Response Quality Check

內文：

"Response Quality Check" can automatically analyze survey responses, helping you quickly identify low-quality answers and improve the accuracy and efficiency of your data analysis. The system classifies each response as "Valid responses" or "Suspicious responses", allowing you to easily understand the overall health and reliability of your data.

The "Response Quality Check" feature is available on the "Data Report" page of your survey results. Using AI technology, it automatically evaluates responses to text questions and classifies them into two categories based on the quality indicators you enable:

• Valid responses: Normal responses with no suspicious signs
• Suspicious responses: Responses with abnormal or questionable text content; further review is recommended

How does AI determine whether a response is suspicious?

The AI fully analyzes the content of each response and checks whether it meets each quality indicator (such as Offensive text, Overly short text, etc.). It then uses the indicators you have enabled to determine the quality classification:

• If you enable the "Offensive text" indicator, responses containing offensive content will be marked as "Suspicious responses"
• If you later disable "Offensive text" and enable "Overly short text", the system will recalculate the quality classification for all responses
• When adjusting quality indicators, the system only reclassifies responses and does not perform AI analysis again

Start Response Quality Check

Go to the "Data Report" page of your survey results and click the "Response quality" button in the top-right corner.

In the sidebar, click "Start checking" to enable Response Quality Check.

In the "Quality indicator settings" window, enable the suspicious-response indicators you want to check. There are currently four quality indicators: "Offensive text", "Overly short text", "Nonsensical text", and "Copied text". You may freely choose which indicators to enable.

After clicking "Apply", the system will begin analyzing survey responses based on the selected indicators. During analysis, the sidebar will show the status "Checking quality...", and the following actions will be disabled:

• Date filter
• Filter
• Export as
• Share responses
• Display formatting adjustments

The analysis will take some time. You may leave the page and return later to see whether it has finished. Once complete, a message "Response quality analysis completed." will appear, and the quality check results will be displayed automatically.

If you need to adjust the indicators later, click "Quality check settings" in the Response Quality Check window. Turn indicators on or off as needed, then click "Apply" to save.

When adjusting quality indicators, the system only reclassifies responses and does not perform AI analysis again. Therefore, it will not consume additional quality-check quota.

Quality Check Results

After AI analysis, the Response Quality Check sidebar will display:

• Quality score: Color-coded based on the proportion of high-quality responses (Valid responses / analyzed responses)
• High-quality ratio > 80%: Green
• 60% < High-quality ratio ≤ 80%: Yellow
• High-quality ratio ≤ 40%: Red
• Valid responses: Number of valid responses
• Suspicious responses: Number of suspicious responses
• Quality check results: Number of suspicious responses detected for each indicator (only indicators you enabled are shown)

Once quality checking is enabled, the Data Report will include two additional fields:

• Response quality: Displays the quality classification for each response
• Valid responses
• Suspicious responses
• -: Not analyzed or no text response provided
• Quality check results: Shows which indicators were triggered for a suspicious response
• Only enabled indicators will be displayed
• -: No indicators triggered
• Quality check failed: Character limit exceeded or AI error

Reviewing Suspicious Responses

After reviewing the overall quality results, you can further examine suspicious responses to support your analysis.

Click "Suspicious responses" or any quality indicator in the sidebar to view flagged responses in the review window.

For selected responses, you may perform the following actions:

• Change quality: Manually reclassify the response as "Valid responses" or "Suspicious responses"
• Isolate: Move the response to the isolation area
• Delete: Permanently delete the response

After review, both the Data Report and Response Quality Check window will update accordingly.

Quality Check Quota

The quality-check quota refers to the "total number of responses analyzed during the month". Responses are counted only after AI analysis completes, and the quota covers all surveys in the account—not per individual survey.

If a response contains no text questions or the respondent left all text questions blank, quality checking cannot be performed. However, such responses still count toward your quota. To avoid wasting quota, we recommend enabling this feature only for surveys with text answers.

When the monthly quota limit is reached:

• Responses beyond the limit will not be analyzed and will display -
• The sidebar will show: "Quality check response limit reached. {response} responses cannot be checked."
• You may wait for the quota to reset next month or select "Upgrade now" to increase your limit.

After the monthly quota resets or you upgrade to obtain more quota, the sidebar will display:

• "You have {response} responses pending quality check."
• A "Starting checking" button. Clicking it continues analysis for unchecked responses.

Downgrade impact:

• Existing analysis results remain available
• Indicator settings are retained
• New responses cannot be analyzed until quota resets or you upgrade again

Notes:

• Quality-check quota limits by plan:

• Quotas reset monthly
• Enterprise quotas apply at the organization level, not per user
• Responses with "Quality check failed" are not included in total usage

FAQ

• Why do some responses show "Not analyzed"?

Possible reasons include:

• Quality-check quota has been reached
• Total character count exceeds 20,000
• The survey contains no text questions or no text answers were provided
• If I modify survey questions, will it affect analysis results?

Yes, but the system handles this automatically:

• Deleting non-text questions: No impact on quality check results
• Deleting text questions:
• If deleted questions previously triggered issues (e.g., Offensive text, Overly short text), the system will automatically recalculate classifications after you publish the survey
• When you return to the Data Report page, you may see "Checking quality..." while recalculation occurs
• Once completed, the system displays updated results based on remaining questions
• Recalculation reuses original analysis results and does not consume additional quota

Example:

A response was originally flagged for "Offensive text" in Q1 and "Overly short text" in Q2, so it was classified as "Suspicious responses".

• If you delete Q1 and publish:
• The system recalculates
• Q2 still has issues → The response remains "Suspicious responses"
• Quality check results will now show only "Overly short text"
• If you delete both Q1 and Q2:
• The system recalculates
• If Q3 has no issues, the response becomes "Valid responses"
• Adding text questions:
• If you have remaining quota, AI will automatically analyze new text content
• If I manually change the quality classification, will adjusting indicators affect the classification?

Yes. After manually changing a response's classification, if you later enable or disable indicators, the system will recalculate the classification based on the new indicator settings.

Example:

• "Offensive text" and "Overly short text" are enabled
• A response is flagged for "Offensive text"
• You manually mark it as "Valid responses"
• Later, you enable "Nonsensical text" and "Copied text"
• If the response meets any newly enabled indicators, it will be reclassified as "Suspicious responses"

Recommended approach: Finalize your indicator settings before performing manual adjustments to avoid repeated rework.

• Why can't I use filtering, exporting, and similar functions?

When the system is analyzing responses, certain features are disabled to prevent data inconsistency. Once analysis completes, all features become available again.