Compare results over time

The Compare with control on the results dashboard puts each question’s results beside an earlier set, with the change. Compare one period of a survey with the period before, such as this month with last month. Or compare the survey with earlier surveys, such as this year’s engagement survey with last year’s.

Comparing periods is available on the Team plan and above. Comparing with other surveys, a longer series, the significance test and the CSV download need the Growth plan or above.

Turn it on

Open the survey and click Responses in the top navigation.
On the dashboard, find Compare with under the filter bar.
Choose what to compare with:
- Previous week, Previous month or Previous quarter.
- Previous send, shown when the survey’s recurring email campaigns have sent.
- Another survey… (Growth and above). See Compare with other surveys.

Each card that can be compared gets a panel under its chart. The line under Compare with says what is being compared. Choose Off to remove the panels.

Questions that can be compared

Question	The panel shows
NPS	The score, and the share of promoters, passives and detractors
Rating	The average and the share in the top box
Grid with one answer per row	Each row’s average
Multiple choice, image choice, yes/no	Each option’s share of the people who answered

Open-text questions have no figures to compare, but on Growth an open-text question with AI themes gets a themes panel (see What’s new in the comments). Grids that allow several answers per row and other question types have no panel.

How periods work

A week starts on Monday. Months and quarters are calendar months and quarters.
A send runs from the day a recurring campaign sent to the day before its next send. The last one runs to today.
Response rate: when you compare sends, the line under Compare with also gives each send’s response rate, for example “Response rate: 62% → 48%”. It counts the responses completed in the send’s window through that campaign, out of everyone the send went to, including people added to the audience and invited after it went out. Responses through other links, such as a web link, don’t count towards it. The rate is for the whole send, so it doesn’t change with your filters. A send that didn’t record who it went to shows no rate.
The latest period is the one with the latest answer to a compared question, not necessarily today’s. A quarterly pulse with nothing yet this quarter compares its last two quarters that have answers.
A response counts in the period it was started.
A period still running is shown and marked “(so far)”, for example “October 2026 (so far)”. Its figures can still change. The period before is not cut short to match: the figures are shares and averages, so a shorter period doesn’t bias them.

Filters and anonymity

Your other filters apply to every period. Your own date range is set aside, because each period is a whole week, month, quarter or send. The line under Compare with says so.

On an anonymous survey, each period must meet the survey’s minimum group size on its own. A period below it shows Withheld, never 0, and the others still show.

When either period has fewer than 30 answers, the panel labels the change a small sample.

Questions edited between periods

If a question was edited and published between the start of the earlier period and the end of the latest, its card says so: “Wording edited”, “Answer options changed” or “Question settings changed”, with the day.

Wording: the change is shown and tested as usual. Read it knowing the question was asked differently.
Answer options or settings (options or matrix rows added, removed or renamed, or a change of question type or scale): the change is shown but not tested, and on Growth, where changes are tested, the card says “Not tested: the question was edited during the cycles compared”, because the two periods answered different questions.

Reordering options and editing help text don’t count as edits.

Growth: series, significance and CSV

On the Growth plan and above:

Up to 12 periods. Open Last 12 cycles (or however many there are) under a panel to see the series.
A significance test on the latest change:
- Unlikely to be chance (p = …): the answers differ more than chance alone would explain.
- Could be chance (p = …): they don’t.
- Too few answers to test the change.
- Not tested: people could choose several answers, for a multiple-choice question that allows several answers.
- Not tested on weighted results, when the survey is weighted.
The test is a chi-squared test of the two periods’ answers. A grid’s rows are tested one at a time, with no correction for testing several.
Download comparison (CSV): every period and figure in one file. You need permission to export results.

Dimension scores over time

If your survey has dimensions (set up on its Team reports page, Growth and above), comparing adds a Dimensions panel under Compare with. It shows each dimension’s score in the two latest periods or surveys and the change, and on Growth, Last N cycles shows the series.

Your filters apply, including a team filter, so you can see any team’s dimension scores over time. With no filter, the panel shows the whole survey.
Highlighting: a change is coloured only when it is larger than one person’s answers could make alone: at least 5% of the scale, and more when fewer people responded. Smaller changes are shown in grey.
Other surveys are scored with this survey’s dimensions, using the same question matching as the cards. An earlier survey shows “–” for a dimension whose questions it did not ask on the same scale.
Anonymity: a period or survey below the survey’s minimum group size shows Withheld.
Restricted questions: a dimension that includes a question you can’t see is left out of the panel.

There is no significance test on dimension changes: a dimension combines each person’s answers to several questions, and the test assumes independent answers.

What’s new in the comments

On the Growth plan and above, an open-text question that has been through Text Analysis with AI themes, with no filter applied, gets a Themes panel when you compare periods. It compares the latest period with the one before:

New: themes that no analysed comment had in the period before and at least 5 have now, with their share of the analysed comments.
Growing: themes whose share rose by more than chance alone would explain, for example “Workload 15% → 50%”. Each theme’s share in the two periods is checked with a chi-squared test, corrected for testing every theme of the question at once. A theme mentioned by very few people isn’t tested.

The themes are the question’s themes in Text Analysis, so you can open any of them there. Counting them uses no AI credits.

Comments that are not in the analysis yet, usually because they came in after it ran, have no theme and are not counted. The panel says how many there are and when the analysis last ran. To include them, run the analysis again in Text Analysis (this uses AI credits, as any analysis does).

