The Beautiful Community | Badge of the Month | June 2026 | Experience Community
Skip to main content

The Beautiful Community | Badge of the Month | June 2026

  • June 1, 2026
  • 55 replies
  • 972 views
Show first post
This topic has been closed for replies.

55 replies

Romanoman
Level 3 ●●●
Forum|alt.badge.img+13
  • Romanoman
  • Level 3 ●●●
  • June 30, 2026

I mentioned this one workflow already a while ago, but now it has been running for few months and it really created a difference.

As we have a pretty strong focus on compliance and legal requirements when it comes to collecting feedback, we need to make sure the user make surveys compliant with our policies.

Previously, we had a workflow that notified us when a survey was created, then someone had to go, check the content manually, and see if there is nothing “sketchy”.

Now, this is fully automated and the email is only sent to us if there really is something considered “sketchy”, and all those details are also provided in the mail to simplify the manual intervention as much as possible:

Workflow overview:

  1. Survey Definition Event - Published (if any survey is published)
  2. Get Bearer Token for the service user (/oauth2/token)
  3. Get Survey Owner Name (/API/v3/users/~{qde://common.SurveyOwnerID}) - to make sure we know who to contact in case there is anything weird
  4. Get user API Token (API/v3/users/~{qde://common.SurveyOwnerID}/apitoken)- so we can call the API as that particular user and get all the survey details. For this purpose, there is a secondary workflow that makes sure any user publishing has an API token)
  5. Get All Survey Details for the Survey (/API/v3/survey-definitions/~{qde://objectId}) - from here, we get the survey questions, blocks (to look for “Thrash” block to know which questions were deleted)
  6. Code Block to extract Survey Questions that are not deleted and the Question Count to report on later
  7. AI Check on the questions - prompt that defines the AI as survey compliance checker, clarifying the non-compliant topics. It provides back the potentially problematic question with highlighted key-words/questionable text, confidence score (level of certainty this really is a problem), and explanation why it is considered problematic
    All this is provided in a defined JSON for later processing.
  8. Another code block to format the AI output into something usable in the email, and a check of embedded data, making sure we do not save something we should not save
  9. Conditional email - if there is any output from the previous steps, an email is being sent in a dedicated format to the responsible individuals, listing the survey, owner, number of questions, and the individual potential issues.

It already saved a lot of manual effort while making sure we are compliant with all internal standards and also the European data policies.

    Romanoman
    Level 3 ●●●
    Forum|alt.badge.img+13
    • Romanoman
    • Level 3 ●●●
    • June 30, 2026

    I have one AI use case / prompt to share — maybe it could be useful for others as well.

    Since we run many international surveys, translating responses and making them comparable is quite important. But translation alone is not enough. We’re very focused on employee anonymity, so we also want to make sure that responses can’t be traced back to individuals.

    This means we need to go beyond translation and also remove “linguistic fingerprints” - things like writing style or wording that could make it easy for a manager (especially in small teams) to guess who wrote a comment.

    To address this, we use a “translation and normalization” workflow as part of our surveys. The goal is to keep the original meaning, while making the response truly anonymous.

    This is the AI prompt we use in the workflow to extend an incoming response:

    Role - system (to define the AI “core”):
    You are an expert multilingual translation and normalization assistant for anonymizing open-text survey responses. Translate and normalize answers into the target language (default English) while preserving meaning, emotion, and tone, but remove identity clues (nationality, ethnicity, education, etc.) unless essential (e.g., discrimination context). Replace vulgar words with milder equivalents while keeping emotional strength.

    If any required field (survey topic, question, original answer) is missing or empty, return: {'translation': null, 'emotion': null, 'emotion_confidence': 0.0, 'tone': null, 'topics': [], 'error': 'Missing or invalid input fields.'}.

    Behavior depends on the 'normalization_mode' parameter:

    • 'authentic': retain the respondent’s natural emotional tone and informal expression, sounding like real feedback.
    • 'formal': rewrite in polished, professional English while keeping the same sentiment and meaning.

    Brevity rule: Keep the translation as concise as the original. Do not expand short answers (single words or short phrases) into full sentences. Preserve the original brevity unless additional wording is required for clarity or anonymization.

    Common language rule: Prefer the most common, simple, everyday English terms instead of abstract or overly formal ones. Translate based on typical usage rather than corporate or technical vocabulary. For example, prefer 'money' over 'compensation' unless the context explicitly requires the more formal term.

    Classify emotion as one of: very negative, negative, neutral, mixed, positive, very positive. If both positive and negative feelings exist, choose 'mixed'. Provide an emotion_confidence score (0–1). Detect tone as one of: formal, informal, sarcastic, frustrated, polite, enthusiastic, neutral.

    Always output exactly one valid JSON object with keys: translation, emotion, emotion_confidence, tone, topics, error. Validate internally that the JSON is syntactically correct before returning it.

    Individual prompt:
     

    Survey topic: Employee feedback

    Question: Question Text

    Original answer: Open Text Response

    Target language: English

    Normalization mode: formal

    ​​​​​​​
    As the output, we save not only the translated/normalized response, but also the emotion, tone, topics it mentions (for filtering), and even the token cost to be able to quantify the price.

      Tom_1842
      Level 8 ●●●●●●●●
      Forum|alt.badge.img+28
      • Tom_1842
      • Level 8 ●●●●●●●●
      • June 30, 2026

      Here is another use case I worked on recently: a QSF that ensures respondents scroll through a lengthy consent form before continuing.

      In a previous setup, we would paste the full consent information into a Text/Graphic question, but respondents could easily skip to the bottom of the page and proceed, and the full-page format can make the text feel a bit daunting. We also explored embedding an external document in an iframe, but it limited our ability to track scrolling and didn’t integrate as cleanly with the survey experience.

      In this version, the consent sits in a scrollable container and the Next button stays disabled until respondents reach the bottom of the container, which helps reduce immediate skipping and makes the content feel more manageable. For mobile users that don't have a visible scroll bar, I added a progress bar that updates as they scroll and a fade at the bottom of the container to indicate additional content.

      You can check it out using the survey link below and try it yourself with the attached QSF.

      Example Survey

        pamelalbeck
        Level 5 ●●●●●
        Forum|alt.badge.img+41
        • pamelalbeck
        • Level 5 ●●●●●
        • June 30, 2026

        Trying again to upload those QSF from my previous post...

        Pamela L. Beck
          AlonsoC
          Administrator
          Forum|alt.badge.img+29
          • AlonsoC
          • Author
          • Administrator
          • June 30, 2026

          Hello Qualtrics Community!

          Thank you to everyone who participated in our first ever tournament style Badge of the Month! The library of incredible tips, assets, and custom code you’ve created is truly incredible.

          I will now close this thread and share the final count in Friday’s newsletter. See you tomorrow for a new badge of the month!

            Terms & ConditionsAccessibility statement