview"]
lastModified: "2026-11-11"
definition: "Open-ended survey questions are questions that let respondents answer in their own words instead of choosing from options you wrote, such as \"What almost stopped you from buying today?\" They capture reasons, customer vocabulary and problems you didn't anticipate, at the cost of more skipped answers and more work to analyze."
faqs: [{"question": "What is an example of an open-ended survey question?", "answer": "\"What, if anything, almost stopped you from completing your purchase today?\" is a strong open-ended survey question. It asks about a specific recent moment, doesn't suggest an answer, and invites a reason rather than a rating. Weaker examples are vague (\"Any comments?\") or leading (\"What did you love about our new feature?\"). The strongest open-ended questions come with a follow-up, such as \"What made you decide to go ahead anyway?\""}, {"question": "What is the difference between open-ended and closed-ended questions?", "answer": "Closed-ended questions ask respondents to choose from predefined answers, while open-ended questions let them answer in their own words. Closed-ended questions, such as ratings, multiple choice and yes/no, are fast to answer and easy to count. Open-ended questions capture reasons and surprises but get skipped more often: on Pew Research Center's panel, open-ended nonresponse ran from 3% to over 50%, against 1–2% for closed-ended items."}, {"question": "How many open-ended questions should a survey include?", "answer": "Most surveys should include one to three open-ended questions, each placed directly after the closed question it explains. Every extra open-end adds effort, and Pew found that questions asking for several sentences had a median nonresponse of 17–18%. If you find yourself writing five or more open-ended questions, you're probably trying to run an interview inside a survey and should use an interview format instead."}, {"question": "Are AI follow-up questions better than open-ended survey questions?", "answer": "AI survey follow-up questions produced richer answers than fixed open-ended questions in every 2026 study here that compared the two. In the Public Opinion Quarterly experiment, embedded AI follow-ups added 57–70 words and more reasons per respondent, and only 10 of 2,253 starters failed to finish. The trade-off is time: median completion time roughly doubled, so use them where the reason justifies the extra minutes."}, {"question": "Can an AI-moderated interview replace a survey?", "answer": "An AI-moderated interview can replace a survey when your goal is understanding reasons rather than estimating percentages. For discovery, churn and win-loss research, interviews recover more distinct needs per dollar than static questions. For representative public-opinion estimates, keep a survey, because interview completers can differ from the full sample. Many teams run both: a short survey for the metric and AI interviews for the why."}]
---

## What are open-ended survey questions?

Open-ended survey questions are questions that let respondents answer in their own words instead of choosing from options you wrote, such as "What almost stopped you from buying today?" They capture reasons, customer vocabulary and problems you didn't anticipate, at the cost of more skipped answers and more work to analyze.

This guide is for CX, product and research teams who already run surveys and want to know when a text box is enough and when it isn't. It covers 40 open-ended survey question examples grouped by goal, why written answers so often come back as "fine" or "N/A," what five 2026 studies found when they compared written open-ends with AI follow-ups and AI-moderated interviews, and a decision rule for choosing between them.

**Key takeaways**

- Closed-ended questions give you the count. Open-ended questions give you the reason. Most good surveys use both.
- People skip written open-ends far more often than closed questions. On Pew Research Center's American Trends Panel, nonresponse on open-ended items ran from 3% to just over 50%, against 1–2% for closed-ended items.
- In 2026 experiments, AI follow-ups and AI-moderated interviews drew roughly 1.5x to 4.8x more words than the same questions without follow-ups, plus more distinct reasons and needs.
- Completion depends on design. AI follow-ups embedded inside a survey caused almost no dropout in one study, while sending people to a separate interview platform lost most of them in another.
- Use a written open-end for a quick reason at scale, an AI follow-up when you need the reason behind a score, and an AI-moderated interview when the "why" is the thing you're delivering.

