Listen
A practical guide to writing comment questions for Comment Copilot, Narrative Analysis Agent, and the Comments Report.
Open-ended comments can yield powerful insights when you ask the right questions. Every question you write is a decision about the kind of insight you’ll get back. Its wording, placement, and response field influence what employees focus on and how much detail they share. Those responses become the foundation for everything that follows, from sentiment and theme detection to Comment Copilot, Narrative Analysis Agent, and a manager reading their team’s feedback. No analysis tool, however advanced, can pull out insight the question never invited.
The right question depends on who will use the answers. When managers are the primary audience, working from the standard Comments Report, focused questions tied to a strategic priority, process, or improvement give them clear direction for local action. When senior leaders, survey administrators, or people analytics teams will analyze comments with Comment Copilot or Narrative Analysis Agent, a broader, neutrally worded question lets a wider range of perspectives emerge and helps AI surface patterns that might otherwise go unnoticed.
This guide brings together survey research, industrial-organizational psychology, and the capabilities of Perceptyx comment analytics to help you design questions with their end use in mind.
This article walks through
Start with the end in mind
Before you write a single question, get clear on how you’ll use the answers. A question is only “good” relative to the decision someone will make from it. Four planning questions will get you there:
- What tools will analyze the comments? Comment Copilot, Narrative Analysis Agent, the Comments Report, word clouds, and manual reading each surface different types of insights from open-ended data.
- What do you need to learn? A broad read on the employee experience, focused input on a strategic topic, or concrete ideas for local action.
- How will you share results? Executive summaries, theme dashboards, manager reports, or visuals like word clouds.
- Who will see the comments? Just admins and executives, or managers who are expected to act on them.
Your answers will usually point you toward one of four analysis paths. Most surveys use more than one, and that’s fine, as long as each question is written for its own path.
| Analysis path | Who typically uses it | What the question should look like | What answers it needs |
|---|---|---|---|
| Narrative Analysis Agent | Admins, creators, people analytics | Broad, neutral, about the overall experience | Rich and detailed, with examples |
| Comment Copilot | Admins, creators | Broad, neutral, about the overall experience | Rich and detailed (at least 20 characters to be included) |
| Comments Report | Managers, HR business partners, admins | Focused on a priority, process, or local action | Clear and specific; length matters less |
| Word cloud or descriptors | Executives, culture and mergers and acquisitions work | “3 to 5 words” style prompts | Short descriptors in a small text box |
A quick note on access
Comment Copilot and Narrative Analysis Agent are available for closed Point-In-Time events. Event administrators and creators can use them, and so can other admin-level users whose data access is set up for it. Managers usually work from the Comments Report, so questions meant for manager action should be written with that in mind.
How comments become insights
Before comments are summarized, they’re translated into English when needed and classified by theme, intent, and sentiment. Themes identify what employees are discussing, such as workload, recognition, or communication. Intents identify how they frame their feedback, whether they’re offering praise, sharing a preference, making a suggestion, or raising a concern. Sentiment classifies the response as positive, negative, or neutral. Comment Copilot and Narrative Analysis Agent build on that processing, so the question you ask shapes both what employees share and how their feedback is interpreted and organized for analysis.
Four considerations for designing open-ended questions
Once you know your analysis path, four design considerations shape what you’ll get back. Each section below shows how the same topic might be asked for manager-level action and for analysis with Comment Copilot or Narrative Analysis Agent. Neither version is wrong; each is written for a different job.
1. Sentiment: Use neutral framing
The question you ask shapes which experiences come to mind when employees answer (Tourangeau et al., 2000). Ask what they appreciate, and they look for positive experiences. Ask what gets in their way, and they look for barriers. That matters when you interpret the comments. It doesn’t mean employees feel negatively about their overall experience. They’re responding to the part of their experience you asked them to consider.
Comment Copilot and Narrative Analysis Agent perform best with neutrally worded questions that invite open reflection rather than steering responses toward a positive or negative tone. A neutral item gives AI the full range of experience to work with, so Copilot’s prompts can uncover a wider range of insights from a single question. Neutral wording also reduces bias and supports more honest, comprehensive feedback.
| For manager-level action | For Comment Copilot or Narrative Analysis Agent |
|---|---|
| “What improvements would you recommend to make the company a better place to work?” | “What additional comments do you have about your experience working at the company?” |
| Why it matters: The first asks only for what needs fixing, so answers lean critical and focus on suggestions that could inform manager action. That critical tone is expected; it doesn’t mean employees feel negatively overall. The second leaves the direction open, so you hear what’s working as well as what isn’t. | |
Here’s what you can expect from three common framings:
| How the question is framed | Expected sentiment | Expected intents | What tends to come up |
|---|---|---|---|
| Positive: “What do you enjoy most about working here?” | Mostly positive | Mostly Praise | Strengths and motivators like coworkers, mission, flexibility, and manager support |
| Neutral: “What additional feedback do you have about your experience?” | Mixed, often leaning critical | Spread across all five, especially Needs and Concerns and Should and Suggest | Whatever is most on people’s minds, often unresolved issues |
| Negative: “What is your greatest source of frustration at work?” | Mostly negative | Mostly Needs and Concerns, with more Angry and Unfair than other questions | Everyday basics that frustrate when missing, like tools, processes, pay, and staffing |
2. Specificity: Balance breadth with focus
Broad versus specific is a choice, not a rule. A targeted question about a specific change or process gives the direct insight a focused initiative needs and is easier to act on locally. Broad questions cover more ground and suit AI analysis at scale: a single prompt like “What additional feedback would you like to share?” can surface feedback on safety, collaboration, leadership, or communication. Leveraging filters to specific themes, intents, and sentiments can provide a wide range of insights with that single question.
