How to run a workplace culture assessment people take seriously

Culture assessments get a bad reputation for a specific reason. They ask broad, agreeable questions, produce a single number, and leave nobody with anything to do. An assessment that earns attention measures a small number of named dimensions and reports them at the level where somebody can act.

This guide covers the dimensions worth measuring, the question sets for each, and how to score and compare results by team. It does not cover survey mechanics. Designing, distributing, timing, and analyzing employee surveys in general, including anonymity thresholds and channel choice, are covered properly in the guide to employee survey best practices, and this page assumes you have read it.

A workplace culture assessment measures the shared norms that govern how work gets done: what is safe to say, who gets included, whether stated values match observed behaviour, and whether people trust the organization and each other. It describes the environment. An engagement survey describes how people feel about their own job and their intent to stay in it.

The distinction matters because the two produce different actions. High engagement with weak psychological safety is a real and common pattern, and it means your team is motivated and quietly withholding bad news. An engagement score alone will not show you that.

Run both, and read the culture data as the explanation for the engagement data. Many organizations pair them by adding a culture module to the annual employee engagement surveys rather than fielding a separate instrument, which keeps response burden manageable.

Measure four dimensions rather than a general culture score, because each one fails differently and each one has a different owner. A single blended number tells you something is wrong and nothing about where.

These four cover most of what organizations actually need to diagnose:

Keep each dimension to four or six items. Fewer than four and one odd response swings the score; more than six and you are paying in response rate for precision you will not use.

One design rule holds across all four. Write items as descriptions of what happens, not as evaluations of how someone feels, because descriptive items are harder to answer aspirationally. "People here can raise problems without it being held against them" produces a more honest distribution than "I am satisfied with our culture of openness."

The sequence below is specific to culture work. It assumes the general survey mechanics are already settled, and focuses on the decisions unique to measuring norms rather than satisfaction.

Step four is the one most often skipped, and it is the one that determines whether the data is worth anything. Employees discount anonymity promises by default, so you need to make the mechanism concrete rather than reassuring. Anonymous response settings, combined with a stated threshold, let you make a promise you can actually keep while still uploading department and tenure data for segmentation.

Use a five-point agreement scale for all items, and keep the scale identical across dimensions so scores are comparable. Each block below is a usable starting set rather than a validated academic instrument, and you should cut or adapt items that do not describe recognizable situations in your organization.

These items test whether the cost of speaking up is low enough that people actually do it. Look at the distribution, not just the mean, because a bimodal result usually means one part of the team has a very different experience.

These items test membership rather than comfort. Always cut these results by segment, because belonging is the dimension where a healthy average most often conceals a struggling group.

These items test whether stated values survive contact with decisions about money, promotion, and speed. This is the dimension where leadership behaviour shows up most directly.

These items test credibility in both directions, upward toward leadership and downward toward employees. Trust items move fastest after an event, so they are the most useful to track on a short cycle.

Add two open-text questions at the end and no more. One narrow prompt, such as "describe a recent moment when this organization behaved exactly as it says it does, or exactly as it does not," produces more codeable material than a general comment box.

Score each dimension as percent favourable, which is the share of respondents choosing the top two points on the five-point scale. It is easier for managers to interpret than a mean and it does not disguise a split distribution the way an average does. Report the dimension score, the item-level scores beneath it, and the response count for every cut.

Then work through four comparisons in order:

On thresholds, apply one rule without exception: do not report any cut with fewer responses than your published minimum, and do not let a manager talk you into it. A common working default is five, but pick a number, publish it, and hold the line, because the first time a small-team result is traced back to an individual is the last time that team answers honestly.

For the open text, read themes rather than comments. Thematic analysis groups recurring topics automatically and sentiment analysis shows the emotional weight behind them, and text analysis features will surface the two or three patterns that account for most of the volume. Crosstab reports then let you check whether a theme is company-wide or concentrated in one department.

Presenting it is its own problem. Per the SurveyMonkey customer proof survey, before adopting the SurveyMonkey platform, 31% of HR professionals faced challenges when sharing or presenting insights from their survey data, which is worth planning for rather than discovering the night before the leadership review. Build the leadership view as four dimension scores with one named owner each, and keep the item-level detail in an appendix.

Distributed teams need the same four dimensions and two additional considerations. Belonging and psychological safety both degrade through a screen in ways that a co-located assessment will not detect, so the items need to test the mechanism, not just the feeling.

Add a small number of items that name the specific failure modes of distributed work:

Then cut every dimension by work pattern: fully remote, hybrid, and fully on site. A belonging gap between those three groups is the clearest evidence of proximity bias you will get, and it is invisible in a company-level average. Cut by time zone as well if your teams span more than four hours, because meeting-time burden falls unevenly and shows up in trust scores.

Be careful with response-rate differences. Remote employees often respond at a different rate than on-site colleagues, and if you do not check that, a composition effect will look like a culture finding. Compare response rates by work pattern before you interpret any gap.

Action starts with a decision about scope. Pick one dimension and no more than three teams for the first cycle, because a plan addressing all four dimensions across every department will not be executed and employees will read the non-execution as proof that the survey was theatre.

Make each commitment specific enough to verify:

Then tell people what you are not doing. A published list of the things you heard, considered, and parked, with reasons, protects your next response rate far better than silence does. The guide to closing the feedback loop covers how to structure that communication so it reads as a decision rather than a deflection.

Re-measure the specific items you committed to, not the whole instrument. A short pulse on three psychological safety items 90 days later tells you whether the intervention worked, at a fraction of the response burden of refielding the full assessment.

The hard part of culture measurement is not the first assessment, it is the second and third, when the wording has to match, the segments have to line up, and somebody has to be able to show the trend. SurveyMonkey handles that continuity through connected programmes and multi-survey analysis, with anonymous response settings and uploaded demographic data making honest segmentation possible, text analysis features doing the open-text reading, and industry and global benchmarks giving you an outside reference point for your dimension scores.

If your psychological safety scores came back weakest, the guide to building a speak up culture is the right next read, because it deals with the specific behaviours that move that dimension. And if you are also running formal feedback on individuals rather than on the environment, a 360 degree evaluation template covers that separate need without contaminating your culture data.