## Open-ended vs closed-ended questions

Closed-ended questions ask people to pick from answers you wrote. Open-ended questions ask them to write their own. Closed-ended questions include yes/no items, multiple choice, rating scales like CSAT or NPS, Likert agreement scales and rankings. They're fast to answer and easy to chart, but they can only confirm options you already thought of. Open-ended questions are slower and messier, and they're the only way a survey can tell you something you didn't know to ask.

| | Closed-ended questions | Open-ended questions |
|---|---|---|
| **What it captures** | How many people chose each option | Reasons, context, language, surprises |
| **Example** | "How satisfied were you with checkout? (1–5)" | "What, if anything, made checkout harder than it should have been?" |
| **Typical skip rate** | About 1–2% on Pew's panel | 3% to just over 50% on Pew's panel, depending on the question |
| **Analysis cost** | Low: count and chart | High: read, code into themes, then count |
| **Best used for** | Tracking a metric over time, segmenting, benchmarking | Explaining a metric, discovery, finding the words customers use |
| **Main weakness** | Can't surface an option you didn't list | Thin or skipped answers, no way to ask "say more" |

The practical rule is to pair them. Ask the closed question first so you get the number, then ask one open-ended question about it so you get the reason. A satisfaction score with no reason attached tells you that something changed, but not what to fix. Our guide to [CSAT survey questions and templates](/blog/csat-survey-questions-examples-and-templates-2026) shows how to pair the two for satisfaction surveys.

## Open-ended survey question examples by goal

The best open-ended survey questions ask about a specific, recent experience, cover one topic at a time, and don't suggest an answer. "What did you think of our amazing new dashboard?" fails all three tests. "The last time you opened the dashboard, what were you trying to find?" passes. Each example below includes, in italics, the follow-up you'd ask if you could. That second question is usually where the useful answer is, and a written survey can't ask it. The same rules apply to open-ended questions in research interviews, covered in [customer interview questions that get honest answers](/blog/customer-interview-questions-2026-that-get-honest-answers).

### Customer satisfaction and CX questions

Customer satisfaction open-ended questions work best right after a score, while the experience is fresh.

1. What is the main reason for the score you just gave? — *Follow-up: "What would have had to be different for you to give one point higher?"*
2. What, if anything, almost stopped you from completing your purchase today? — *"What made you decide to go ahead anyway?"*
3. Think about the last time you contacted our support team. What happened? — *"What did you expect to happen that didn't?"*
4. What is one thing we could do to make your next visit easier? — *"Can you walk me through where it got hard last time?"*
5. What were you trying to get done when you contacted us? — *"Did you get it done? If not, what did you do instead?"*
6. How would you describe us to a friend who has never heard of us? — *"Why those words?"*
7. What, if anything, surprised you about your experience? — *"Was that a good surprise or a bad one?"*
8. If you could change one thing about our delivery, billing or checkout, what would it be? — *"How often does that come up for you?"*
9. What did you expect before you started using us, and how does it compare now? — *"Where is the biggest gap?"*
10. Is there anything else you'd like us to know? — *"Earlier you mentioned X. Can you say more?"*

Question 10 is the classic catch-all, and it's the one most likely to come back as "N/A." For more questions organized by journey stage, see [50 voice of customer questions to ask by journey stage](/blog/50-voice-of-customer-questions-to-ask-in-2026-by-journey-stage), and for the score-plus-reason pattern on NPS, see [NPS follow-up questions: how to capture the why behind the score](/blog/nps-follow-up-questions-how-to-capture-the-why-behind-the-score). The [voice of customer survey template](/templates/voice-of-customer-survey) runs questions like these as a conversation that asks the follow-up for you.

### Product and UX research questions

Product and UX open-ended questions should ask about recent behavior, not opinions about hypothetical features.

11. What were you trying to accomplish the last time you used the product? — *"What did you do right before and right after?"*
12. Tell us about the last time you got stuck. What happened? — *"How did you get unstuck?"*
13. What did you use before this product, and what made you switch? — *"What did the old way do better?"*
14. Which feature would you miss most if it disappeared tomorrow? — *"What would you use instead?"*
15. What is the most frustrating part of this workflow today? — *"Roughly how much time does that cost you each week?"*
16. What almost made you give up during setup? — *"What kept you going?"*
17. If you could change anything about the product, what would you change? — *"What problem would that solve for you?"*
18. What workaround have you built because the product doesn't do something? — *"Who else on your team relies on that workaround?"*
19. How would you explain what this product does to a colleague? — *"Which part is hardest to explain?"*
20. What would make you recommend this product to a peer without hesitation? — *"What's holding you back today?"*

Question 17 gets feature requests. The follow-up turns them into problems, which are what you can actually prioritize. More stage-by-stage prompts are in [product discovery questions for every stage](/blog/product-discovery-questions-2026-what-to-ask-every-stage), and the [product feedback survey template](/templates/product-feedback-survey) covers questions 11–20 with follow-ups built in.

