Green flags and red flags in a WhatsApp chat: how to read the tone

You have read the message five times. "Ok." Is it fine, or is it not fine? You scroll up looking for the moment the tone changed, and an hour later you are less sure than when you started.

Tone is the hardest part of a conversation to judge fairly, especially when you are in it. This guide covers what is worth looking for, with examples, and how Le Confidant's free analysis groups what it finds into 💚 green flags, ⚠️ things worth attention and 🚩 red flags.

Why tone is hard to read in text

Three things work against you.

A message has no voice. "Sure" can be warm, neutral or cold, and text strips out the cues that would tell you which.

Memory keeps the sharp moments. One curt reply stays with you longer than twenty kind ones, so rereading a chat tends to confirm whatever you were already worried about.

One message is not a pattern. Everyone sends a short reply when they are busy. What matters is what keeps happening, and in which direction.

So the useful question is not "what did this message mean?" but "what does this person tend to do in this conversation, and how does the other one respond?"

💚 Green flags: what a healthy exchange looks like

• Support. Comfort, encouragement or practical help when the other person is struggling. "That sounds exhausting. Do you want me to pick up dinner?"
• Repair attempt. An effort to calm things down during or after tension: an apology, a joke, a "can we start over?". The psychologist John Gottman found that whether repair attempts are made, and accepted, says more about a couple than whether they argue at all.
• Substantive reply. An answer that engages with what was said rather than "ok" or "lol". It shows the message was actually read.
• Interest in the other person. Questions about their day, their plans and how they feel, and remembering what they said last time.
• Emotional disclosure. Saying how you actually feel: "I felt a bit left out yesterday." It is a risk, and taking it is usually a sign of trust.

⚠️ Worth attention: defensiveness

Defensiveness is answering a concern by protecting yourself instead of hearing it: counter-attacking, making excuses, or turning it back on the other person.

A: You said you'd call last night. B: I had a really long day, I don't need this right now.

One instance means little; everyone gets defensive when tired. It deserves attention when it is the usual answer to any complaint, because then problems are never discussed, only defended.

🚩 Red flag: stonewalling

Stonewalling is withdrawing from the conversation instead of engaging: one-word replies to a serious message, changing the subject, or going silent in the middle of a discussion.

A: Can we talk about this weekend? I'm still upset. B: k

Gottman's research lists stonewalling among the four patterns that best predicted couples breaking up, when it becomes a habit. In text it is also the easiest to misread: silence can just mean someone is driving or asleep. Look for it during a discussion, and for whether it repeats.

What the free preview shows you

When you upload an exported chat to Le Confidant, the free analysis gives you the statistics straight away: who writes more, who starts conversations, reply times and the rhythm of the chat (step by step here). Depending on the conversation, it can also show a tone preview:

• It reads a sample of passages spread from the beginning to the end of the conversation, not every message.
• It shows up to four real moments, each quoted from your chat with the pseudonym of who wrote it and the date, and marked 💚, ⚠️ or 🚩.
• A ⚠️ or 🚩 is only shown when the model is highly confident. Some labels are deliberately kept out of the preview: criticism and contempt are too heavy to show as a first impression, without the full context.
• It shows moments, not a score. There is no count, no trend and no "who is worse" comparison, because a sample cannot support any of those.

If the preview shows no 🚩, that does not prove there are none. It means nothing in the sampled passages crossed the bar.

What the full report adds

The preview is a sample. With Pro, the analysis reads the whole conversation and shows:

• How the emotional tone changes week by week, for each participant.
• Conflict patterns: how intense they are, whether they settle or escalate, and who tends to raise or lower the tension.
• Clarity: misunderstandings, requests for clarification, and the moments the exchange got back on track.
• Engagement: who invests, and who mostly reacts.

Findings are backed by passages from the conversation, so you can check them yourself, and you can ask the Confidant your own questions, such as "When did the replies get shorter?". Pro is paid per session, with no subscription, and stays open for 24 hours. If a number in the free analysis puzzles you, what your WhatsApp stats really mean goes through them one by one.

How to read flags fairly

A flag is a moment, not a person. One 🚩 says something about one exchange. Read what came before and after it.

Look at both sides. If one person stonewalls, look at what they were responding to. Patterns usually involve two people.

Check the timing. A cold month can match a new job, exams or a loss. Compare it with what was happening at the time.

Use it to talk, not to win. An analysis is a starting point for a conversation, not proof in an argument. Use it only on conversations you are part of: it is not a way to check someone else's phone.

If a conversation leaves you feeling afraid or controlled, an analysis is not the right tool. Talk to someone you trust, or to a local support service.

No account is required, but messages can still identify people. Access expiry and physical deletion are separate. See the Privacy Policy for cleanup, retained records and provider copies. We do not sell personal data or train models on uploaded conversations. The full journey of your data is in is it safe to upload your WhatsApp chat?

Frequently asked questions

Is the tone preview free? Yes. It is part of the free analysis, with no account and no card. It appears when the sampled passages contain moments clear enough to show, so some conversations show none.

Can AI tell whether a message is passive-aggressive? Not with certainty. The analysis estimates how likely a passage is to show a pattern such as defensiveness, and only shows warnings it is highly confident about. Each one is quoted, so you can judge it in context yourself.

Why does my preview show no red flags? Either none of the sampled passages showed one clearly enough, or there were none to find. The preview reads a sample, not the whole chat, so it is not a clean bill of health.

Does it work for group chats? Yes. Every participant gets a pseudonym, and each quoted moment shows who wrote it.

Can I check the tone of someone else's conversation? No. You analyse an export you made yourself of a conversation you are part of. Nothing here reads another person's phone or messages.

Is my conversation used to train AI? No. Neither we nor our AI providers train models on uploaded conversations.