| For manager-level action | For Comment Copilot or Narrative Analysis Agent |
|---|---|
| “What one thing could we do to enhance our culture of safety?” | “What additional feedback do you have about your experience?” |
| Why it matters: The first zeroes in on one topic and one priority, giving clear direction on safety. The second casts a wider net, letting you filter to safety specifically or explore anything else employees raise. | |
3. Response length: Invite the depth you need
Wording tells people how much to write, how many things to mention, and what counts as a good answer. They tend to give as much as the question seems to ask for, and when effort feels high, they write the least they can that still answers the question (Krosnick, 1991). That’s why “3 to 5 words” gets you about four words and “describe a time” gets you a story. Copilot and Narrative Analysis Agent perform best with longer, more detailed responses.
| For manager-level action | For Comment Copilot or Narrative Analysis Agent |
|---|---|
| “What 3 to 5 words describe our culture?” | “Describe how our culture affects your daily work experience.” |
| Why it matters: The first produces short descriptors that are easy to count and compare. The second produces fuller answers that explain how culture shows up day to day, and why. | |
Wording isn’t the only lever
- Text box size: bigger boxes get longer answers. A tiny box under “describe a time” sends mixed signals.
- Instructions: a short cue like “Please share a specific example” reliably adds detail. “In a few words” does the opposite.
- Character limits: these keep answers short, so use them only for “a few words” or word cloud questions.
- Device: people on phones typically write less. If many employees take surveys on mobile, save rich-answer questions for when you really need them.
Match the question to how you’ll analyze it
| If you want… | Use… |
|---|---|
| A picture or set of descriptors to compare | A “3 to 5 words” prompt with a small box; group similar words before counting |
| A clear priority for action | “ONE thing” wording; count and rank themes. Works especially well for managers |
| Broad insight at scale with Copilot or Narrative Analysis Agent | An open, neutral question with a large box and a cue to share examples |
| The “why,” or stories to share with leaders | “Describe a time” or “How does X affect” wording; expect fewer responses and protect anonymity in small groups |
4. Access: Design for the right audience
Consider who will use the comment data and at what level decisions will be made. Some organizations limit comment access to the highest levels, while others expect managers to act on feedback directly. Write each question so it produces insight meaningful to the people who will read it.
| For manager-level action | For Comment Copilot or Narrative Analysis Agent |
|---|---|
| “What one improvement could your store leader make to help you be most effective?” | “How does your store leader affect your overall experience in your role?” |
| Why it matters: The first gives one store leader specific, actionable input. The second captures how leaders shape the experience overall, which is more useful for seeing patterns across many stores. | |
Designing for your analysis tool
Comment Copilot
- Include at least one broad, neutral question so each of Copilot’s prompts returns distinct results.
- Lean on theme filters to explore specific topics like safety or leadership instead of adding more questions, especially when survey space is tight.
Narrative Analysis Agent
- The Copilot guidance above applies here too.
- Plan demographics with analysis in mind. To ask things like “How do comments about leadership vary by region?” you need clean, meaningful groups that are large enough to answer the question and protect anonymity.
- A few purposeful questions (one broad experience question plus one tied to a strategic priority) usually beat many narrow ones. Narrative Analysis Agent lets you dig deeper through conversation.
Manager-level and traditional analysis
- Managers usually work in the Comments Report without generative AI, so keep questions focused and tied to a process, priority, or specific improvement. That makes it easier to go from reading comments to a team conversation.
- Paired questions like “what’s working” and “what would help” work well here because they map neatly to “keep doing” and “change.”
Placement, number of questions, and respondent effort
When it comes to open-ended questions, fewer and better beats more. Writing takes far more effort than clicking a rating. Pew Research Center found people skipped open-ended questions about 18% of the time on average, compared with 1% to 2% for closed-ended questions, and high-effort questions went unanswered about three times as often as low-effort ones (Asare-Marfo, 2021).
- Stick to 1 to 3 open-ended questions for most engagement and pulse surveys. Each extra one lowers response rates and depth on the ones that follow.
- Make them optional. Required open-ended questions tend to get filler like “n/a” or “nothing” and can make people less candid elsewhere. Copilot skips short filler answers anyway.
- Put the broad question at the end. It catches anything you didn’t ask about and isn’t colored by one topic section.
Confidentiality and candor
People are candid when they feel safe. Employees who doubt confidentiality hold back, which steers comments toward safe, generic answers and away from the issues leaders most need to hear.
- Explain confidentiality plainly in the survey introduction and near the comment questions: who will see comments, that results are reported in groups, and that minimum reporting thresholds apply.
- Ask people not to name names. A short cue like “Please avoid naming individuals” protects coworkers and respondents. Perceptyx also automatically masks detected names before comments are sent for AI analysis, but it’s better not to collect them in the first place.
- Match questions to who will read them. If managers will read comments from small teams, avoid questions that invite very personal or identifying stories.
Getting the most from your comment data
Great survey design isn’t just about gathering data; it’s about creating a conversation between employees and their organization. When each question is written for its analysis path, Comment Copilot, Narrative Analysis Agent, and the Comments Report can surface the themes that matter most and help you plan next steps with confidence.
References
Asare-Marfo, D. (2021, October 14). Why do some open-ended survey questions result in higher item nonresponse rates than others? Pew Research Center.
Krosnick, J. A. (1991). Response strategies for coping with the cognitive demands of attitude measures in surveys. Applied Cognitive Psychology, 5(3), 213–236.
Tourangeau, R., Rips, L. J., & Rasinski, K. (2000). The psychology of survey response. Cambridge University Press.
Was this article helpful?
That’s Great!
Thank you for your feedback
Sorry! We couldn't be helpful
Thank you for your feedback
Feedback sent
We appreciate your effort and will try to fix the article