Every finding arrives with its count and its evidence.
We read every response, at any scale, and count each finding the same way every time. Differences are tested before we call them, serious reports are read in full against a rubric set in advance, and every claim links to the words respondents wrote, so your team can check any number in the report.
- Count and compareWho says what, with a test behind every difference.
- Change over timeWhat moved between waves, and how sure we can be.
- Severity reviewRare, serious reports kept, each with its reason.
How we count and compare
You learn who says what, and where groups agree.
Every finding is counted across all the responses, then split by group. We call a difference only when it holds up to a statistical test; otherwise the report says “about the same”. Each difference we call comes with both counts and a quote from the side that says it more.
Compare a group with everyone else
Family and caregiving: parents of young children mention it moreThe loss of flexibility makes it increasingly difficult to manage the logistical demands of childcare and my morning routine with my small children.graphic designer, under a year
Drag a group onto the findings, or select it.
Illustration: return to the office, 1,806 simulated responses, none written by a real person. Each group is compared with everyone else using Fisher’s exact test, corrected across the seven findings (Holm). The groups come from the simulated respondents’ profiles, which were inputs to the generator, so some differences exist by design.
Change over time
You see what changed between waves, and how sure we can be.
When the same question runs every quarter, each sub-theme is compared between neighbouring waves and every change is tested. A clear change is called clear, a possible one is labelled possible and not confirmed, and the rest read “no clear change”. We count each response by its main point, not by the words it mentions, because mention counts can hide a real shift.
What changed between quarters
Between Q1 and Q2Office consolidation: desks become shared (no assigned desks) on several floors.
- All 32 sub-themes: no clear change
List them
- Hot-desking, no assigned desk15 → 22
- Environmental impact of daily commuting15 → 14
- Exposure to colds, flu and other illness at the office15 → 14
- Feeling distrusted or treated like a child by the mandate15 → 14
- Fewer distractions when working from home15 → 14
- Flexible hours that match when one works best15 → 14
- Friction in hybrid meetings with some people in the room and some on screen15 → 14
- Leaders not following the policy they impose15 → 14
- Badge tracking and attendance monitoring20 → 20
- Time lost to commuting30 → 30
- Cost of commuting20 → 20
- Childcare and school schedules15 → 15
- Proximity bias in promotions15 → 15
- Loneliness and isolation when working remotely15 → 15
- Losing everyday time with family15 → 15
- Mentoring and learning from experienced colleagues15 → 15
- No clear rationale given for the return-to-office decision15 → 15
- Office equipment worse than the home setup15 → 15
- Office perks such as free lunch as incentives to come in15 → 15
- Open-office noise and constant interruptions15 → 15
- Pets that cannot be left alone all day15 → 15
- Routine and social contact at the office supporting mental health15 → 15
- Spontaneous in-person collaboration and problem solving15 → 15
- Stress and fatigue from the commute itself15 → 15
- Suspicion that the mandate exists to justify office leases15 → 15
- Time wasted on office logistics such as finding rooms and setting up equipment15 → 15
- Unequal rules where some teams are exempt15 → 15
- Blurred work-life boundaries when working at home14 → 14
- Burnout and exhaustion from the added demands14 → 14
- Caring for elderly or ill relatives14 → 14
- Considering leaving the job because of the mandate14 → 14
- Disability or medical accommodations14 → 14
Not every event shows a clear change
After desks became shared, hot-desking went from 15 to 22 responses out of 500. That’s too small to tell apart from chance at 500 a quarter, so it isn’t called a change.
Main point is hot-desking
Q1: 15. Q2: 22.
Between Q2 and Q3Badge tracking introduced: office attendance is now recorded by badge scans.
