Every call your team handles tells a story about how your customers feel. dialnote automatically analyzes the sentiment of every call — both AI-handled and human-handled — so you can spot trends, catch problems early, and coach your team without listening to hours of recordings.
How It Works#
dialnote assigns one of three sentiment values to each call:
- Positive 😊 — The caller was happy, satisfied, or had a good experience
- Neutral 😐 — The conversation was matter-of-fact, neither positive nor negative
- Negative 😟 — The caller was frustrated, upset, or had a bad experience
The way sentiment gets analyzed depends on the call type:
AI voice agent calls are analyzed by the AI engine (Retell) that powers your agents. After each call wraps up, Retell evaluates the full conversation and sends back a sentiment score automatically. There's nothing to configure — it just works.
Human-handled calls go through dialnote's built-in LLM evaluation. Once a call's transcript is ready, the system runs it through a language model that reads the entire conversation and determines the caller's overall sentiment. This runs alongside call scoring, so you get both a numeric score (1–10) and a sentiment label from the same analysis.
Requires AI Transcription
Sentiment analysis for human-handled calls only works when AI transcription and AI call tagging are turned on for that phone number, and your team has at least one call tag set up. Go to Settings → Phone Numbers → [your number] → AI & Intelligence to check these toggles. Both are on by default for new numbers. AI agent calls always include sentiment regardless of these settings.
Viewing Sentiment Data#
You'll find sentiment data in a few places across dialnote:
Analytics Dashboard#
Go to Reports in the sidebar and click View Detailed Report on the AI Voice Agent card to see the Sentiment Analysis chart. It shows your total calls, then one horizontal bar each for positive (green), neutral (gray), and negative (red), with the percentage and call count. It counts AI agent calls only. This gives you a quick read on overall customer satisfaction across your AI agents.
The Call Insights section on the same page uses sentiment data to generate actionable recommendations. If more than 20% of calls show negative sentiment, dialnote flags it as an opportunity and suggests reviewing common pain points. When positive sentiment exceeds 60%, you'll see a success indicator confirming strong customer satisfaction.
Individual Call Records#
Each call record in the AI calls table includes a sentiment indicator. You can filter the table by sentiment to quickly find all negative calls that might need follow-up, or all positive ones you can use as training examples.
For human-handled calls, sentiment shows as a chip on the call in your Inbox. The Call Tags report (View Detailed Report on the Call Tags & FCR card in Reports) also has a sentiment column and filter covering both human and AI calls.
Focus on negative calls first
Use the Sentiment filter on the AI calls table or the Call Tags report to show Negative calls and quickly find the ones that need attention. These are your best opportunities to improve agent instructions, update knowledge bases, or identify gaps in your call handling.
Using Sentiment to Improve#
Sentiment data is most useful when you track it over time and act on patterns:
For AI agents: If you notice a spike in negative sentiment, check the transcripts of those calls. Common causes include missing information in your knowledge base, confusing agent instructions, or scenarios your agent isn't set up to handle. Update your agent's instructions or add a new tool to address the gap.
For your team: Negative sentiment trends on specific phone numbers can point to training opportunities. Pair this with call scores to get the full picture.
For reporting: Click Export CSV on the AI calls table to download your AI calls with a sentiment column and share trends with your team. The cost savings metric also ties into sentiment indirectly — calls that are fully contained by AI with positive sentiment are your highest-value automation wins.
Short calls may not get sentiment
Human-handled calls only get sentiment when there's a transcript to analyze. If transcription is turned off for the phone number, or no speech was captured, the call won't have a sentiment value in reports. There's no minimum word count — even a short transcript with any text is analyzed. (Note: full call summaries do require at least 50 words, but sentiment does not.)
Sentiment and Call Scoring#
Sentiment and call scoring work together but measure different things. Call scoring rates how well the call was handled on a 1–10 scale: was the issue resolved, and was the caller satisfied? Sentiment captures how the caller felt about the interaction.
The two usually line up: a high score (8–10) means the issue was resolved and the caller was satisfied, and a low score (1–4) means it wasn't resolved and the caller left unhappy. A mid-range score often points to a partial fix, so sentiment helps you see how the caller took it.
Both metrics are generated from the same transcript analysis for human calls, so there's no extra setup required. Check the Call Scoring docs for more details on how numeric scores work.
What's Next#
- Call Scoring — the 1–10 numeric score that runs alongside sentiment
- Call Transcription — turn on the transcripts that power sentiment
- AI Agent Analytics — where the Sentiment Analysis chart and Call Insights live
- Testing Your Agent — run a test call before customers hear it