How to use ChatGPT for dating app replies — without sounding like ChatGPT

ChatGPT can write good dating-app replies once you give it three things: the conversation itself (a screenshot is fine), five or six messages you have actually sent, so it can copy how you text, and one line on what you want from this chat. Without them it writes like a customer-service agent who has been told to flirt. A dedicated app does the same job in fewer taps — it is a shortcut, not a better writer.

A disclosure, because it matters here: we make one of those apps, Rizz Wingman. This page shows how to get most of what it does from ChatGPT for free, and says where the app is still faster. You can judge the trade for yourself.

Why ChatGPT’s first draft sounds wrong

Paste a chat into ChatGPT and ask “what should I reply?” and you get the median polite stranger:

• Too long. Two sentences where you would send six words.
• Too correct. Capital letters, full stops, a comma before “haha”.
• Three questions at once, so the other person has homework instead of a reply.
• The tics — “I love that!”, em dashes, “honestly”, a compliment nobody asked for.
• Nothing from the chat. A reply that would fit any conversation fits none.

None of that is ChatGPT being bad at writing. It simply does not know you, so it writes like nobody in particular — and “nobody in particular” is exactly what people have learned to spot. The seven tells are here, and an unedited ChatGPT reply usually has four of them.

Set it up once

Put this in ChatGPT’s custom instructions (under Personalization in Settings), or in a Project called “Texting” so it only applies there. Fill in the brackets with your own messages — real ones, copied from your own chats, typos and all:

When I ask for dating-app replies:

• Write the way I text. Messages I’ve actually sent: [paste five or six of your real messages]
• Match my punctuation and capitals. No emoji unless my examples use them.
• Under 25 words. One question at most.
• No em dashes, no “haha I love that”, no compliments on looks.
• Give three options that do different things, not three versions of one.
• Use something specific from the chat or their profile.
• If I ask for something pushy or manipulative, tell me instead of writing it.

The examples do most of the work. Five real messages teach it more about how you write than any description of your “vibe” could, and they are the difference between a draft you edit lightly and a draft you rewrite.

The prompt, every time

Upload the screenshot — cropped, so their name and photos are not in it — and add one line:

Here’s our chat. I want to [keep it going / ask them out this week / bring it back after a few days of silence]. Three replies, my style.

The goal line matters more than it looks. “Keep it going” and “ask them out” need different messages, and without it ChatGPT defaults to keeping it going forever, which is how a good chat turns into a pen pal.

Four situations, four goal lines

• They opened with “hey”. “They opened with ‘hey’. Here’s their profile text: [paste]. Pick one specific thing from it and ask one easy question.” Why that works is in how to respond to “hey”.
• You are writing the first message. “No chat yet. Here’s their profile: [paste]. Three openers that each pick a different detail.” More on the detail worth choosing in first messages on Hinge, Tinder and Bumble.
• It went quiet. “The last message was mine, four days ago. Something that gives them a new, easy thing to answer — not ‘hey stranger’.” The longer version: how to revive a dead conversation.
• You want to meet. “Ask them out: a specific day and a specific thing, in one message, easy to say no to.” Why specific beats open-ended: asking someone out over text.

Where ChatGPT is the wrong tool

When you want an answer in ten seconds. The setup is a one-time job, but every chat still needs the screenshot, the goal line and a moment to pick and edit. At the exact moment you are stuck, that minute is the friction — and it is the whole reason dedicated apps exist.

When the draft is pushier than you are. ChatGPT has no rules written for dating. Ask it for something manipulative and it may oblige; the last line of the instructions above helps, but it is a request, not a guardrail. If a draft says something you would not say across a table, delete it — that is not confidence, it is a different person.

When privacy matters to you, or to them. By default your chats stay in your history and can be used to train OpenAI’s models. Turn off “Improve the model for everyone” under Settings → Data controls, or use a Temporary Chat, which is not used for training and is not kept in your history (OpenAI may hold it for up to 30 days for safety). And the messages you are uploading belong to someone else too — crop their name and photos before anything leaves your phone.

ChatGPT or a rizz app?

If you will do the setup, ChatGPT is a fair place to start, and it costs nothing. If you find yourself skipping the setup — most people do, which is why the drafts come out generic — that gap is what a dedicated app fills. The full comparison of rizz apps puts the paid ones side by side, ours included.

Whichever writes the draft, the rule is the same: edit it, never send the first option, and never send a line you would not say out loud. The draft is the idea. The message is yours.