Deck

Typeform Guide9 min read

How to Turn Typeform Survey Responses into Product Insights

Last updated: · By Matt Teixeira, Founder and CEO

On this page
Short answer

To analyze open-ended Typeform responses with AI, export them as a CSV with each respondent's email attached, then work one question at a time: have AI label every answer against a fixed theme list, count unique respondents per theme in a spreadsheet, and split each count by the score people gave. Fix first the theme that is common among detractors and rare among promoters, not the one mentioned most. Typeform's own AI summarizes one form well. Product insights take two more steps: tying each answer to a customer, and checking each theme against your tickets and calls.

Key Takeaways

  • Typeform's Smart Insights summarizes open-ended answers into topics and sentiment, one form at a time, on higher-tier plans (Typeform Help Center).
  • Attach identity before you send. Pass email or account ID as URL parameters (formerly Hidden Fields), because you can't add it to anonymous answers later.
  • Theme each open-ended question on its own. Let AI label rows and a spreadsheet count them.
  • Rank themes by how over-represented they are among low scorers, not by raw mentions. The most-mentioned theme is often praise.
  • Deck reads Typeform's native CSV export and puts survey themes next to your tickets and calls. It has no documented live Typeform sync.

What Can Typeform Do with Open-Ended Responses on Its Own?

Typeform gives you a fast read of one form, but it isn't built to track themes across surveys or sources.

Typeform optionWhat it doesLimit for product work
Smart InsightsAI summary, topic detection, and sentiment (positive, negative, neutral, or mixed) for open-ended questions (source)Higher-tier plans only, after a minimum number of responses
Ask AIAnswers plain-language questions about your results (source)Sees only what is in Typeform
CSV or XLSX exportAll responses, or only filtered or selected ones (source)You bring the analysis
Google Sheets and the Responses APISheets syncs as results come in (source); the API returns responses as JSON (source)Raw data, no themes

What's missing is context. Smart Insights can tell you a third of this survey's comments mention reporting, but not whether support hears the same complaint or which accounts are behind it.

Step 1: Attach Customer Identity Before You Send the Survey

How well you can analyze a survey is decided when you build the form. A Typeform answer has no customer attached unless you ask for an email or pass one in. Passing it is better: use URL parameters, which Typeform used to call Hidden Fields (Typeform developer docs).

  • Add email and plan as URL parameters and let your email tool fill each link: .../to/AbC123#email=jane@acme.com&plan=pro.
  • Plan lets you split themes without a CRM join. Pass only fields you'll use.

Step 2: Put Each Open Question Right After the Score It Explains

An open answer is most useful as the reason behind a number. Ask "What's the main reason for your score?" right after an NPS question, and you can split themes by detractors, passives, and promoters, the groups Typeform's NPS question already reports (Typeform Help Center).

  • Ask about a moment. "What almost stopped you from finishing setup?" beats "Any other feedback?"
  • Freeze the wording. If the question changes between waves, you're comparing two different questions.

Step 3: Export Only the Responses You Haven't Analyzed

Filter to the dates since your last analysis, then download the CSV. With a filter on, Typeform exports only the filtered responses (Typeform Help Center). Keep the file as exported; hand-reshaping is how answers land in the wrong rows.

Step 4: Theme Each Open-Ended Question Separately

Never pool answers from different questions. "What do you love?" and "What's missing?" produce opposite sentiment by design, so pooled counts are noise.

Read 50 answers to one question yourself and draft 5 to 10 themes, each a specific problem such as "Reports take too long to load," not "Performance" (how to build a feedback taxonomy). Then give ChatGPT or Claude the CSV and this prompt.

Markdown template
You are labeling open-ended survey answers to ONE question.
Question: "What's the main reason for your score?"
Columns: response_id, email, plan, score (0-10), answer.

Themes (use only these, "Other", or "No clear reason"):
1. [Theme name]: [one-sentence definition]
2. [Theme name]: [one-sentence definition]

For every row, return CSV columns: response_id, theme_1, theme_2, quote.
- quote is the exact words that justify the label. Never paraphrase.
- Use at most two themes per answer.
- Do not count or summarize. Only label.

After the CSV, suggest up to 3 new themes from the "Other" answers.

Let AI label, and let a spreadsheet count. Models tag single answers well but miscount hundreds of rows, so count with a pivot table. Spot-check 20 labels before you trust the rest.

Step 5: Count Respondents and Split Themes by Score

Rank themes by who raised them, not how often they came up. Count unique respondents per theme, split each count by score group, and fix the theme that is over-represented among low scorers.

  • Drop empty answers such as "N/A," "good," and "nothing" first.
  • As a rule of thumb, wait for about 30 real answers per segment. A theme from fewer than five people is a lead, unless they are your largest accounts.
Sentiment Scores Mislead on Problem Questions
Ask "What should we improve?" and nearly every answer reads as negative, so the sentiment chart tells you nothing. Use the rating to measure feeling and the text to learn why.

