You catch a social media crisis early by deciding, in advance, which words, which spikes and which accounts deserve a human's eyes within 15 minutes. Set up keyword alerts, define a spike in numbers, and send both to one named person. Everything else in a crisis plan depends on noticing in time.
TLDR: Monitor your names plus six risk words. Define a spike as a multiple of your normal comment volume. Send flags to one owner. Triage with three questions: how many people, how fast, how serious.
This post is about the first 15 minutes: detection and triage. If you already know a crisis is under way and need the roles and statements, use the crisis communication playbook instead.
List the words that should wake someone up
Open a note and write two lists.
- Names: brand, product names, the founder, any spokesperson, common misspellings and the handle on each platform.
- Risk words: refund, scam, fake, lawsuit, recall, unsafe, "never again". Add two or three that fit your industry, for example "allergic" for food or "data" for software.
Keep it short. A list of 60 words produces 60 kinds of noise, and the person watching will stop looking within a week.
Turn the list into rules, not a habit
Relying on someone to scroll comments "when they get a chance" is how a crisis gets found by a journalist. Put the lists into tools that do the looking.
In Mydrop, trigger-word rules can flag comments that contain a risk word, and social monitoring tracks mentions of your names. Both feed the same place your team already works, so a flagged comment appears next to every other message instead of in a tool nobody opens.
Define a spike with a number
"A lot of comments" is not a trigger. Pick a rule you can check:
- Comments or mentions in an hour at least 3 times your usual hour.
- Five or more flagged messages with the same risk word in 30 minutes.
- Any single post of yours being quote-posted by an account much larger than yours.
Your first week of numbers will be rough. Write down the normal range from the last 30 days, then adjust the multiple after your first false alarm.
Send every flag to one owner
Detection fails at the handoff. Decide who gets the flag, and give that person a backup for evenings and weekends. One owner means nobody assumes someone else saw it.
Put the owner's name, backup and phone number at the top of the plan. Alerts that go to a shared channel nobody is responsible for are the same as no alerts.
Triage with three questions
When the owner sees a flag, they answer three questions in under five minutes and write the answers down.
| Question | Low | High |
|---|---|---|
| How many people? | One or two accounts | Many unrelated accounts or a big one |
| How fast? | Flat for an hour | Doubling every 15 minutes |
| How serious? | Annoyance, a service issue | Safety, legal, discrimination, a lie about you |
Any "High" in the third row is a crisis on its own, even with a small audience. Two or more "High" answers in any rows go to the decider immediately. All "Low" stays in the normal inbox with a same-day reply.
Route it by severity in the inbox
Triage only works if the message is where you can act on it. Mydrop's social inbox holds comments from every connected channel in one list, so the owner can assign a flagged thread, add a note, and let the decider see it without forwarding screenshots. A screenshot in a chat is an hour lost, because nobody can tell if it is still current.
For the replies themselves, keep to facts, say what you are doing, and avoid arguing in public. Take detail into a private message and say so.
Stop the self-inflicted version
Some of the worst crises start from your own post. A caption that reads badly, a link to the wrong page, a joke scheduled for a bad day. A pre-publish check catches the cheap ones. See the pre-publish checklist for links and media for the mechanical errors, and give sensitive posts a second reader before they go live.
Test it once a quarter
Every three months, post a harmless test comment containing a risk word from a spare account and time how long it takes the owner to notice and answer the three questions. Fix whichever step was slowest. A detection plan you never test degrades quietly, as people change roles and rules go stale.
Your next step: write your name list and risk list today, create the rules, and put one person's name on the alerts before you close this tab.
Google review
Trustpilot review
Product Hunt comment
