Skip to main content

Sentiment analysis

See how every customer feels at a glance, with each freeform response automatically classified by sentiment and tagged with key topics.

Sentiment analysis automatically reads every freeform response and tells you how each customer feels, so you can see at a glance whether feedback skews positive or negative without combing through each answer yourself. It runs quietly in the background on your text responses and surfaces the result right alongside each one.

Sentiment analysis is part of Research Assistant. If you don't see sentiment on your freeform responses, contact our support team to learn more about adding it to your plan.

How It Works

Every freeform response is analyzed automatically. There's nothing to switch on per question, once Research Assistant is active on your account, each new text response is classified as it comes in.

Each response is sorted into one of four sentiments:

  • Positive: the response expresses satisfaction or a good experience.

  • Negative: the response expresses dissatisfaction or a poor experience.

  • Mixed: the response contains both positive and negative elements.

  • Unsure: the sentiment is unclear, or the question doesn't lend itself to sentiment analysis.

The analysis is context-aware. It weighs the meaning of an answer against the question that was asked, rather than scanning for positive or negative words in isolation. A customer saying they wish something were better, for example, is read as negative when the current experience isn't meeting that expectation.

Alongside each classification you'll see a confidence score, shown as a percentage in the dashboard. It reflects how certain the analysis is about its read, and the lower-confidence responses are the ones most worth a quick human glance. If you ever disagree with a classification, you can change a response's sentiment manually from the response view.

Topics

Sentiment tells you how a customer feels; topics tell you what they're talking about. Alongside sentiment, each response is tagged with the key topics it mentions, short two- to three-word phrases like "shipping delays" or "easy checkout," each weighted by how prominent it is in the answer. Together they let you filter to, say, negative feedback about one specific part of your experience.

Working Across Languages

Sentiment analysis works across languages. Responses written in any language are classified the same way, so you capture how your global audience feels without translating anything first.

Exporting Sentiment and Topics

Sentiment and topics export alongside your response data. When you export a survey to CSV, each freeform question adds a Sentiment column and a Topics column next to its responses, so you can slice and pivot the results in your own tools. If you'd rather export the raw responses on their own, you can exclude sentiment and topics from the export.

Best Practices

  • Use it where feeling matters. Sentiment is most useful on open-ended questions like "What could we improve?" where the emotional read adds context to the words.

  • Filter by sentiment to triage. Sort to negative responses first to surface problems that need attention, even when only a handful of customers raised them.

  • Glance at low-confidence reads. The confidence score flags the responses where a quick human check is worthwhile.

  • Pair with Insights reports. Sentiment analysis classifies individual responses; Insights reports roll the same signal up into a summary of how sentiment is shifting over time.

Data Privacy

We take data privacy seriously. When using Sentiment analysis, your survey responses are:

  • Processed for analysis only: Responses are sent to AWS Bedrock (US region), where they're analyzed in real time by Anthropic's Claude models. They're used only to classify your feedback, not retained by the AI provider afterward.

  • Isolated per account: Your data is never mixed or compared across accounts. Even while using shared AI infrastructure, your responses remain logically separated from those of other Iterate customers.

Your survey data is not used to train any AI models, and we only transmit it to trusted processors under strict confidentiality and data-handling agreements. If you have further questions about how your data is handled, feel free to contact us.


If you have questions about sentiment analysis or how to act on your results, reach out to our support team.

Did this answer your question?