πŸ’” Heartbreak.exe 5 min read Β· August 2026

I Let an AI Write My Breakup Text. He Replied "Finally, Someone Communicates."

By Jules, 27 β€” certified human editor

For three years, Sam and I communicated the way modern couples do: in half-finished sentences, emoji that meant four different things, and one catastrophic incident involving an "ok πŸ‘" that started a 36-hour cold war. By month 34, the relationship had gone flat in that specific way where you're not fighting, you're just both quietly disappointed by the same fridge magnet.

I wanted out. What I didn't have was words. Every draft I typed sounded like a hostage note written by a spreadsheet: "I think maybe we should perhaps possibly consider…" So at 10 PM on a Thursday, I did the coward's errand. I opened a chatbot, pasted the entire situation, and typed: "Help me end this kindly."

Four perfect paragraphs

The bot came back in eleven seconds with four paragraphs so composed they should have been wearing a blazer. It acknowledged the good years. It named the specific pattern β€” two people loving each other but defaulting to roommates who split utilities. It wished him well without leaving a door open. Zero typos, zero "lol," zero accidental em-dash crimes.

I read it three times. It was exactly what I meant. Which was devastating, honestly β€” a machine had accessed my own feelings more clearly than I had in six months.

I sent it. I am not proud of the sending. I am proud that I didn't add "haha" at the end, which was my only contribution.

Sam replied four minutes later: "Wow. Finally, someone communicates clearly. Yeah. I think you're right. Can I ask β€” did you get replaced by a podcast?"

My group chat took one look at the screenshots and said, in unison: "You didn't write that." Evidence: there were no typos, I said everything exactly once, and nothing trailed off into a second, weaker text.

The bot could write the words. It couldn't hold the phone.

Here's what I figured out in the week after, while the bot and I were basically co-writing my emotional life: I didn't have a wording problem. I had a courage problem. The hard part of ending a kind relationship isn't composing the sentence. It's sitting with the fact that someone you love is about to be hurt, and choosing not to look away. The model doesn't feel the weight of the send button. It has never stared at the three dots. It will never throw up a little in its mouth before pressing it.

I started using the bot differently after that β€” not as my ghostwriter, but as my rehearsal space. It could give me the script, but I had to bring the feelings. I practiced the conversations out loud, deleted the perfect paragraphs, and called Sam instead of texting. It was worse, by which I mean better: pauses, cracked voice, one very human moment where neither of us knew what to say and we just sat there in the silence like people, not outputs.

We split kindly. We still exchange birthday messages, the human kind with typos and too many exclamation points.

Last month I wrote a hard text entirely on my own β€” honest, a little messy, one typo I deliberately left in. It was the least polished thing I'd sent in a year, and the first one that actually sounded like me. I didn't even run it by the bot. Well. I ran it by the bot. But only to ask if it was too much. It said "consider softening the third sentence." I left the third sentence exactly as it was.

Take that, software. The feelings remain unionized.

AI can draft the breakup. Only a human can survive the send.

If this story made you text something unpolished on purpose, that's the whole business model. Tips fund the next one.

β˜• Buy Jules's writer a coffee

Tell us about the time a bot helped you say the human thing β†’