Step 6: Check Top Themes Against Your Other Feedback

A survey theme that also shows up in tickets or sales calls beats one that lives only in the survey. Surveys capture what people say when asked; tickets capture what blocks them unprompted. Before a theme reaches the roadmap, count the accounts raising it in your help desk (how to analyze support tickets and sales calls with AI).

Worked Example: Explaining a Quarterly NPS Score

This example is illustrative. A B2B SaaS team's quarterly Typeform NPS survey passes email and plan and asks "What's the main reason for your score?" Of 600 responses, 260 give a real reason: 90 detractors, 60 passives, and 110 promoters.

ThemeRespondentsShare of detractorsShare of promoters
Support is fast and helpful642%45%
Reports take too long to load5834%5%
Setup took longer than promised4124%3%
Wants a Salesforce sync3710%13%
  1. The biggest theme is praise. Protect it, but it isn't the work.
  2. The top request isn't the driver. Salesforce sync is as common among promoters as detractors, so it doesn't explain low scores.
  3. Slow reporting is the driver. It covers a third of detractors and almost no promoters, and the help desk logged the same complaint from 25 accounts last quarter.
  4. The decision: report speed tops the backlog with ten quotes attached, and next quarter's survey keeps the same wording to measure the change.

Common Mistakes with Typeform Survey Analysis

  • Reading only the summary. Summaries flatten the details engineers need. Read 20 raw answers per top theme.
  • Ignoring "Other." If more than about 15% of answers land there, your theme list is missing something.
  • Calling one wave a trend. Call a theme rising only after it grows across two waves with the same wording.

How Do You Analyze Typeform Responses in Deck?

You upload Typeform's native CSV export unchanged, and Deck detects the format. Deck has no live Typeform sync today, so the export is the path to use. The sources Deck connects to directly are on the integrations page.

  1. Open Manage → CSV Uploads, click Upload CSV, and choose Survey (Deck docs).
  2. Set each column's role to Question, Metadata, or Skip. Keep email as Metadata so each response links to a contact.
  3. Confirm question types and check the credit counter: respondents times free-text questions. Ratings, choices, and NPS scores cost 0 credits.

Each free-text answer becomes an insight with a verbatim quote, sentiment, and a category such as Pain Points, read with its question text and linked to your existing Themes. Survey evidence then counts alongside tickets, calls, and Slack in the Opportunity Backlog.

An NPS question also feeds the NPS Score dashboard, with top themes for detractors, passives, and promoters (Deck NPS docs; see the NPS breakdown playbook). Switching survey tools keeps your history, because Deck also reads Qualtrics, SurveyMonkey, and Google Forms exports (survey import playbook).

Limitations:

  • No documented live sync. Upload each wave filtered to new responses, never the same export twice.
  • Files are capped at 10 MB. Split larger exports by rows, keeping the header in each part.
  • Deck skips low-content answers such as "N/A" or "good."
  • NPS Analytics needs Starter or above. Retried rows use credits again, and deleting an upload doesn't refund them.

By the counter's formula, a 600-respondent survey with one open question shows about 600 credits. Treat that as an estimate, since Deck bills credits on the AI processing each answer actually uses, so leave headroom: it is already more than the Free plan's 200 a month. Test your column roles with a 20 to 50 row file on the Free plan first.

Tools That Help

OptionHow it handles open textBest for
CSV plus ChatGPT or ClaudeThe labeling prompt above, counted in a spreadsheetOne survey of a few hundred answers
Qualtrics Text iQTopics, plus sentiment for each topic from very negative to very positive (source)Research teams already surveying in Qualtrics
SurveyMonkey AI analysisThematic analysis, sentiment labels, and an Analyze with AI chat (source)Teams whose surveys already live in SurveyMonkey
DeckSurvey themes next to tickets, calls, and Slack, with NPS broken down by theme; CSV upload only for TypeformProduct teams checking survey themes against every other source

If Typeform is your only feedback source, Smart Insights or the prompt above is enough. A dedicated tool earns its cost once surveys have to be read next to tickets and calls.

Frequently Asked Questions

Yes. Typeform's Smart Insights adds an AI summary, topic detection, and sentiment labels to open-ended questions once a form has enough responses, on Typeform's higher-tier plans. Both it and Ask AI work on one form at a time.

Yes, one survey at a time. Upload the CSV, give the model a fixed theme list, and ask it to label every row with an exact quote. Count the labels in a spreadsheet, because models miscount large sets of rows.

Ask for the main reason behind the score right after the NPS question, label each answer by theme, then compare each theme's share among detractors (0 to 6) with its share among promoters (9 and 10). A theme common among detractors and rare among promoters is what pulls the score down; one common in both groups is not.

Pass the respondent's email or account ID into the form link as URL parameters, formerly called Hidden Fields. Every response then carries that ID, so you can join it to your CRM or feedback tool.

Not through a documented live sync. Deck's documented path for Typeform is the native CSV export, uploaded as a Survey in Manage → CSV Uploads, so each new wave is a new upload. Deck detects the format and links responses to contacts by email.

Sources

Sources checked November 2, 2026.