- Badge tracking and attendance monitoring20 → 60 (+40)Clear change
- 31 other sub-themes: no clear change
List them
- Loneliness and isolation when working remotely15 → 13
- Losing everyday time with family15 → 13
- Mentoring and learning from experienced colleagues15 → 13
- No clear rationale given for the return-to-office decision15 → 13
- Office equipment worse than the home setup15 → 13
- Office perks such as free lunch as incentives to come in15 → 13
- Open-office noise and constant interruptions15 → 13
- Pets that cannot be left alone all day15 → 13
- Routine and social contact at the office supporting mental health15 → 13
- Spontaneous in-person collaboration and problem solving15 → 13
- Stress and fatigue from the commute itself15 → 13
- Suspicion that the mandate exists to justify office leases15 → 13
- Time wasted on office logistics such as finding rooms and setting up equipment15 → 13
- Unequal rules where some teams are exempt15 → 13
- Environmental impact of daily commuting14 → 13
- Exposure to colds, flu and other illness at the office14 → 13
- Feeling distrusted or treated like a child by the mandate14 → 13
- Fewer distractions when working from home14 → 13
- Flexible hours that match when one works best14 → 13
- Friction in hybrid meetings with some people in the room and some on screen14 → 13
- Leaders not following the policy they impose14 → 13
- Blurred work-life boundaries when working at home14 → 13
- Burnout and exhaustion from the added demands14 → 13
- Caring for elderly or ill relatives14 → 13
- Considering leaving the job because of the mandate14 → 13
- Disability or medical accommodations14 → 13
- Time lost to commuting30 → 30
- Hot-desking, no assigned desk22 → 22
- Cost of commuting20 → 20
- Childcare and school schedules15 → 15
- Proximity bias in promotions15 → 15
Counted two ways, the same answer here
Counting mentions of distrust or surveillance shows the same clear rise. Here, the two measures agree.
Mention distrust or surveillance
Q2: 24. Q3: 53.
Main point is badge tracking
Q2: 20. Q3: 60.
Badge-tracking concerns tripled after badge scans started
Clear change 20 in Q2, 60 in Q3 (of 500 each). Holds after correcting for all 96 tests.
Honestly, it feels like I'm being watched every time I walk through the entrance just to make sure I'm sitting at my desk for the right amount of hours.
software engineer, about 3 years
I feel micromanaged by the new system that logs every time I tap my ID card at the door.
accountant, under a year
Between Q3 and Q4Hybrid schedule changed from 3 to 2 required office days per week.
- Time lost to commuting30 → 15 (−15)Possible change
- 31 other sub-themes: no clear change
List them
- Cost of commuting20 → 13
- Badge tracking and attendance monitoring60 → 55
- Hot-desking, no assigned desk22 → 23
- Loneliness and isolation when working remotely13 → 14
- Losing everyday time with family13 → 14
- Mentoring and learning from experienced colleagues13 → 14
- No clear rationale given for the return-to-office decision13 → 14
- Office equipment worse than the home setup13 → 14
- Office perks such as free lunch as incentives to come in13 → 14
- Open-office noise and constant interruptions13 → 14
- Pets that cannot be left alone all day13 → 14
- Routine and social contact at the office supporting mental health13 → 14
- Spontaneous in-person collaboration and problem solving13 → 14
- Stress and fatigue from the commute itself13 → 14
- Suspicion that the mandate exists to justify office leases13 → 14
- Time wasted on office logistics such as finding rooms and setting up equipment13 → 14
- Unequal rules where some teams are exempt13 → 14
- Environmental impact of daily commuting13 → 14
- Exposure to colds, flu and other illness at the office13 → 14
- Feeling distrusted or treated like a child by the mandate13 → 14
- Fewer distractions when working from home13 → 14
- Flexible hours that match when one works best13 → 14
- Friction in hybrid meetings with some people in the room and some on screen13 → 14
- Leaders not following the policy they impose13 → 14
- Blurred work-life boundaries when working at home13 → 14
- Burnout and exhaustion from the added demands13 → 14
- Caring for elderly or ill relatives13 → 14
- Considering leaving the job because of the mandate13 → 14
- Disability or medical accommodations13 → 14
- Childcare and school schedules15 → 15
- Proximity bias in promotions15 → 15
Mentions held steady; the main point moved
About a third of all responses mention commute time in every quarter, whatever their main point.
Mention commute time
Q3: 167. Q4: 158.
Main point is time lost to commuting
Q3: 30. Q4: 15.
Mention counts go 167 → 158, which reads as no change. Counted by each response’s main point, it went from 30 to 15: a possible change, worth watching but not confirmed.
Time lost to commuting may have fallen after office days were cut
Possible change 30 in Q3, 15 in Q4 (of 500 each). Worth watching, not confirmed at 500 responses a quarter.
I’ve found myself dreading the mandatory in-office days more than I expected, mainly because of the sheer amount of time slipping out of my day just traveling back and forth.
project manager, about 3 years
The transition back to the office has been logistically challenging, as the commute now consumes a significant portion of my working day that was previously dedicated to productive tasks.
data analyst, under a year
Pick two quarters to see what changed between them.
Illustration: return to the office, four quarterly waves of 500 simulated responses (2,000 in all), none written by a real person. Every sub-theme is compared between neighbouring quarters with Fisher’s exact test, and all 96 tests are corrected together (Holm); a possible change passes a looser bar (p < 0.10 before correction). The main points are the labels the simulation assigned, standing in for an analyst’s coding, and we knew which changes were built in before the checks were written, so this is not a blind test.
