How FlagMyChat Analyzes a WhatsApp Chat

1. Your file is opened on your device

The WhatsApp .zip is unzipped by your browser in a background worker. Folders, media and unexpected files are skipped, and only the chat .txt is read (up to 20 MB of text). System lines such as “Messages are end-to-end encrypted” and “<Media omitted>” are removed before anything is sent.

2. Exact statistics

Message counts, share of the conversation, who starts conversations (after 6+ hours of silence), typical reply time, emoji use, questions asked and active hours are counted directly, with no AI involved.

3. AI reading of tone and intent

The chat text is sent over HTTPS to an AI model configured not to retain or train on your data. The Free preview uses a fast model on the newest 50 messages; the Full analysis uses an advanced model on the whole chat. Very long chats are split into time segments, analyzed in parts and merged. The AI follows a fixed rubric:

• Toxicity (0–100 per person): 0–15 kind; 16–35 occasional friction; 36–60 recurring unhealthy patterns; 61–80 frequent control, contempt or manipulation; 81–100 pervasive abuse or threats.
• Flags need evidence: every red or green flag cites exact messages, in their original language with a translation.
• Fairness: both people are read; teasing between affectionate partners is not counted as disrespect.
• Languages: chats in any language, including mixed-language and transliterated chats.

4. Limits

AI reads words, not your whole life. Love language, attachment style and personality are estimates. Results are for entertainment and self-reflection, not professional or psychological advice. If a chat shows signs of abuse, the report shows helpline contacts.

5. What we keep

Only payment records (transaction ID, amount, date, status) and an email send log (address, time, plan, delivery status). Your chat and report are never stored. Read the Privacy Policy.