To analyse open-text responses from employee engagement surveys: build a coding framework across management, career growth, culture, compensation, workload, and inclusion; apply it consistently; segment by department and tenure; and connect themes to your quantitative scores. Skimle can code hundreds of responses in minutes and filter any theme by department, tenure, or any other variable you collected.
Employee engagement surveys are one of the most common instruments HR teams run, and also one of the most commonly misread. The Likert scores go into a dashboard. The open-text responses (the part that actually explains what the numbers mean) get skimmed at best, ignored at worst.
This guide is for HR business partners, people analytics teams, and HR directors who want to change that. It covers the full process: building a coding framework, connecting open text to quantitative scores, segmenting by department and tenure, presenting to leadership, and avoiding the mistakes that make engagement survey analysis feel circular rather than useful.
Why Likert scores alone are not enough
A score of 3.8 out of 5 on "I feel valued by my manager" tells you there is a problem. It does not tell you what the problem is, who experiences it most acutely, or what a plausible solution would look like.
According to Gallup's State of the Global Workplace 2024, only 21% of employees worldwide are engaged at work. Disengagement costs the global economy $8.9 trillion per year, roughly 9% of global GDP. Those figures have been cited in board discussions and budget presentations, but they rarely change anything at the company level. The reason is that they are aggregate. A score without a story is not enough to act on.
The open-text box is where the story lives. When an employee writes "my manager never responds when I raise concerns and I've stopped trying," they are giving you the mechanism behind a low score on manager effectiveness. When someone writes "I've had two conversations with my skip-level this year, which is two more than anyone on my team," they are giving you a reference point. These are not anecdotes. Coded systematically across several hundred responses, they become the evidence base that makes engagement findings credible to a sceptical leadership team.
The challenge is not that HR teams fail to see the value in open text. The challenge is volume and time. A 500-person company with a 60% response rate generates around 300 respondents, and if each of your five survey questions has an open-text field, that is potentially 1,500 individual responses. Most HR teams read through them, note what stands out, and call that analysis. The result is a subjective impression shaped by whichever comments were most vivid on the day, not a systematic picture.
The scale challenge: why most teams read rather than code
Industry benchmarks suggest a response rate of around 60% is typical for employee engagement surveys, with 75% or above considered good. SHRM data indicates that companies with response rates above 70% are 2.3x more likely to implement meaningful workplace improvements based on survey insights, which suggests that higher response rates go hand in hand with the organisational commitment needed to act on findings.
A higher response rate is better for statistical confidence, but it also means more open-text responses to process. Teams that receive 400 or 500 open-text responses face a real analytical bottleneck if they are relying on manual reading.
The typical pattern looks like this: one or two HR analysts read through all the comments over a day or two, highlight what seems significant, and write a summary. They produce a list of themes ("people care about career development," "some managers need more training") and present those themes to the leadership team. Leadership accepts the themes because nobody has a better source of information, but they cannot act on them because "some managers" and "people" are not specific enough to address.
This is not a failure of effort. It is a structural limitation of reading-based analysis. Reading produces recall of the most striking comments. Coding produces a frequency distribution, a segment breakdown, and a connection to quantitative scores. Those are very different outputs.
If this problem sounds familiar, our guide on analysing open-ended employee survey responses covers the foundational process in detail, including how to prepare and anonymise your data before analysis begins.
Building a coding framework for engagement survey open text
A coding framework is the set of categories you use to label each response. It determines what you can measure, compare, and act on, so it is worth building it deliberately rather than improvising as you go.
Predefined thematic areas
For employee engagement surveys, most of what you will find falls into a manageable set of recurring areas. Starting with these as your framework gives you speed and year-on-year comparability:
| Theme area | What it captures | Example code labels |
|---|---|---|
| Management quality | Feedback quality, approachability, fairness, support | Manager feedback, manager visibility, manager fairness |
| Career growth | Promotion pathways, skill development, progression clarity | Promotion transparency, learning opportunities, career stagnation |
| Culture and values | Team dynamics, psychological safety, organisational norms | Team cohesion, values alignment, psychological safety |
| Compensation and benefits | Pay fairness, benefits satisfaction, equity concerns | Pay equity, benefits adequacy, pay transparency |
| Workload and wellbeing | Capacity, burnout risk, flexibility | Unsustainable workload, flexibility, burnout signals |
| Inclusion and belonging | Representation, fair treatment, voice | Voice and inclusion, belonging, bias concerns |
These six areas will cover the majority of what employees write about. The risk of starting with only a predefined framework is that you miss what is actually new or specific to your organisation. A restructure that happened three months before the survey will generate a cluster of comments that no standard framework will have a label for. The answer is to combine predefined categories with emergent coding.
