The Complete Guide to
Chat Analysis, Step by Step
You want to look at a conversation with fresh eyes. This guide covers the practical side: choosing the chat, exporting it from each platform, picking a tool, reading the output in the right order, and keeping the whole thing private.
Key takeaways
- Five steps: choose the conversation, export it, pick a tool, read the results in a fixed order, decide what to do.
- Every major platform exports plain text. WhatsApp and Telegram take a minute; Instagram and Messenger take hours to days.
- Free counters give you message and emoji statistics. AI tools add sentiment, personality and compatibility. Choose by the question you are asking.
- Read trends first, then balance, then style. Skipping straight to a score is how people misread their own relationship.
- Analyze with consent, strip what you do not need, and prefer tools with a stated retention policy.
Chat analysis is the systematic reading of a message history for patterns: how tone moves, who replies when, how each person writes. Unlike a questionnaire, it works from what people actually sent. This guide is about doing it. For what the results mean, and what the research supports, read Chat Analysis: The Ultimate Guide to What Your Conversations Reveal.
The five steps
1Choose the conversation and get consent
Pick a one-on-one thread with at least a few hundred messages over several months. Timing statistics settle quickly, but trend lines need range, and a two-week chat gives you counts rather than patterns. Group chats work for volume statistics only.
Then tell the other person. The export is half their words, and analysis done in secret tends to become ammunition rather than insight.
2Export the chat
Every major platform has a built-in export. Choose text without media unless a tool specifically asks for it; media adds gigabytes and no analytical value.
| Platform | Where | Format | Time | Guide |
|---|---|---|---|---|
| Chat info → Export chat → Without media | .txt | Under a minute | WhatsApp export guide | |
| iMessage | Mac: Messages database or a third-party exporter | .txt / .csv | Minutes, Mac required | iMessage export guide |
| Accounts Center → Download your information → Messages | JSON | Hours to days | Instagram export guide | |
| Messenger | Accounts Center → Download your information → Messages | JSON | Hours to days | Messenger export guide |
| Telegram | Desktop: Settings → Advanced → Export Telegram data | HTML / JSON | Under a minute | Telegram export guide |
| Snapchat | Account settings → My Data | JSON | Hours | Snapchat export guide |
3Pick a tool that matches your question
Tools fall into three tiers. Choose by what you want to know, not by feature count.
Free web counters
Browser-based analyzers such as ConvoAnalyzer, Chatilyzer, WhatsAnalyze and ChatVisualizer produce message counts, emoji statistics, active hours and word clouds, often without uploading anything. Good for "who talks more" and group-chat leaderboards; no sentiment or personality.
Open-source scripts
Projects such as chat-analytics on GitHub and the many Python WhatsApp analyzers run locally and let you define your own metrics. Best for people comfortable with a terminal who want full control of their data.
AI relationship tools
MosaicChats reads the export with language models and returns a sentiment timeline, reply-time and initiation statistics, MBTI and attachment-style estimates, a word cloud, and a 0–100 compatibility score, with Myrah available to explain any chart. Analyzing chats is free (up to 50 uploads a week); the free plan includes five Myrah messages a week and Pro raises that to 200. For a wider survey, see our comparison of chat analysis tools.
Pasting texts into ChatGPT has become common enough to make the news. It gives a quick qualitative read, but it cannot compute timing statistics from a screenshot, and you should check the retention settings before pasting private messages.
4Read the results in order
Most people jump to the score. Read in this order instead, because each layer gives the next one context.
- Trends. Look at the sentiment timeline and the reply-time chart across the whole period. Is the line stable, rising, or sliding? Where are the dips, and how long did recovery take?
- Balance. Who initiates, and has that changed? Do message lengths match, or does one person write paragraphs to one-word replies?
- Style and personality. Treat MBTI and attachment estimates as descriptions of how each of you writes, not diagnoses. A 2024 study found GPT-4 classified full MBTI type correctly for 76% of participants, good but not certain.
- Topics. The word cloud shows what the relationship is about. Logistics-only vocabularies and feelings-heavy vocabularies tell different stories.
Healthy signs
- Initiation near balanced, or stable if uneven
- Reply times consistent across months
- Tone that recovers quickly after dips
- Questions asked in both directions
- Future-oriented language
Warning signs
- Message frequency falling over months
- Reply times lengthening in one direction only
- Tone sliding without recovery
- Messages getting shorter and flatter
- Topics that used to appear and no longer do
5Decide what to do with it
Pick one thing. If initiation is 85/15, ask for one unprompted message a day. If tone dips every Sunday night, name the Sunday problem. Share the finding as a question rather than a charge: "I noticed our chat got a lot quieter in March, what was going on for you?"
Do not let the tool write your messages. A study in Scientific Reports found that when people suspect a partner is using AI to compose texts, they rate that partner more negatively. Use the analysis to understand the conversation, then write your own. Re-run it in three months to see whether the change stuck.
Privacy checklist
Before you upload
- The other person knows and is comfortable with it.
- You have read how long the tool keeps the data and whether it trains models on it.
- You have removed anything the analysis does not need: addresses, financial details, third parties' messages.
- You prefer tools that anonymize before processing or run locally.
- You plan to delete the upload once you have what you came for.
MosaicChats anonymizes chats before processing; the details are in our privacy mission. Whatever tool you choose, the checklist above applies.
Run the five steps on your own chat
Export a conversation from any supported platform, upload it, and read the timeline, timing stats and style analysis in the order this guide recommends.
Frequently asked questions
How many messages do I need for a useful analysis?
A few hundred messages spread over at least two or three months. Timing statistics stabilise quickly, but trend lines and personality estimates need range; a two-week chat will give you counts, not patterns.
Can I analyze a group chat?
Free counters handle groups well and will rank members by volume and emoji use. Relationship-focused tools are built for two people, so a group export will either be rejected or produce a muddled compatibility read. Export the one-on-one thread instead.
Will the other person know I analyzed our chat?
No tool notifies them. That is exactly why you should tell them yourself: the export is half their words, and analysis done in secret tends to be used as ammunition rather than insight.
Which export format should I choose?
Text without media for WhatsApp, JSON for Instagram and Messenger, HTML or JSON for Telegram. Media adds gigabytes and no analytical value for most tools. Check the tool's supported formats before you export.
How long does an export take?
WhatsApp and Telegram export in under a minute. Instagram and Messenger requests go through Meta's Accounts Center and typically arrive within a few hours, occasionally days. iMessage requires a Mac and either a database export or a third-party tool.
Related articles
References & Sources
- "Woman Asks ChatGPT To Analyze Her Date's Texts, Left 'Shocked' by Response." Newsweek, 2024.Source
- "Artificial Intelligence and Personality: Large Language Models' Ability to Predict Personality Type." SAGE Journals, 2024.Source
- Hohenstein, J., et al. "Artificial intelligence in communication impacts language and social relationships." Scientific Reports, 2023.Source