“No answer given” responses are left out of the shares. Comments under “Other / unassigned” count towards the shares but are never listed as new or growing. On an anonymous survey, each period must meet the survey’s minimum group size, as on the cards. A period below it lists no themes.

Themes are compared between surveys too: see Themes between surveys.

Summarise the changes with AI

On Growth and above, once a comparison is on screen, Summarise the changes writes a short summary of it: a few sentences on the most important changes, up to six of the largest changes with both figures, and anything you shouldn’t read too much into, such as small samples, changes that could be chance and questions edited between the two periods. It costs 10 AI credits. Asking again within an hour for the same comparison, with the same results, shows the same summary without using more credits.

How it works:

It uses only the figures in your comparison: each question’s figures and change, the answer counts, the significance test, dimension scores, new and growing comment themes, and response rates for sends. It never sees what respondents wrote, or questions you can’t see.
It describes what changed, not why. It’s instructed not to suggest causes, so read any explanation you think you see as your own, not the data’s.
Every figure is checked. A sentence that gives a figure the comparison doesn’t contain is removed before you see the summary.
Your filters apply, as they do to the comparison.
Very large surveys: the questions with the smallest changes may be left out to keep the summary short. The summary says how many.

The summary is written by AI. Check the figures on the cards before you share it.

Summaries are off for workspaces where AI features are turned off.

Compare with other surveys

Use this when each wave of a study is its own survey, for example a yearly engagement survey you duplicate each year, or a client’s wave 1 and wave 2.

In Compare with, choose Another survey….
Tick the surveys to compare with and click Compare. The list shows the surveys in this workspace whose results you can see. The likeliest earlier waves come first, marked Likely an earlier wave, and the likeliest is ticked to start with.
To change the selection, click Choose surveys.

You can compare with up to 11 other surveys. They read oldest first, by when each survey was made, and this survey is the latest.

How questions are matched

Each question of this survey is matched to one question in each other survey:

A match you set by hand.
Copied from it: a question copied from the other survey, or from the same original, matches even if it was reworded since. Copies made before 29 October 2026 don’t have this link, so their questions match by wording or by hand.
Same wording: a question of the same type with the same wording, ignoring case, punctuation and spacing.

Answer options and grid rows are matched the same way, so reordered options still line up. An option that is new in this survey shows “–” for the earlier one.

Click Match questions to see how each question matched against the survey before this one: Copied from it, Same wording, Matched by hand or No match found. If you can edit this survey, you can choose a question to match, No match, or Match automatically. A match you set is kept for everyone who compares these two surveys.

What changes when you compare surveys

Different scales: a rating or grid asked on a different scale shows both figures and no change.
Different options: a choice question whose options differ is shown but not tested: Not tested: the answers offered differ.
Reworded questions: a card says when the wording differs from the earlier survey.
Anonymity: each survey is checked against its own minimum group size. A filtered survey below it shows Withheld.
Filters: completion status, team, contact and respondent attribute filters apply to every survey. A filter on a question’s answer is carried across through the match. A filter that can’t be carried, such as one on a question with no match or on a distribution source, is left off for that survey, and the line under Compare with names the survey.
Dates: your date range is set aside. Each survey is compared whole.
Weighting: results are weighted only when every survey in the comparison is weighted.
Access: you need permission to see the results of every survey you compare with. Surveys you can’t see are not offered.

Themes between surveys

On Growth, an open-text question that both surveys have analysed in Text Analysis with AI themes gets the Themes panel when you compare surveys too. It compares this survey with the survey just before it.

The question is matched as the cards’ questions are: copied from it, the same wording, or matched by hand.
The themes: each analysis names its own themes, so they are matched in this order: a match you set by hand, then the same name, then a similar meaning. A growing theme that had another name in the earlier survey shows it, for example “Workload (was “Heavy workload”)”.
A theme with no match in the earlier survey is listed as new when at least 5 analysed comments have it.

Open Match themes under the panel to see each theme’s match and how it was found: Same name, Similar meaning, Matched by hand or No match found. If you can edit this survey, you can choose the earlier theme, No match, or Match automatically. A match you set is kept for everyone who compares these two surveys.

Matching by meaning sends only the theme names, never the comments, to our AI provider, and uses no AI credits. If AI is turned off for the workspace, themes match by name and by hand only.

In the PDF report

With Compare with on, the PDF button on the results dashboard includes the comparison you’re looking at:

A line at the top naming the two periods or surveys compared, and each send’s response rate when you compare sends.
The Dimensions table, if the survey has dimensions.
Under each comparable card, a table of the two periods or surveys with the change, and the same notes as on screen: answer counts, small samples, the significance test, wording or scale differences and edits.
New and growing comment themes, on Growth.

The PDF shows the latest two periods or surveys only. For the full series, use Download comparison (CSV).

The PDF includes only what you can see: a comparison your plan doesn’t include, or a survey whose results you can’t see, is left out, and the PDF is made without it.

Troubleshooting

Compare with isn’t on the dashboard. It needs the Team plan or above. Another survey… needs Growth or above.

Previous send isn’t offered. It appears only for surveys whose recurring email campaigns have sent. A one-off campaign has no previous send.

Only one period shows. All the answers fall in one period so far, so there is nothing to compare yet.

A question has no match in the earlier survey. Usually its wording changed and it wasn’t copied from the earlier survey. Click Match questions and choose its match by hand.