Emergent coding for unexpected themes
Emergent codes are categories that arise from what you actually find in the data rather than from what you expected to find. After an initial pass through your responses using your predefined framework, look for responses that do not fit any existing category. When a cluster of responses shares a common concern that lacks a code, create one.
Common sources of emergent themes in engagement surveys:
- Reactions to a specific recent event (a leadership change, a merger announcement, a policy change)
- Concerns about a particular tool, process, or internal system
- References to an informal norm or expectation that was never articulated but is clearly shared
- Positive themes about something the organisation does well that you did not think to measure
Emergent codes give you the finding you would not have thought to look for. In Skimle's automatic thematic analysis, this discovery process happens before you start reviewing: the system surfaces the theme structure from the data, and your job is to refine it rather than construct it from scratch. For a deeper look at the inductive approach, see our guide on inductive coding and when to use it.
Deductive vs inductive vs hybrid
If you have run engagement surveys in previous years and want to track changes, a deductive approach (applying the same codes as last year) preserves comparability. If this is a first survey or you want to understand what is salient without presupposing the categories, inductive coding gives you a richer picture. Most HR teams find a hybrid approach most practical: start with a standard framework for the recurring themes, stay open to emergent codes for anything new.
For a detailed treatment of when to use each approach, see inductive, deductive, and abductive coding explained.
Connecting open-text themes to quantitative scores
The most powerful analysis comes from treating open text and quantitative data as two parts of the same picture. Each informs the other.
A standard approach is to flag all responses where the respondent also gave a low score (say, 2 or below) on a specific quantitative question, and then look at what those respondents wrote in the corresponding open-text field. If 80% of people who scored "I have opportunities to grow" at 2 or below also mentioned the word cluster around career stagnation in their open text, you have a corroboration that strengthens both findings. The qualitative explains the quantitative; the quantitative gives the qualitative a frequency count.
The reverse is also worth doing. Take every open-text response coded under a particular theme and compare the average quantitative scores of those respondents against the average for the rest of the population. If employees who mentioned psychological safety concerns score significantly lower on engagement overall, you have a signal about what is driving the headline number.
This kind of cross-analysis is covered in more depth in how to analyse employee survey results, which addresses the full quantitative side of engagement analysis.
Connecting themes to attrition risk requires one additional data element: whether respondents are among the people who later leave. If you run your survey in Q1 and track voluntary departures through the year, you can retrospectively ask whether employees who wrote certain kinds of open-text comments had a higher departure rate. Themes around "not feeling valued" and "unclear career path" tend to show up at higher rates in the pre-departure population. This connection between open text and retention data is the subject of exit interview analysis, which shares methodological elements with engagement survey work.
Demographic subgroup analysis: why it matters and how to do it
The most important thing to know about an engagement theme is not that it exists in your data, but who experiences it and where it is concentrated.
A company-wide observation that "people feel underrecognised" is interesting. The finding that this theme appears in 55% of open-text responses from engineers with three to five years of tenure, compared to 14% across the rest of the population, is something a manager can take to their team. The difference is segmentation.
What first-year employees say vs five-year employees
Employees in their first year and employees with five or more years of tenure often experience the same organisation in meaningfully different ways. First-year employees are more likely to write about onboarding, clarity of expectations, and whether they feel welcomed by their team. Employees with five or more years are more likely to write about progression, the pace of organisational change, and whether their experience is being recognised.
