Getting a chat analyzed, without being reckless

What is in the file you hand over

An export without media is still a dense document: the messages from both sides, word for word, dated to the minute, over the whole length of the conversation. Often years. In the corpus used to calibrate RevealChat, one thread alone covered nearly seven years.

It also quietly contains the contact cards of anyone shared in the chat, so sometimes third party phone numbers. And the participants’ names, exactly as they sit in your address book.

Put plainly: half of that file belongs to the other person. That is the real question in this guide, before any technical consideration.

The three questions to ask any tool

1. 01Where is the file read?On your device, or on a server? A website that has you drag a file into a browser can do either, and the difference is total. The answer must be written down, not inferred.
2. 02What is kept, and for how long?The messages themselves, or only results? A specific duration has to be stated. "We do not keep your data" with no period and no scope means nothing.
3. 03Who else receives them?A tool that writes prose almost always leans on a third party AI model. That third party must be named, and what it does must be written down: retention, training, country.

Signals that should make you close the tab

• No publisher identity. No name, no address, no registration number. In Europe that is a legal obligation, not a design choice.
• No privacy policy, or a generic page that names no provider at all.
• An absolute promise. "Everything stays on your device" from a service that produces written prose is almost always false: writing needs a model, and the model has to receive something.
• A rating and thousands of reviews on a service nobody has heard of. Unverifiable social proof numbers are the cheapest signal there is to fabricate.
• A tool offering to read somebody else’s conversation. That is not analysis, that is surveillance, and it is illegal in most countries.

The consent question

Analyzing a conversation you took part in, to understand yourself, is one thing. The simple rule, and it applies to every tool: only import conversations you are a participant in.

What comes next matters just as much: what you do with the result. A result card shared publicly can expose an identifiable person over private words. The name can be blurred, the conversation stays recognisable to whoever lived it.

An analysis tool cannot tell you what somebody thinks, and none should claim to. It counts written messages. A result phrased as a diagnosis about a person, "he is avoidant", "she is a narcissist", is a badly written result, whatever the quality of the computation behind it.

What RevealChat does, precisely

We may as well turn the three questions on ourselves, and the honest answer has more nuance than the usual slogan.

• The cleanup is local. Reading the .zip, splitting it, merging identities, computing the counters: all of that runs on your phone, offline. Your export file is never sent.
• Nothing leaves before you start an analysis, including for the 3 free insights. You are the one who triggers it.
• When you do, excerpts leave. Not your contact list, not your media, not your address book: excerpts and counters, selected to limit what goes out. We do not write your messages to our servers, they pass through for the length of the call.
• Two model providers are named, Mistral AI and OpenAI. They keep what they receive for a rolling thirty days to detect abuse, then delete it, and do not use it to train their models.
• The database is in Ireland, inside the European Union.

So the accurate wording is not "everything stays on your phone". It is: the cleanup is local, writing the insights is not. The full detail is in the privacy policy, and deletion is requested from this page or in one tap inside the app.

One last thing RevealChat cannot do, by design: read a conversation you do not have access to. The app connects to no WhatsApp account. It opens a file you exported yourself, and nothing else.