How to report survey results back to the people who answered

Because people want to know, and almost nobody tells them. In one study of Syrian refugees, researchers asked whether participants would like to eventually see the results. "Seventy-four percent of respondents answered yes" (Innovations for Poverty Action). The same post says "sharing results is neither straightforward nor universal."

That is a research study, not a newsletter poll. Your readers may care less or more. But the principle is the same one university ethics offices use. The Australian National University's ethics committees "encourage researchers to proactively provide a summary of the project results to all participants (not upon request or in an "opt-in" manner)" (ANU).

There is a practical reason too. The next time you ask, people remember whether the last answer went anywhere. Imperial College London's guidance says that when you ask students to take a new survey you should "always mention how the previous survey results have been used" (Imperial). The sources here do not measure how much this raises later response, so don't promise yourself a number.

Four things, in this order.

1. What you asked and how many people answered. One sentence. "We asked 212 readers what nearly made them unsubscribe. 140 answered."
2. The top three counts, in plain words. Each count is a number plus what people said, such as "58 people said the emails were too long."
3. What you will change. One concrete action per count, with a date.
4. What you won't change, and why. Imperial's page says to "say what will or won't happen and why." This is the part people trust most, because it shows you read the replies and didn't only pick the easy ones.

Leave out the full data, long charts and anything that could point to one person. The ANU page warns that sharing results "can be a benefit but can also be a source of risk (social, legal, psychological etc.)", so report groups, not individuals.

Group results. Friedlander and colleagues describe one project that showed "average treatment effects, rather than impacts for specific teachers or schools, in part to protect students' privacy" (Friedlander et al.). The same rule works for a survey. "Twenty people mentioned the login screen" is safe. A quote that only one person could have written is not.

If you want to quote someone, ask first. A one-line "Can I quote your reply, without your name?" is enough. Keep quotes short and pick ones that make a count clearer, not ones that are merely vivid.

The note should fit on one screen. Here is a template with invented numbers. The figures below are not from a real survey.

You told us what got in the way. Here is what we're doing.

212 of you answered. Thank you.

1. 58 people said the weekly email was too long. From next week it stops at five items.
2. 41 people couldn't find old issues. The archive moves to the top of the site on 14 November.
3. 23 people wanted a daily version. We won't do that. It would double the work and we'd rather keep the weekly one good.

The next update comes on 1 December. Reply to this email if we got something wrong.

Every line has a count, an action and a date or a clear no. There is no "valued feedback" and no "we hear you."

This is also where a short deck helps if you need to show the same results to a team or a board. One finding per slide, the count on top, the replies behind it. The note to respondents is the same content with fewer slides.

Soon, and before the analysis is perfect. Imperial's guidance is to "send a thank you email with headline results, and initial response and details of what will happen with the results next" when the survey closes.

There is a tension. The Friedlander paper says "preliminary results can be misleading" and that the timeline should be set "not by researchers' convenience but rather by when the information can be acted on." For a small survey the answer is simple: send the headline counts once you have read every reply, and say the next update date. If your counts change later, say so in the next note.

Be honest about what you don't know. The same paper lists "being honest about uncertainties" as a first step. "We can't tell yet whether this is a few loud people or many" is a fine sentence.

Three reasonable ways to get from replies to a report-back:

• A spreadsheet. Sort the replies, label them by hand and count. It is free and you read everything. It gets slow past a hundred or so replies.
• A general AI chat. Paste the replies in and ask for the top reasons. Fast, but check the counts against the originals. Can ChatGPT analyze survey responses? covers what to check.
• A survey tool. Google Forms and SurveyMonkey collect answers, and our comparison pages say how each handles written replies.

Enform writes the survey from a description, you share one link, it reads every reply and counts what people said. Every count opens to the replies behind it, so you can check a number before you put it in the note, and it can draft a deck from the results. If you only have a handful of replies, a spreadsheet is enough. See pricing for what is available.

If the survey gave you fewer replies than you hoped, what to do with few survey responses covers how to report small numbers honestly.