Crisis early warning: reading the signals before they trend

A social media crisis rarely appears from nowhere. It usually starts as a small cluster of posts with an unusual emotion, then spreads into new communities before volume explodes. Early warning means watching for that combination of shifting emotion, spread and velocity on sensitive topics, rather than waiting for raw mention volume to spike.

Why do volume alerts fire too late?

Most listening tools offer a volume alert: tell me when mentions double. By the time mentions double, journalists have often noticed, screenshots are circulating and the story has a shape you no longer control. Volume is a lagging indicator.

It is also a noisy one. A product launch, a holiday or a viral meme can double your mentions with no risk at all. Teams that receive volume alerts learn to ignore them, which is the worst possible outcome for an early-warning system.

Somonitor's view is that the earliest crisis signal is a change in meaning, not a change in volume.

What signals actually come first?

Definitions: the four early signals

• Emotional shift: a topic that usually carries mild frustration starts to carry anger, fear or disgust.
• Community spread: a complaint that lived in one community starts appearing in unrelated ones, such as parenting groups or local news commenters.
• Narrative crystallisation: scattered complaints converge on one shared story, often with a phrase or image that repeats.
• Velocity: the rate of new posts accelerating, even from a low base.

Any one of these on its own is usually noise. Two together deserve a look. Three together on a sensitive topic deserve a call.

Somonitor detects these using SOMIN's concept engine, which decodes tones, emotions and tensions from each post. Because the engine keeps every concept linked to its evidence, an alert arrives with the posts that triggered it. Your team reads ten posts and decides, rather than debating a chart.

How should a team triage an alert?

We recommend a simple three-question triage, completed within the hour.

1. Is it true? Read the triggering posts. Is there a real incident, a misunderstanding, or coordinated activity?
2. Is it spreading? Has it crossed from one community into others? Who is amplifying it?
3. Is it about values? Complaints about service are usually recoverable. Stories that suggest a brand does not care about safety, fairness or honesty are much harder to recover from.

Map the answers to a response level: monitor, prepare a holding line, or escalate to leadership. Write the levels down in advance. Crisis decisions made at 11pm without a framework are rarely good ones.

A worked example

Picture a food delivery app. On a Tuesday afternoon, Somonitor flags a small cluster of posts about riders being penalised for delays caused by heavy rain. Volume is low, a few dozen posts. But the emotion has moved from irritation to indignation, and the posts have started appearing outside rider forums, in general community groups where customers are sympathising with riders.

Volume alone would not trigger anything. The combination of emotional shift and community spread on a fairness topic does. The team reads the posts, confirms the policy is real, and has a clear decision to make before the story crystallises. The cost of acting on Tuesday is a policy review. The cost of acting on Friday, after the story has a hashtag, is very different.

For a further read on how a delivery brand used audience evidence in a competitive market, the Keeta Hong Kong case study on the SOMIN site is worth your time.

How do you avoid crying wolf?

An early-warning system lives or dies on trust. If it alerts too often, people stop reading. Three practices help.

• Flag sensitive topics deliberately. Safety, pricing fairness, staff treatment, data privacy. Alert sensitivity should be higher on these and lower elsewhere.
• Require two signals. An emotional shift without spread, or spread without emotional shift, goes into a daily digest rather than an alert.
• Review every alert monthly. Which were useful? Which were noise? Tune thresholds accordingly.

What happens after the crisis?

Listening does not stop when the story fades. Watch whether the narrative returns, whether the emotion attached to your brand on that theme recovers, and whether competitors are discussed differently as a result. That aftermath belongs in your share-of-conversation grid, because a crisis often shifts which themes you own.

It is also a moment to check how AI assistants now describe you. Generative answers can carry an old crisis long after social media has moved on. Our sister agency SNMRush works on visibility in AI answers, and its perspective is useful when a story lingers.

Who writes the holding line?

Decide this before you need it. A holding line is a short, factual statement that acknowledges the issue, says what you are doing and promises an update. It should be drafted by comms, checked by legal and approved by a named leader, with templates ready for your sensitive topics. When an alert arrives, the team edits a template rather than starting from a blank page. Speed matters, but accuracy matters more: a holding line that later proves wrong becomes its own story.

The question is never whether people are talking. It is whether the meaning of what they say has changed.

Checklist: is your early warning ready?

• Sensitive topics listed and agreed with comms and legal.
• Alerts based on emotion, spread and velocity, not volume alone.
• Every alert includes the posts that triggered it.
• A one-hour triage routine with named owners.
• Pre-written response levels and holding lines.
• A monthly review of alert quality.

Crises are rarely surprises in hindsight. The posts were there. Early warning is simply the discipline of reading them while it still matters, and SOMIN's social media tooling and Somonitor are designed to make that discipline practical.

Frequently asked questions

How early can Somonitor detect a crisis?

It depends on how the story develops. Somonitor looks for emotional shifts and spread across communities, which often appear before volume spikes. We do not promise a fixed lead time; we aim to give your team the posts to judge while a response is still cheap.

Who should receive crisis alerts?

A small named group: usually a comms lead, a social lead and someone with authority to escalate. Broad distribution lists slow decisions and encourage people to assume someone else is handling it.

Should we always respond publicly?

No. Many alerts end with monitoring or a quiet fix. Respond publicly when people are addressing you directly, when the facts are clear, or when silence would itself become the story.