Decoded
Can ChatGPT analyze my text messages?
The short version, the four blind spots nobody warns you about, and an honest account of where her words end up once you paste them.
Yes. ChatGPT can analyse a message you paste into it, and it does it better than most people expect — but it cannot see your phone, it agrees with you far too easily, and everything you paste leaves your device.
Two different questions hide inside this one. If you mean “can ChatGPT read my messages app and tell me what’s going on with her” — no. It has no access to your inbox. It sees the message you paste in, or a screenshot you upload, and nothing else. If you mean “can it take the one text I’ve now read eleven times and tell me what it probably means” — yes, and reasonably well.
What it genuinely gets right
Pasting a message into a chatbot at 1am is not a stupid thing to do, and we’re not going to pretend otherwise. Reading likely intent out of a short piece of text is close to the thing these models are actually good at. Three things it does well:
- It names the emotion. You can’t see past your own panic. It can. Ask what someone might be feeling when they send “it’s fine, do what you want” and you get a fair answer that doesn’t make anyone the villain.
- It writes better replies than you do at 1am. Not cleverer — calmer. It will not send the four-paragraph message you were drafting.
- It is awake, free, and it doesn’t tell your friends. Not a small thing at 1am.
Give it a proper prompt and it gets better again. We wrote one you can copy: the ChatGPT prompt that decodes her texts. It’s free and it works.
Four things it gets wrong
None of these are hypothetical. They happen every time, unless you fight them.
- It agrees with you. Tell it she was being unreasonable and it will show you how unreasonable she was. Tell it you ruined everything and it will agree with that too. It is trying to be useful to the person in the chat — and she is not in the chat. Paste her message with no commentary, or you are just getting your own theory back with better grammar.
- Its severity scale drifts. Ask the same question in two fresh chats and you can get “she sounds a bit tired” and “this is serious” about the same seven words. Nothing anchors it — so people re-roll until they get the answer they wanted, which is worse than not asking.
- It won’t warn you off the wrong reply unless you make it. It hands you three good replies and says nothing about the one you were already drafting — the defensive one that starts “well if you’d just said”. Most of the damage in a text argument is one message long. Ask it explicitly what not to send.
- It doesn’t know her. Not that she always types like that, not that her mum is in hospital, not that you had this exact argument in March. It is reading seven words cold.
And that last one, we can’t fix either. Decoded reads one message too. You know her; it doesn’t. The last call is always yours.
She has an opinion and has decided not to spend it. “Do whatever you want” is usually withdrawal rather than permission — she has stopped expecting the conversation to go anywhere useful, so she’s handing it back to you.
Don’t: take it literally and go ahead, then act surprised later. Don’t answer with “fine, I will then.”
That structure — the reading, the tone flag, the replies, the thing not to send — is what a chatbot gives you only if you remember to ask for all four, at midnight, with shaking hands. Closing that gap is the whole of what Decoded is. It runs on Claude, the analysis is identical in the free and paid tiers, and it isn’t out yet: get early access.
Where your messages actually go
When you paste her text into ChatGPT, it leaves your phone, is processed on OpenAI’s servers, and sits in your chat history until you delete it. Consumer ChatGPT also has a data-controls setting governing whether your conversations can be used to improve the model — go and look at where yours is set rather than assuming.
The same question applies to us, so here is the answer. Decoded stores your decodes against your account — that is what makes your history work — and the message text is sent to Anthropic’s Claude to be analysed. We don’t sell your data. We’re not going to tell you we store nothing, because we do. Read the privacy policy →
And there is the part nobody in this category says out loud: they are her words. One short message with the names taken out is a very different thing from uploading a year of your relationship as screenshots. Decide where your line is before you’re at 1am with a phone in your hand.
The point where you stop asking a chatbot
Here is the tell. If you have run the same message through ChatGPT three times, rewording your side of it each time until you got a reading you could live with, the tool is no longer analysing anything. It is keeping you company while you avoid a two-line text.
Send this instead:
Hey — I’ve read your last message a few times and I honestly can’t tell if you’re annoyed with me. Are you?
It is not clever. It is not supposed to be. It ends the guessing in one move. Use the AI to work out what to send; don’t use it as a reason not to send anything — and if you’re not even sure there’s something wrong, start with how to tell if she’s actually mad at you over text.
Questions people also ask
No. ChatGPT has no access to your phone’s message history. It only ever sees what you paste in or upload as an image. There is no setting that hands it your inbox, and any app claiming to read your messages in the background is describing something ChatGPT does not do.
It depends what you mean by safe. The text leaves your phone, is processed on OpenAI’s servers, and stays in your chat history until you delete it. Consumer ChatGPT also has a data-control setting governing whether your conversations can be used to improve the model — check where yours is set rather than assuming. And the words are hers, not yours.
For one message and a good prompt, ChatGPT gets you most of the way there. Decoded is that idea on a button: the same five-part structure every time, a tone flag pinned to the same three levels — green, yellow, red — instead of a fresh adjective each time, and a “what not to say” you never have to remember to ask for. It runs on Claude, it isn’t released yet, and it doesn’t know her either.