Free-form questionnaire: open questions, AI analysis

What is the free-form questionnaire?

The free-form questionnaire is the method without a predefined framework: you combine question types the way your research question requires. Available types are free text, single and multiple choice, Likert scales and sections.

That makes it two things: the tool for general feedback that doesn't fit any method — and the starting point when it isn't even clear yet what needs to be asked.

When does the free-form questionnaire fit — and when does a structured method?

The free-form questionnaire fits as long as there is no hypothesis that could be scored. Typical situations:

The metrics are good, and complaints are piling up anyway.
A new target group needs to be understood before a product is built for it.
A quick pulse check after an event, for example after onboarding.
A question that none of the structured methods is built for.

A structured method is the better choice as soon as the question is clearly defined: prioritizing improvements with JTBD ODI, understanding a switch with JTBD Switch, measuring usability with SUS or UEQ+. They deliver comparable values that a free-form questionnaire does not.

How do you write open questions that yield something?

Ask about the sequence of events, not the verdict. A verdict yields a score; a sequence of events yields the point where things go wrong.

Yields little	Yields the point
“How satisfied were you with our support?”	“Describe your most recent support request from start to finish.”
“What did you dislike?”	“What held you up last time?”
“What do you think of the new feature?”	“What did you last use the feature for — and what happened next?”

Three more rules:

Don't presuppose dissatisfaction. “What did you dislike?” pushes satisfied participants out of the analysis.
Ask satisfied participants too. If you only collect the loud voices, you conclude that everything is bad.
Closed questions for context. Role, length of use, occasion — they turn free text into comparable segments.

How are open answers analyzed?

The responses are grouped into themes, counted and given supporting quotes. There are three ways to do this:

Approach	Effort for 700 responses	Strength	Weakness
Coding by hand	15 to 25 hours	highest quality, finds the unexpected	hard to repeat
Spreadsheet with keywords	1 to 3 hours	quick to set up	only finds what you're looking for
AI clustering	minutes plus review	finds the unexpected, even across thousands	verifiable only if themes trace back to responses

What matters is the same for every approach: every theme must lead back to the raw responses. A cluster you can't expand is a claim. The detailed comparison with time estimates is in the post Analyzing open answers: by hand, with Excel or with AI (in German).

Example: the metrics measure tickets, customers experience issues

In the Customer support case study (in German), all metrics are green — 3.2 hours to first response, 94% resolution rate, 4.3 out of 5 points satisfaction. Still, complaints reach management, and the budget includes more first-level staff. 730 requesters describe their most recent request:

Theme	Responses	Share
Having to explain the issue again with every handoff	277	38%
Ticket closed although the problem remained	154	21%
First response was fast and helpful	139	19%
No status update for days	117	16%

Satisfaction drops from 4.7 without a handoff to 1.9 from three handoffs on. Every handoff creates a new ticket with good metrics — and a customer who gives up. More first-level staff is rejected, because the first response is the best-rated part of the process. Instead, the issue will be measured from now on, not the ticket.

What are the limits of the free-form questionnaire?

Exploring, not measuring. Frequencies from free text show what people mention on their own, not how strongly a theme applies across everyone.
No reference values. Without a standardized instrument, there is no benchmark and no score.
The wording carries everything. A leading question distorts the themes before the analysis begins.
Personal data in free text. What participants type in can't be fully predicted — including names and details nobody asked for.

How does the free-form questionnaire work in usabiq?

The free-form questionnaire is included in every plan, including Free Discovery, the free plan. AI clustering of open answers is part of Decision Pro — and therefore part of the 14 days every sign-up starts with.

Themes with sentiment. The AI reads every open answer, groups by theme, adds the sentiment and shows the distribution per cluster.
Traceable. Clusters can be expanded, renamed, merged and split; every assignment can be overridden individually.
Large volumes. With a very large amount of free text, the model reads a sample balanced across sources; the shares are calculated against the full number of responses, and the unassigned remainder stays visible.
Data protection before the AI call. Structured profile fields are removed, email addresses and phone numbers are masked. European providers can be selected, and a workspace switch disables all AI processing.
Repeat instead of rebuilding. As a campaign, the same questionnaire keeps running regularly — one campaign from Research Core, unlimited from Decision Pro.

Sources and further reading

Udo Kuckartz, Stefan Rädiker: Qualitative Inhaltsanalyse (2022), building categories from open answers
Norman M. Bradburn, Seymour Sudman, Brian Wansink: Asking Questions (2004), question wording and typical biases

Worked example

Questionnaire in a case study.

Software · Customer service

“How satisfied are customers with our support — and why?”

The ticket metrics look good: response times on target, high resolution rate. Yet more and more complaints go straight to management.

It is not the wait that bothers people, but explaining their issue again at every handover — 38% of open answers.

Questionnaire + AI clustering 730 requesters surveyed
Case study (in German)

Constructed example scenarios based on typical projects — no real customer data. The full case studies are currently available in German.

FAQ

Frequently asked questions about Questionnaire

How many open questions can a questionnaire handle?

Only a few. Open questions take more effort from participants than clicks. Two well-worded core questions yield more than six arbitrary ones. The Customer support case study (in German) uses two open core questions, five closed questions for context and two about the respondent's role.

Can you trust an AI analysis of open answers?

Under certain conditions: when every assignment leads back to the raw response, themes can be checked and corrected, and the unassigned remainder stays visible. The five checks for this are described in the post Can you trust an AI analysis of customer feedback? (in German)

What happens to personal data in free text?

In usabiq, structured profile fields are removed and email addresses and phone numbers are masked in free text before every AI call. Other personal data that participants type in freely cannot be fully detected technically — the survey owner remains responsible for that. If you don't want AI, you can switch it off in the workspace.

When do I move from the free-form questionnaire to a scored method?

As soon as the themes are clear and you want to measure how strong they are or what comes first. From the clusters of an exploration you can, for example, derive the outcome list for JTBD ODI.

More methods

Would another method fit your question better?

All methods compared

Run Questionnaire without doing the math yourself.

usabiq builds the questionnaire, collects responses via a link and analyzes Questionnaire. Included from Free Discovery — try Decision Pro for 14 days, no credit card required.

Send a request All methods

Free-form questionnaire: exploring before you have a hypothesis