Getting answers is the easy part. The harder part is working out what they actually tell you. A pile of 400 responses isn’t useful until you can say something like “most customers are happy, but people on our cheapest plan aren’t, and pricing is why.”
Here are the 5 steps to analyze survey data:
- Clean the data: find duplicates and junk answers, review them, and tidy up labels that mean the same thing.
- Summarize the results: count and chart the answers to each multiple-choice and rating question.
- Compare groups: see how answers differ between groups, and check that the difference isn’t just chance.
- Read the written answers: sort open-ended answers into a few themes.
- Report what you found: for each finding, show a chart, the number behind it, and how many people answered.
To keep things concrete, we’ll follow one made-up example all the way through: a customer satisfaction survey that asks people to rate us from 1 to 5, which plan they’re on, and “What could we improve?” The screenshots come from a separate PlatoForms Insights sample report filled with generated data, so their numbers won’t match the example.
On this page
One quick note on words. A survey is the whole project: deciding who to ask, collecting answers, and making sense of them. The questionnaire is just the list of questions people fill in. (There’s more on this in survey vs questionnaire.) If your answers live in Google Forms, our guide to analyzing Google Forms responses covers the Google-specific clicks.
1. Clean and prepare the data
Bad answers in means bad conclusions out, so look through the responses before you count anything. Watch for:
- Duplicates: the same person answering twice. A repeated email address is usually the giveaway.
- Speeders: people who finished so fast they can’t have read the questions.
- Straight-liners: people who picked the same spot question after question, for example always the first option.
- Answers that mean the same thing: “UK”, “U.K.” and “United Kingdom” should count as one answer, not three.
- Skipped questions: decide once whether a blank counts as “no answer” or is left out, and stick to it.
- Questions where people can pick several options: the percentages can add up to more than 100%. That’s fine, as long as you describe each one as “the share of people who picked it.”
Don’t delete doubtful answers straight away. Mark them first, so you can check whether they change the results.
In PlatoForms: the Data quality section of the Insights report does the first pass for you. It flags speeders (people who finished in under 20 seconds), straight-liners (the same option position across three or more choice questions), and duplicate emails (if the form has an email field). Flagged answers aren’t removed, they still count in every chart, and you decide what to do with them. To merge answers that mean the same thing, open Clean up categories on a chart. This only changes how answers are grouped in the report, not the submissions themselves (how it works).
2. Summarize the results
Now go through the closed-ended questions one at a time, the ones where people picked from a list or a scale.
- Multiple-choice questions: give the count and the percentage for each option. A simple bar chart is usually all you need.
- Rating questions: look at the median (the middle answer) and how spread out the answers are, not just the average. An average of 3.5 could mean everyone is lukewarm, or that half your customers love you and half don’t.
- Agree/disagree scales (Likert scales): a handy single number is the share of people who picked the top option (“top-box”) or the top two (“top-two-box”), such as “Agree” plus “Strongly agree.” Our Likert scale guide shows how to set these questions up.
In our example: 412 customers answered. The median rating is 4 out of 5, and 68% gave a 4 or 5. That 68% is the top-two-box score.
In PlatoForms: you don’t have to build these charts. Every form gets a default Insights report with a chart of submissions over time and a chart for each field it can report on. Choice questions get answer charts, and rating, slider, and number fields can show a statistics card with the median, the mean, and a plain-English label for how spread out the answers are. On the Free plan, a report can hold up to 5 charts (more on charts).
3. Compare groups
The overall number often hides the interesting part. Split each important question by a group, such as plan, team, region, or new vs returning customers. A table that crosses one question with another like this is called a cross-tab.
Before you call something a real difference, check two things:
- Is the group big enough? If only 8 people are on a plan, a gap between them and everyone else could easily be luck. Always say how many people are in each group.
- Does a test back it up? The usual check for this kind of data is a chi-square test, which tells you whether two answers are related or the gap could be chance.
In our example: people on the Starter plan seem less happy than people on Pro or Business. We’d only report that as a real difference if each group is big enough and the test agrees.
In PlatoForms: on a paid plan, you can break a chart down by another question and switch it to a cross-tab, showing row percentages or counts. Once there’s enough data (at least 30 responses), a label based on a chi-square test tells you Meaningful difference or No clear difference. To look at one group on its own, click a value on any chart that shows answer counts and choose Filter report to this value (charts and comparisons).
4. Analyze open-ended answers
Written answers can help explain why people gave the scores they did. They also take the longest to go through. The usual approach:
- Read a sample first, so your themes come from what people actually wrote, not from what you expected.
- Make a short list of themes, around 5 to 10, plus “Other.”
- Put each answer under a theme (researchers call this “coding”), and note whether it’s positive or negative.
- Count how often each theme comes up, and save one or two quotes per theme for your report.
In our example: the themes that come up most in “What could we improve?” are pricing, onboarding, and support.
In PlatoForms: a word cloud shows which words come up again and again. To go further, Analyze with AI suggests themes based on the answers. You review them, edit them, and approve them, and only then does it put every answer under a theme and tag its sentiment. AI can also write a short summary with a few example quotes. These AI features use your team’s monthly AI allowance (AI in Insights).
5. Visualize and report
Most people will only skim your report, so make every finding easy to grab:
- Say the finding first, then the number: “Starter customers are less happy: 41% gave a 4 or 5, compared with 78% on Pro.”
- Always say how many people answered, overall and in each group.
- Use the simplest chart that makes the point: bars to compare options, lines to show change over time.
- Compare against something: last quarter, another group, or a goal. “68%” means more next to “61% last quarter.”
- Add a real quote to show what’s behind the number.
In PlatoForms: turn on Compare to previous period and time charts show the last period as a dashed line, with labels such as New top answer when something changes. You can share a read-only link to the report on a paid plan, download it as a PDF, or download the chart totals as a CSV. And if you’d rather just ask, Ask your data answers questions about the report in plain language and includes the supporting number for each claim (sharing and exports).
Why people stop before the end
Your results only include the people who finished. If lots of people start the survey and give up halfway, the ones who finished might not speak for everyone, so it helps to know where people drop off. For common reasons and fixes, see why people abandon forms, and for keeping surveys with branches short, see our branching survey design guide.
In PlatoForms: the visitor journey in Insights shows how many people opened, started, and submitted your form. Once the form has had at least 100 visitors, it highlights the question that loses the biggest share of people who started. A long or confusing question is one possible reason, but look at that question and the ones around it before deciding. Forms published before Insights launched need to be published again to collect this question-by-question data (visitor journey).
FAQ
How do you analyze data from a survey?
Clean the responses first, then summarize each closed-ended question with counts and percentages, compare groups with cross-tabs, group open-ended answers into themes, and report each finding with the number and sample size behind it.
How do you summarize the results of a survey?
Lead with the main finding, then give the numbers behind it: how many people answered, the percentage for each key answer, and how groups differ. Add one or two representative quotes from the open-ended answers to explain the numbers.
How do you interpret survey results?
Compare each result with something: a previous survey, another group, or a target. Check the sample size before drawing conclusions, especially for small groups, and use the open-ended answers to understand why people answered the way they did.
Can ChatGPT analyze survey data?
AI assistants can help summarize exported responses and suggest themes, but check every number against the data and think about privacy before sharing responses with a third-party tool. In PlatoForms Insights, Ask your data reads the report’s aggregated numbers and cites the number behind each claim, and for themes you approve the AI’s proposals before any answer is classified.