When both groups surface the same theme (say, "lack of transparency about decisions") the meaning is often different. A new joiner may mean they do not understand how decisions get made at all. A long-tenured employee may mean they have watched the decision-making process change over time and feel cut out of something they used to be part of. Treating these as the same finding and the same solution would be a mistake.
What one department says vs another
Department-level segmentation is often where the most actionable findings live. A theme about workload that is spread evenly across the organisation is an HR policy question. The same theme concentrated in one department is a management and resourcing question for that department's leader.
This is why metadata is critical. Your analysis is only as segmentable as the variables you collected. At minimum, every engagement survey should collect department, tenure band, seniority level, and manager (or team). These four variables let you cut most engagement findings in ways that produce department-specific action plans rather than generic recommendations.
Skimle's metadata analysis capability lets you filter any coded theme by any metadata variable and see the distribution immediately. You can ask "which departments mention workload most often?" and "does workload concern vary by seniority?" in a single session, without any additional data manipulation. The guide on discovering themes using metadata variables covers this workflow in practical detail.
For HR teams running engagement analysis as part of a broader AI interviewing approach, the Skimle Ask feature complements survey analysis by collecting deeper qualitative data through structured AI-conducted follow-up conversations. This is particularly useful when you want to explore an emergent theme in more depth without running another survey round. See HR surveys: moving from meaningless numbers to deeper insights for context on how AI interviewers fit into the broader engagement workflow.
Presenting findings to leadership
HR directors and CHROs are not looking for a list of themes. They are looking for three things: what is driving our engagement score, what does this mean for attrition and performance, and what should we do about it.
A standard engagement survey presentation structured around themes ("Theme 1: Career Development, Theme 2: Management Quality, Theme 3: Compensation") makes the audience do all the interpretive work. They have to decide which theme matters most, how the themes relate to each other, and what action each theme implies. That cognitive load means most of the content gets mentally filed under "noted, let's revisit" and nothing changes.
A more effective structure for leadership presentations is:
1. The headline finding with a number. Not "people have concerns about career development" but "career development was the most frequently mentioned theme in our engagement survey, appearing in 41% of open-text responses, and employees who mentioned it scored 1.2 points lower on overall engagement than those who did not."
2. The subgroup that matters most. "This concern is concentrated among employees in their second and third year across our product and engineering functions, where it appears in 67% of open-text responses. This is the cohort most likely to be actively exploring external options."
3. One or two verbatim quotes as evidence. Not as decoration, but as proof. The quote shows the leadership team that this is what employees actually said, not an interpretation. Select quotes that are specific and representative, not the most dramatic ones.
4. A prioritised action. "We recommend a quarterly career conversation for all employees in year two and three, with a template provided to managers. We also recommend an internal mobility announcement process so open roles are visible before they are advertised externally."
This is the structure that produces decisions. The themes are the support material, not the main content.
For a deeper treatment of how to present qualitative findings to senior audiences, presenting qualitative research findings to executives and skip-level interview analysis for HR and people leaders both address the challenge of translating qualitative evidence into leadership conversations.
4 common mistakes in engagement survey open-text analysis
Mistake 1: reading all comments without coding them
Reading produces recall. Coding produces evidence. When you read 300 comments and summarise your impressions, your summary reflects the comments that were most vivid or most emotionally salient, not necessarily the ones that were most frequent or most predictive of outcomes. Two analysts reading the same set of responses will produce noticeably different summaries. Coding eliminates that variability.
If time is the constraint, a hybrid approach works: use AI-assisted analysis to code all responses systematically, then read a sample of each coded category to verify that the groupings make sense. This takes a fraction of the time of manual reading and produces defensible, quantified findings.
Mistake 2: reporting top themes without checking the low-engagement subgroup
The most common finding from an engagement survey open-text analysis is "people care about career development and management quality." Those will be the top themes in almost every survey at every company. Reporting them without checking whether they appear disproportionately in the responses of low-engagement employees adds little value.
The meaningful analysis is: do employees who gave an overall engagement score of 2 or below mention this theme at a higher rate than employees who scored 4 or 5? If career development concerns appear at 60% frequency among low-engagement respondents and 25% frequency among high-engagement respondents, you have a potential driver. If the rates are similar across engagement levels, the theme is widespread but may not be what is depressing your scores.