### Market research and messaging questions

Market research open-ended questions should capture the buying context: what triggered the search, what else people considered, and how they describe the problem.

21. When you first realized you needed a solution like this, what was happening? — *"Why then, and not six months earlier?"*
22. What other options did you consider, and how did you compare them? — *"What was the deciding factor?"*
23. What words would you type into a search engine to find a product like this? — *"What would you expect to find?"*
24. Based on this description, what do you think the product does? — *"Which part was unclear?"*
25. What, if anything, makes you skeptical about this offer? — *"What would you need to see to believe it?"*
26. Who else is involved when you make a decision like this? — *"What does that person care about most?"*
27. What would have to be true for you to pay for this? — *"At what price would you stop and think?"*
28. What is the hardest part of this task for you right now? — *"What have you already tried to fix it?"*

### Churn and lost-customer questions

Churn and lost-customer open-ended questions should separate the headline reason from the moment of decision. "Too expensive" is usually where the story starts.

29. What was the main reason you decided to cancel? — *"When did you first start thinking about it?"*
30. What were you hoping we would do that we didn't? — *"Did you ever tell us? What happened?"*
31. What are you using instead, if anything? — *"What does it do better for you?"*
32. What could we have done to keep you? — *"Would that alone have been enough?"*
33. Was there a specific moment when you decided to leave? — *"Walk me through that day."*
34. For a lost deal: what made you choose another provider? — *"Was that settled before or after you saw our pricing?"*
35. What would bring you back? — *"How likely is that to change in the next six months?"*
36. What would you tell someone who is considering us? — *"What kind of customer are we a good fit for?"*

Cancellation surveys are where thin answers cost the most money. For more, see [churn survey questions that surface why customers really leave](/blog/customer-churn-survey-questions-that-surface-why-customers-really-leave).

### Employee feedback questions

Employee open-ended questions get more honest answers when they ask about specific, fixable things rather than general morale.

37. What is one thing that would make your work easier next week? — *"What's stopping that from happening now?"*
38. What do you wish leadership understood about your day-to-day work? — *"Can you give a recent example?"*
39. What would make you more likely to stay here for the next two years? — *"And what would make you leave?"*
40. What is one process we should stop doing? — *"What would you replace it with?"*

The [employee engagement survey template](/templates/employee-engagement-survey) turns these into a short conversation instead of a form.

## Why open-ended answers come back thin

Open-ended answers come back thin because writing a considered answer takes effort, and respondents look for the cheapest acceptable way to finish. Survey methodologist Jon Krosnick applied the term [satisficing](https://doi.org/10.1002/acp.2350050305) to this behavior: instead of searching memory and composing a full answer, respondents settle for one that's just good enough to move on. On a closed question that might mean picking the first reasonable option. On an open-ended question it means "good," "N/A," a single word, or a blank.

The skip data is consistent. Pew Research Center analyzed 92 open-ended questions on its American Trends Panel and found [nonresponse ranging from 3% to just over 50%](https://www.pewresearch.org/decoded/2021/10/14/why-do-some-open-ended-survey-questions-result-in-higher-item-nonresponse-rates-than-others/), while closed-ended items on the same panel average 1–2%. Questions asking for multiple sentences had a median nonresponse rate of 18%, compared with 13% for questions asking for a word or phrase. A [2023 Pew follow-up](https://www.pewresearch.org/decoded/2023/03/07/nonresponse-rates-on-open-ended-survey-questions-vary-by-demographic-group-other-factors/) found median nonresponse of 15% among people answering on a phone or tablet, against 11% on a desktop or laptop.

