Beyond sentiment: reading tones, emotions and tensions

Sentiment analysis sorts posts into positive, negative or neutral. That hides most of what matters. Tones describe how people speak, emotions describe what they feel, and tensions describe the conflicts they are trying to resolve. Reading all three turns a flat sentiment line into a clear picture of what your audience needs and why.

What is wrong with sentiment scores?

Sentiment was a sensible start for social listening. It is cheap, easy to chart and roughly right on average. But averages hide the things marketers need most. "Negative" covers mild irritation and genuine fury. "Positive" covers polite approval and fierce loyalty. "Neutral" is often where the most useful questions live, the people asking for help before they buy.

Sentiment also struggles with how people actually talk online: sarcasm, irony, affectionate complaint ("I hate how much I love this"), and the dry humour common in many Southeast Asian online spaces. A model trained to find polarity reads all of these badly.

Somonitor's position is simple: sentiment tells you the weather; tones, emotions and tensions tell you the climate.

What are tones, emotions and tensions?

Definitions

• Tone: how something is said. Playful, sarcastic, earnest, weary, authoritative.
• Emotion: what the writer feels. Pride, anxiety, delight, frustration, nostalgia, disappointment.
• Tension: a conflict the writer is caught in. "I want to treat my kids, but I worry about sugar." "I want the cheapest option, but I do not trust cheap."
• Concept: any of the above, tagged consistently across posts and linked to evidence.

SOMIN, an AI audience-research platform, built its concept engine to decode exactly these layers from real posts. Somonitor uses that engine as its listening layer. You can see how SOMIN describes the difference from conventional social tools on its comparison page.

Why are tensions the most valuable signal?

Tones and emotions describe the conversation. Tensions explain it. A tension is a reason to act: people caught between two things they want are looking for a way through. Brands that resolve a real tension earn attention without shouting.

Consider a few tensions we hear often in marketing conversations themselves: wanting to use AI while fearing it makes everyone sound the same; wanting proof of results while distrusting vanity metrics; wanting personalisation while disliking brands that know too much. Each one is a brief waiting to be written.

How do you read a conversation in three layers?

1. Start with emotion. Which emotions dominate the topic, and which are rising?
2. Add tone. Is the anger sarcastic and resigned, or earnest and mobilising? The second is far more urgent.
3. Find the tension. What are people torn between? Phrase it as "I want X, but Y".
4. Check the evidence. Read the posts behind each layer. Our guide to checking AI conclusions covers how.
5. Write the implication. What would resolve the tension? That is your product, service or message idea.

A worked example

A home-cleaning brand sees a steady "neutral to slightly negative" sentiment line and assumes nothing is happening. The three-layer reading tells another story. Emotion: growing guilt and weariness. Tone: self-mocking, often funny. Tension: "I want a clean home, but I do not have the time or energy, and I feel judged for that."

That tension rewrites the brief. Advertising that shows spotless homes makes the guilt worse. A message about "good enough, done fast" resolves it. The sentiment line never moved, but the opportunity was there all along.

For a further read on how audience tensions became a brand decision, see the Miraflora case study on the SOMIN site.

How do tensions help content teams?

Content built on trends tends to look like everyone else's. Content built on a real tension sounds like it understands the reader. That is why we share tension maps with content teams, not only with insight teams. Our sister consultancy GPT3 Marketing grounds generative content systems in what audiences actually respond to, and tensions are a natural input to that work.

The weekly show from Marketing Mondays often takes this approach too: one signal, decoded until the tension underneath is clear, then turned into something a team can act on.

How do you track tensions over time?

Tensions change slowly, which makes them useful for strategy. Keep a short list of the five tensions most relevant to your brand and review them monthly. Is each one growing, steady or fading? Has a competitor started to resolve one? Is a new tension forming? Emotions can swing week to week with news and memes; tensions show the underlying shape of what people need. When a tension you have been tracking starts to ease, check whether that is because someone, perhaps a competitor, has resolved it. That is a positioning signal worth acting on before it shows in market share.

How does Somonitor present the three layers?

Each topic view in Somonitor shows the leading emotions, the dominant tones and the tensions people express, side by side, with a trend line for each over time. Selecting any of them opens the tagged posts. That keeps the analysis honest and quick: you can move from "frustration is rising" to the actual sentences behind it in one step, then decide whether it needs a product fix, a message change or simply watching.

Sentiment tells you whether people are happy. Tensions tell you what would make them happier.

Checklist: are you reading beyond sentiment?

• Do your reports name specific emotions, not just polarity?
• Do you distinguish sarcastic from earnest tone?
• Is at least one tension written as "I want X, but Y"?
• Is every layer linked to example posts?
• Does each tension come with an implication for action?

Sentiment is not wrong; it is just not enough. The richer reading is where the useful decisions come from.

Frequently asked questions

Should we stop tracking sentiment?

Not necessarily. Sentiment is a useful headline and easy to trend over time. Treat it as a smoke alarm, then use tones, emotions and tensions to understand what is actually happening.

How does Somonitor detect sarcasm?

SOMIN's concept engine reads tone as a separate layer from emotion, which helps distinguish sarcastic from literal statements. It is not perfect, which is why every tag links to its posts for human checking.

What is a consumer tension?

A conflict a person is caught in, usually between two things they want or value. Writing it as 'I want X, but Y' makes it practical, because resolving that conflict is a clear brief for product, service or messaging.