Mistake 3: treating one vivid comment as representative
A single memorable comment (especially one that is emotionally charged, sharply written, or describes something dramatic) can dominate an HR team's discussion of the survey far beyond its weight in the data. The comment gets quoted in the presentation. Leaders discuss it. Action plans get built around it.
This is a known failure mode in qualitative analysis. Vivid examples are memorable; frequency is not. The antidote is simple: for any comment you want to feature or act on, check how many other responses say something similar. If it is representative of a pattern that appears across multiple respondents, that is worth showing. If it is a single outlier, note it as a potential individual concern rather than a systemic one.
Mistake 4: not connecting themes to turnover or performance data
Open-text themes on their own are descriptions of what employees feel. Connected to turnover data, they become indicators of what is driving departures. Connected to performance data, they become indicators of what is enabling or limiting productivity.
Running a retrospective linkage between engagement survey responses and subsequent departure is not always possible in a small team, but where the data exists, it is the most compelling analysis you can produce. "Employees who mentioned workload concerns in January were 2.4x more likely to have left voluntarily by December" converts an engagement finding into a business case for resourcing decisions. For a detailed look at this connection, see exit interview analysis and the related question design guide at open-ended questions: design, examples, and analysis.
Frequently asked questions
What is the best way to analyse open-ended employee survey responses?
The most reliable approach is systematic coding: apply a consistent set of categories to every response, quantify how often each theme appears, segment by department and tenure, and connect qualitative themes to quantitative scores. Reading without coding produces impressions, not findings. Tools like Skimle can apply a coding framework across hundreds of responses automatically, with every theme linked back to the supporting quotes and filterable by any metadata variable.
How long does it take to analyse open-text survey responses manually?
Manually coding 300 open-text responses (reading each one, assigning codes, tallying by segment) takes a trained analyst roughly two to three full working days. Writing up and preparing the presentation adds another day. With AI-assisted tools like Skimle, the initial coding and segmentation takes under an hour for the same dataset, with the subsequent review and refinement taking a few hours more. The total time saving is typically three to four days per survey cycle.
Should HR teams code engagement survey comments or just read them?
Code them. Reading produces a subjective impression shaped by whichever comments were most memorable on the day. Coding produces a frequency count, a segment breakdown, and an audit trail. The practical constraint is time, which is where AI-assisted coding tools make the structured approach feasible for teams that do not have a dedicated qualitative researcher. If you have to choose between reading 300 comments thoroughly and coding 300 comments with AI assistance, the coded analysis will be more defensible and more useful to leadership.
How do you present employee survey open-text findings to the leadership team?
Lead with a prioritised finding rather than a theme list. "Career stagnation is the most frequently mentioned concern among employees in their second and third year in product and engineering, and this cohort scored 1.2 points below the company average on overall engagement" is something a leader can act on. "People mentioned career development" is not. Support the headline finding with a frequency count, a subgroup analysis, and one or two representative verbatim quotes. Close with a specific, time-bound recommendation.
Ready to get real insight from your engagement survey open-text responses? Try Skimle for free and see how systematic coding, subgroup analysis, and traceability back to source quotes changes what you can present to leadership.
Want to go deeper? Read our guides on how to analyse employee survey results, exit interview analysis, and the best employee engagement survey tools in 2026.
About the authors
Henri Schildt is a Professor of Strategy at Aalto University School of Business and co-founder of Skimle. He has published over a dozen peer-reviewed articles using qualitative methods, including work in Academy of Management Journal, Organisation Science, and Strategic Management Journal. His research focuses on organisational strategy, innovation, and qualitative methodology. Google Scholar profile
Olli Salo is a former Partner at McKinsey & Company where he spent 18 years helping clients understand the markets and themselves, develop winning strategies and improve their operating models. He has done over 1000 client interviews and published over 10 articles on McKinsey.com and beyond. LinkedIn profile
Sources
- Gallup State of the Global Workplace 2024 (Gallup)
- Employee survey response rate benchmarks (Heartcount)
- SHRM research on survey response rates and workplace improvement (paraphrased from SHRM benchmarking data)