Five things make it worse:

1. **Effort.** The more a question asks for, the more people skip it, as the Pew medians show.
2. **Vague prompts.** "Any other comments?" gives people nothing specific to remember, so most have nothing to say.
3. **Small screens.** Typing a paragraph on a phone is work, which is why Pew's phone and tablet respondents skipped more often than desktop respondents.
4. **Placement.** Open-ends at the end of a long survey get whatever patience is left.
5. **No second question.** The first answer is usually a headline: "too expensive," "hard to use," "support was slow." The useful detail sits one question further down, and a form can't ask it.

The fifth problem is the one software can now address. The italic follow-ups in the examples above are what a good human interviewer asks automatically. The open question is whether an AI can ask them well enough, and at what cost to completion.

## What the research says: written open-ends vs AI follow-ups vs AI interviews

The 2026 research shows depth at a cost. AI follow-ups and AI-moderated interviews consistently get more words, more reasons and more distinct needs than written open-ends, but completion and engagement depend heavily on how the AI is delivered. The five studies below test different things, so read each row on its own terms.

| Study | Design | N | Depth result | Completion / engagement result | Caveat |
|---|---|---|---|---|---|
| [Geiecke & Jaravel, *Review of Economic Studies* (2026)](https://doi.org/10.1093/restud/rdag114) | AI-led text interview vs open text fields, random assignment; U.S. sample asked about meaning in life | 461 | 470 vs 195 words (141% more); graders judged 75% of interview transcripts more informative than matched open-text answers (2.5% the reverse) | 77% said they would not have preferred open-ended questions; "captured my views" rated 3.57 out of 4 | Expert graders scored the AI 2.93 on a 1–5 scale where 3 = an average human expert in text chat; the authors note it never matched the best human experts |
| [Deng, Liu, Toubia & Jain, arXiv (2026), pre-registered](https://arxiv.org/abs/2609.29143) | Spoken answers to the same questions: AI-moderated with follow-ups vs static (no probes) vs live human moderator; home-improvement buyers | 317 (139 AI / 154 static / 24 human) | 952 words vs 609 static; humans drew 1,404. Budgets were matched at about $5,000 per condition; at realized costs ($33.33 per AI or static interview, $163.33 per human interview) that bought 238 unique customer needs (AI) vs 188 (static) vs 109 (human), or $19 vs $27 vs $36 per need | Participants "sound more emotionally engaged with a human interviewer" | Small human arm; one category (windows and doors); AI platform supplied by a startup partner; richer transcripts didn't make digital twins more accurate |
| [Austin et al., *Public Opinion Quarterly* (2026)](https://doi.org/10.1093/poq/nfag063) | AI follow-ups (up to three) embedded inside a survey vs closed-only and fixed open follow-ups; five randomized arms on immigration and tariffs | 2,243 analyzed | 70 more words (immigration) and 57 more (tariffs) than a fixed open-ended follow-up; about 0.6–0.75 more reasons per respondent | Only 10 of 2,253 starters failed to finish, all before the AI section; satisfaction change negligible; median time rose from under 7 to 14.5 minutes | Opt-in panel; tested CloudResearch's own AI interviewing tool on CloudResearch's panel |
| Verasight report (Morris et al., August 2026) | Four written open-ends vs an AI-moderated video/audio interview on a separate platform, random assignment | 3,160 | 4.8x the words per assigned respondent (128 vs 27); 1.8x the information per assigned respondent on a length-neutral quality measure | Completion 99.4% (written) vs 40.5% (AI); completers were more AI-optimistic and demographically different from non-completers | Respondents were redirected mid-survey without warning; 16% of AI completes were fraud-flagged; Verasight sells survey panel data |
| [Valli, *Survey Practice* (2026)](https://www.surveypractice.org/article/171999-vocal-vs-typed-responses-in-ai-moderated-interviews-a-comparative-study-of-data-quality-disclosure-and-participant-experience) | Voice vs text vs hybrid answers inside AI-moderated interviews; UK participants recruited via Prolific | 252 (78 voice / 89 text / 85 hybrid) | Voice: 236% more words and 28% more thematic codes than text; hybrid in between | Completion 81% voice, 94% text, 92% hybrid; 55% would rather use text, for comfort and privacy | Author is Head of Research at Glaut, whose AI interview platform ran the study |

Four lessons hold across the table:

1. **Probing adds depth even when nothing else changes.** Deng and colleagues kept the format constant, with spoken answers in both arms, and the AI follow-ups alone added 56% more words and recovered more distinct needs per dollar. Valli's study isolates a separate effect, the answer mode. Voice adds length on top of probing.
2. **Completion is a design choice.** Follow-ups embedded in the survey flow lost almost no one in the *Public Opinion Quarterly* study. Redirecting people to a separate video interview without warning lost most of them in the Verasight report. Voice-only lost about 13 more points than text in Valli's study. These figures measure different designs, so never average them into a single "AI interview completion rate."
3. **Who finishes matters.** Verasight found that the people who completed the AI interview leaned more positive about AI than those who didn't. If you need a representative estimate of public opinion, that's a real bias. If you need to understand why a specific group of customers churned, it matters less, but check who answered.
4. **More words don't automatically mean better answers.** Humans drew more emotional engagement and still produced the most words per interview in Deng's study, and the best human experts outperformed the AI in Geiecke and Jaravel's grading. Deng and colleagues also found that "the richer conversation does not necessarily yield more accurate predictions" for their digital-twin models.

For a deeper look at how the conversations themselves work, see our explainer on [how AI-moderated interviews work, when to use them and what they replace](/blog/ai-moderated-interviews-how-they-work-when-to-use-them-and-what-they-replace).

## When to use an open-ended question, an AI follow-up, or an AI interview

Use an AI-moderated interview when the reason is the deliverable, an AI follow-up when you need the reason behind a score you already collect, and a written open-ended question when you need a quick reason code across thousands of people. The table below sorts common situations by goal, sample size, stakes and audience.

| Your situation | Best tool | Why |
|---|---|---|
| The decision depends on *why*: churn, lost deals, product-market fit, the job customers hired you for | AI-moderated interview, such as a [Perspective AI interviewer](/agents/interviewer) | Follow-ups recovered the most distinct needs per dollar in the pre-registered Deng et al. study |
| You don't yet know the likely answers (discovery, a new segment, a new market) | AI-moderated interview | You can't write closed options or targeted follow-ups for problems you haven't heard yet |
| You already run CSAT or NPS and want the reason behind every score | AI follow-up attached to the score | Embedded probes added words and reasons with almost no breakoff in the *Public Opinion Quarterly* study |
| A large tracker (thousands of responses) where you need a reason code for trend lines | Closed question plus one written open-end | Cheap to field and code, and depth per person matters less than consistency |
| You need a representative estimate of public opinion | Written open-end or embedded follow-up, with no redirect | The redirected AI arm in the Verasight report lost 59.5% of respondents, and not at random |
| Mobile-heavy audience or a sensitive topic | Text first, with voice as an option | 55% preferred text in Valli's study, and voice completion was 13 points lower |
| A dozen high-stakes interviews with senior buyers | Human moderator, or an AI interview reviewed closely by a researcher | Humans drew more emotional engagement, and the best human experts still beat AI |

If you want a shorter rule, ask two questions. Will you act on the number or on the reason? If it's the number, use a closed question with one open-end. If it's the reason, ask whether you already know the likely answers. If you do, attach an AI follow-up to the score. If you don't, run an interview. For a fuller survey vs interview comparison, see [AI vs surveys: when each method actually wins](/blog/ai-vs-surveys-when-each-method-actually-wins-in-2026), and for the CX program view, [survey-based CX measurement vs conversational VoC](/blog/survey-based-cx-measurement-vs-conversational-voc-why-the-model-is-shifting-in-2026).

Perspective AI is built for the first two rows. Its interviewer runs text or voice conversations in 57 languages, asks follow-ups on vague answers, and can be embedded inline, as a popup, slider or chat. That last point matters because of the completion lesson above: a conversation that lives where the customer already is avoids the redirect where the Verasight arm lost most of its people. Teams usually start from a template, such as [the customer interview template](/templates/customer-interview) for discovery, [the market research interview template](/templates/market-research-interview) for buying context, or [the churn interview template](/templates/churn-interview) for questions 29–36 above. If your team owns research, [Perspective for research teams](/roles/research-teams) shows how it fits an existing practice.