Severity review
Rare, serious reports reach you, each with its reason.
A broad search turns up every response that might be serious, and each one is read in full against a severity rubric written in advance. Of 2,001 responses, the search turned up 513; reading cleared 443, and 70 were kept for action, each with a one-line reason. Select any dot to read the response behind it and why it was kept or cleared.
- 2,001 responses
- 513 turned up by a broad search
- 443 read and cleared, including
- 70 kept: 56 at level 4, 14 at level 3
Intimidation by a manager
Level 4, high 16 of 2,001 responses (0.8%)
Rubric, level 4: A credible report of harassment or intimidation, a physical safety hazard, or a denied disability accommodation. Escalate promptly through the right channel.
I am reluctant to return on schedule because I feel compelled to avoid my manager’s office to prevent further belittling remarks.
Reason: Repeated belittling by a manager, leading them to avoid the manager's office: meets level 4 (credible report of intimidation).
I am very anxious about returning to the office because my direct supervisor uses intimidating language and frequently makes me feel unsafe through unpredictable outbursts.
Reason: A supervisor's intimidating language and unpredictable outbursts: meets level 4 (credible report of intimidation).
All 16 reasons
- qs-0154 Repeated belittling by a supervisor in front of others, so they avoid all contact: meets level 4 (credible report of intimidation).
- qs-0176 A manager's belittling comments and subtle threats about their future at the company: meets level 4 (credible report of intimidation).
- qs-0183 A manager's constant belittling and threats about their job: meets level 4 (credible report of intimidation).
- qs-0236 A manager whose intimidating, belittling behaviour makes them anxious to share a room: meets level 4 (credible report of intimidation).
- qs-0439 Repeated belittling and consistent intimidation by leadership: meets level 4 (credible report of intimidation).
- qs-0571 A manager's persistent intimidation and belittling: meets level 4 (credible report of intimidation).
- qs-0920 A manager's constant belittling and threats about job security: meets level 4 (credible report of intimidation).
- qs-1059 Repeated belittling by a manager, leading them to avoid the manager's office: meets level 4 (credible report of intimidation).
- qs-1060 A manager's frequent undermining comments, leaving them feeling intimidated: meets level 4 (credible report of intimidation).
- qs-1119 Belittling comments and veiled threats about job security: meets level 4 (credible report of intimidation).
- qs-1173 Persistent intimidation by their direct manager, with a request for HR to act: meets level 4 (credible report of intimidation).
- qs-1200 Belittling comments and vague threats from a supervisor: meets level 4 (credible report of intimidation).
- qs-1214 A supervisor's intimidating language and unpredictable outbursts: meets level 4 (credible report of intimidation).
- qs-1355 Persistent intimidation and belittlement by a supervisor: meets level 4 (credible report of intimidation).
- qs-1492 A supervisor who uses their job security as a threat: meets level 4 (credible report of intimidation).
- qs-1637 A manager whose conduct has made the office hostile and intimidating: meets level 4 (credible report of intimidation).
Select a group, or any dot, to read the response and the reason for its level.
Illustration on 2,001 simulated responses, none written by a real person. The rubric was written before the data was generated, and the review was done without the data’s answer key. Scored afterwards against what was built into the data: every response we rated level 4 was right, and 56 of the 60 built in were found; the other 4 were rated level 3, so they still reach review.
How the home page replay was made
The responses are simulated: the organizational ethics study from Katz, A., Fleming, G. C., & Main, J. (2026), GATOS: Generative AI-enabled Theme Organization and Structuring for Qualitative Research, Humanities & Social Sciences Communications. No real person wrote them; the paper’s ethics statement notes the research involves no human participants or their data.
The ideas and highlights come from our own re-run of the method’s first step (information extraction) on the 10 responses the replay shows, using the paper’s published prompt with the open-weights model gpt-oss:20b at temperature 0 (seed 7). Each highlight marks where an idea’s quoted words were found in the response: an exact match, a match ignoring case, or a close match (similarity of at least 0.9). None was placed by hand.
The counts and names (823 responses, 246 codes, 75 themes, 5.07 ideas per response on average, and the theme names) are as published in the paper. We did not re-run those steps for this figure.
The marks in the replay’s overview of every response are illustrative, not one mark per extracted idea.
Bring us the question your data has to answer.
On a scoping call we’ll say what an analysis of your interviews or open-ended responses can and can’t answer, and how your team will be able to check it, before any work starts.