7 Warning Signs a Customer Is About to Leave a Bad Review
Every bad review on Google, every BBB complaint, every angry LinkedIn post was a support email first. The customer told you they were unhappy. Probably more than once. The signals were there. You just didn't catch them in time.
After analyzing thousands of customer interactions, we've identified seven specific patterns in support emails and call transcripts that reliably predict a customer is about to go public with their frustration. These aren't vague "sentiment indicators." They're concrete, detectable signals that show up days or weeks before the review gets posted.
1. They mention a specific platform by name
Signal: Platform name-dropping
Risk Score: 80-95When a customer names Google Reviews, BBB, Yelp, LinkedIn, or any review platform in a support email, they're not thinking about it abstractly. They've already looked up where to post. The decision tree has shifted from "Should I complain publicly?" to "Where should I complain publicly?"
This is the highest-urgency signal we track. In our data, customers who name a specific review platform in a support email post a review within 72 hours over 60% of the time.
2. They reference documentation or evidence
Signal: Evidence gathering language
Risk Score: 75-90Phrases like "I have screenshots of every conversation," "I've documented everything," or "I'm keeping records" signal a customer who is building a case. They're not just frustrated. They're preparing for something.
Evidence-gathering language appears an average of 11 days before a formal complaint. It's one of the earliest reliable signals because the customer is still in preparation mode, which means you still have time to intervene.
3. Legal language appears
Signal: Attorney, BBB, or regulatory mentions
Risk Score: 85-95Any mention of attorneys, legal teams, the BBB, consumer protection agencies, or "consulting with my lawyer" is a critical alert. Even if the customer never follows through legally, the mindset that produces this language almost always produces a public review.
Legal language and review threats are highly correlated. In service businesses, customers who mention attorneys are 4x more likely to also post a negative review than those who simply ask to cancel.
4. Contact frequency accelerates
Signal: Repeat contacts in short timeframe
Risk Score: 60-80A customer who emails once a month suddenly emailing three times in a week. A customer who has never called now calling every other day. Acceleration in contact frequency is one of the most overlooked churn signals because it requires tracking patterns over time, not just reading individual messages.
The risk isn't just the frequency. It's what happens when they stop. A customer who goes from 4 contacts in a week to zero didn't calm down. They gave up on you and moved to a public channel.
5. They name your competitors
Signal: Competitor comparison or switching intent
Risk Score: 65-80When a customer mentions a specific competitor by name, they've already researched alternatives. This isn't theoretical. They've visited the competitor's website, maybe even talked to their sales team. They're telling you they have options.
Competitor mentions in frustrated contexts are different from feature comparison questions. A calm customer asking "Does your product do X like CompetitorY?" is evaluating. A frustrated customer saying "CompetitorY already offered me a better deal" is leaving.
6. Tone shifts across interactions
Signal: Declining sentiment trajectory
Risk Score: 45-65The first email is polite: "Hi, just checking on my order." The second is shorter: "Still haven't heard back." The third drops pleasantries entirely: "This is unacceptable." Tracking tone across a customer's interaction history reveals escalation patterns that a single email can't show.
Tone trajectory is harder to detect manually because each individual email might seem reasonable in isolation. You need to read them as a series. AI-powered analysis can score sentiment across a customer's full history and flag when the trajectory turns negative.
7. They set deadlines or ultimatums
Signal: Time-bound threats
Risk Score: 70-85"If I don't hear back by Friday." "You have 24 hours to resolve this." "This is your last chance before I..." Ultimatums are the customer drawing a line. They've decided on a consequence and given you a countdown. Miss the deadline and the consequence is almost guaranteed.
Deadline language is particularly dangerous because it comes with a built-in timer. If you see it on Monday and don't respond until Thursday, you've already lost.
The Common Thread
All seven signals share one trait: they're hiding in text you already have. These patterns are in your support inbox, your CRM notes, your call recordings. The data exists. The problem is that no human can read every email, listen to every call, and track sentiment trajectories across thousands of customer interactions.
That's exactly the gap AI analysis fills. Not replacing your CS team's judgment, but making sure the right conversations reach the right people before it's too late.
What to do when you spot these signals
- Escalate immediately. Critical signals (legal language, platform mentions, documented evidence) should reach a senior manager within the hour, not sit in a queue.
- Call, don't email. A phone call from a real person defuses situations that email threads escalate. It's harder to post a bad review about someone who just called you personally.
- Acknowledge the full history. "I've reviewed your previous messages and I can see this has been going on too long" is more powerful than "How can I help you today?"
- Offer a concrete resolution with a timeline. Vague promises trigger more frustration. "I'll have a corrected invoice to you by 3 PM today" is specific and verifiable.
- Follow up after resolution. The customer who almost left a bad review can become your strongest advocate if you genuinely fix the problem and check back in.
Stop reading signals manually
RiskDetect analyzes every support email and call transcript for these exact signals automatically. Try it with your own customer data.
Try the live demo