## How to analyze open-ended survey responses

You analyze open-ended survey responses by reading a sample, building a small set of themes, coding every response against those themes, and then counting and segmenting the themes like any other variable. A workable process:

1. **Read 50–100 responses before you code anything.** Write down the themes you see in customers' own words.
2. **Keep the codebook small.** Ten to fifteen themes plus "other" is easier to apply consistently than fifty.
3. **Code with AI, then check it.** LLM coding is fast, but in the *Public Opinion Quarterly* study the model's estimates of how common each reason was didn't match the observed prevalence. Hand-code a random sample and compare.
4. **Segment by the closed question.** Themes among detractors and promoters tell different stories, so split them before you report.
5. **Pull quotes for each theme.** A quoted customer moves a roadmap meeting more than a bar chart does.

At volume, this becomes its own discipline. Our post on [verbatim analysis for 40,000 open-ended responses](/blog/verbatim-analysis-40000-open-ended-responses) covers the large-scale version, and [voice of customer analytics](/blog/voice-of-customer-analytics) covers how coded themes feed a VoC program. Perspective AI does the first-pass synthesis automatically, with themes, summaries and extracted quotes across every conversation, so a researcher can spend their time checking and interpreting rather than tagging.

## Frequently Asked Questions

### What is an example of an open-ended survey question?

"What, if anything, almost stopped you from completing your purchase today?" is a strong open-ended survey question. It asks about a specific recent moment, doesn't suggest an answer, and invites a reason rather than a rating. Weaker examples are vague ("Any comments?") or leading ("What did you love about our new feature?"). The strongest open-ended questions come with a follow-up, such as "What made you decide to go ahead anyway?"

### What is the difference between open-ended and closed-ended questions?

Closed-ended questions ask respondents to choose from predefined answers, while open-ended questions let them answer in their own words. Closed-ended questions, such as ratings, multiple choice and yes/no, are fast to answer and easy to count. Open-ended questions capture reasons and surprises but get skipped more often: on Pew Research Center's panel, open-ended nonresponse ran from 3% to over 50%, against 1–2% for closed-ended items.

### How many open-ended questions should a survey include?

Most surveys should include one to three open-ended questions, each placed directly after the closed question it explains. Every extra open-end adds effort, and Pew found that questions asking for several sentences had a median nonresponse of 17–18%. If you find yourself writing five or more open-ended questions, you're probably trying to run an interview inside a survey and should use an interview format instead.

### Are AI follow-up questions better than open-ended survey questions?

AI survey follow-up questions produced richer answers than fixed open-ended questions in every 2026 study here that compared the two. In the *Public Opinion Quarterly* experiment, embedded AI follow-ups added 57–70 words and more reasons per respondent, and only 10 of 2,253 starters failed to finish. The trade-off is time: median completion time roughly doubled, so use them where the reason justifies the extra minutes.

### Can an AI-moderated interview replace a survey?

An AI-moderated interview can replace a survey when your goal is understanding reasons rather than estimating percentages. For discovery, churn and win-loss research, interviews recover more distinct needs per dollar than static questions. For representative public-opinion estimates, keep a survey, because interview completers can differ from the full sample. Many teams run both: a short survey for the metric and AI interviews for the why.

## Conclusion

Open-ended survey questions are still the cheapest way to attach a reason to a number. The 40 examples above will get you better answers than "Any other comments?" But the evidence from 2026 is clear about their ceiling: written open-ends get skipped, answered in a word, and never followed up. AI follow-ups and AI-moderated interviews get more depth, more reasons and more distinct needs, with completion that depends on keeping the conversation inside the respondent's flow rather than sending them somewhere else.

Use the decision rule. Count with closed questions, explain with one open-end, and when the "why" matters more than the count, have a conversation. The quickest way to try that is to attach a follow-up to a score you already collect with [the AI CSAT template](/templates/ai-csat) or [the NPS follow-up template](/templates/nps-survey-template). Then [start an AI-moderated interview](/research/new) for the questions a text box can't answer. Every plan includes 100 free credits, enough for about 10 conversations, and [pricing](/pricing) scales from there. Read 10 transcripts next to your last batch of open-text answers and judge the difference for yourself.