Listening vs tracking: the ethics of hearing your audience

Listening means studying public conversation in aggregate to understand what people think, feel and need. Tracking means following identifiable individuals across contexts to predict or influence them. Both use data, but they build opposite relationships: listening earns trust by responding to what people say, while tracking erodes it by knowing more than people chose to share.

Why does this distinction matter now?

In the conversations we analyse about brands and AI, one theme keeps returning: people are uneasy with brands that seem to know them too well. An ad that references something you only mentioned to a friend. A recommendation that feels like being watched. The reaction is rarely gratitude. It is suspicion.

The insight that shaped Somonitor came from that listening: trust is bred in the soil of conversation, not in the data of algorithms. Brands earn trust when they visibly respond to what people say in public. They lose it when they appear to be assembling private files.

That gives marketers a clear choice of posture. You can try to know each person, or you can try to understand the conversation. Somonitor is built for the second.

What is the difference in practice?

Definitions

• Listening: analysing public posts at the level of themes, concepts and communities. The unit of analysis is the conversation.
• Tracking: linking behaviour to an identifiable person over time and across places. The unit of analysis is the individual.
• Aggregate insight: a finding that holds across many posts and does not depend on who wrote any one of them.
• Evidence trail: the set of posts that support a finding, available for review, without building profiles of the authors.

A listening question sounds like "what are parents worried about when choosing a preschool?" A tracking question sounds like "which parents in this postcode looked at preschools last week?" The first leads to better products and messages. The second leads to targeting that may work for a quarter and cost you trust for years.

Is listening automatically ethical?

No. Public posts are public, but that does not make every use acceptable. A small community discussing a health condition, a fan group with its own norms, a teenager venting: all public, all deserving care. Ethical listening is a practice, not a property of the data.

We think of it the way the best researchers do. You are a guest in other people's conversations. You learn from them; you do not exploit them.

A practical code for ethical listening

1. Report patterns, not people. Findings should describe themes and communities. Never build lists of individuals from listening data.
2. Quote with care. When you share example posts internally, remove handles unless the author is a public figure speaking publicly. Never quote vulnerable people in external material.
3. Do not respond to everything you hear. Joining a conversation you were not invited into can feel intrusive. Respond where people address you or where you can clearly help.
4. Respect community norms. Some spaces are public by platform but private by culture. Read the room before you act on what you learned there.
5. Make conclusions checkable. Every AI-generated finding should link to the posts that support it, so a human can confirm the machine did not over-reach.
6. Keep the purpose honest. Use listening to understand and serve, not to manipulate a moment of vulnerability.

Why does checkability matter for ethics?

AI summarisation makes listening faster, and it also makes it easier to be confidently wrong. A model might label a group "angry" when they are joking, or merge two communities that would be offended to be merged. If you cannot see the posts behind the conclusion, you cannot catch the mistake, and you may act on a caricature of real people.

This is why Somonitor is built on SOMIN's approach to AI that listens: every concept tag traces to the posts that evidence it. Checkability is not only a quality feature. It is how you keep humans accountable for what the machine says about other humans. We expand on this in why every AI conclusion should be checkable.

What does good listening look like for a brand?

Consider a family-focused retailer that hears, across many public posts, that parents feel judged by baby-product advertising showing impossibly calm homes. A tracking mindset would look for those parents and retarget them. A listening mindset changes the creative, so the next campaign shows real mess and real tiredness, and nobody had to be followed anywhere.

That is the difference in one example. The brand responded to the conversation, visibly, and everyone benefited. For further reading on how this kind of audience understanding becomes a campaign, see the Mothercare Singapore case study on the SOMIN site.

How do teams keep each other honest?

Ethics rarely fails in a policy document. It fails in a rushed Tuesday meeting. Make a habit of asking two questions whenever a listening insight becomes an action: would the people who wrote these posts be comfortable seeing how we used them, and can we show the evidence? Our sister consultancy AgentC writes about keeping human judgment in charge of automated marketing, and the same principle applies here: people own the decision, machines prepare it.

What about personalisation?

Teams sometimes worry that giving up tracking means giving up relevance. It does not. Listening produces relevance at the level of the message rather than the person: you learn which worries, hopes and phrases are common in a community, and you write for them. Everyone who shares that concern feels understood, and nobody feels followed. That is a different kind of personalisation, earned rather than extracted.

The brands audiences trust are the ones that clearly heard them, not the ones that clearly watched them.

Checklist: are we listening or tracking?

• Is the unit of analysis a theme or community, not a person?
• Could we explain this use to the authors without embarrassment?
• Are sensitive communities handled with extra care?
• Can every conclusion be traced to posts and checked?
• Does the action respond to what people said, rather than exploit what we inferred?

Listening is the more modest choice, and the more effective one over time. It is the reason Somonitor exists.

Frequently asked questions

Is social listening legal?

Analysing public posts is generally lawful, but platform terms and data-protection rules such as PDPA or GDPR still apply, especially to storing personal data. Aggregate, theme-level analysis without building profiles keeps you on much safer ground. Take legal advice for your market.

Does Somonitor build profiles of individuals?

No. Somonitor reports patterns across conversations, themes and communities. Example posts are available so findings can be checked, but the product is not designed to follow or profile individual people.

Can listening replace first-party data?

They answer different questions. First-party data tells you what your customers did with you. Listening tells you what the wider market thinks and feels, including people who are not yet customers. Most teams need both, used with consent and care.