How KOPY generates replies that actually sound like you

The hardest part of building an AI reply tool isn't generating text. That's the easy part. The hard part is generating text that doesn't sound like AI wrote it.

Most AI writing today has a very recognizable fingerprint: overused words like "delve," "tapestry," and "nuanced." Sentences that start with "Certainly!" or "Great question!" Em dashes everywhere. Perfect grammar that no real person uses in a text message.

KOPY works differently. Here's how.

Banned vocabulary

The first layer of our prompt is a list of banned words and phrases that AI models reach for by default. Words like "leverage," "utilize," "transformative," and dozens more. We block them entirely and replace them with the simpler words a real person would use: "use" not "utilize," "start" not "embark," "help" not "facilitate."

We also ban structural habits: no em dashes, no semicolons in casual contexts, no bullet points in replies, no affirmation openers like "Absolutely!" or "Great point!"

Mode and tone stacking

A reply to a Slack message from your manager should sound completely different from a reply on Hinge. KOPY handles this through a layered prompt system: a Mode sets the register (Social, Work, Dating, Insight) and a Tone adjusts the specific angle (Funny, Agree, Rizz, Question, etc.).

Each mode has its own rules for punctuation, capitalization, and sentence structure. Social mode uses lowercase and omits periods at the end of messages. Work mode uses contractions but keeps standard capitalization. Dating mode prioritizes specificity over cleverness.

Energy matching

One thing humans do naturally that AI usually misses: matching the energy of the incoming message. If someone sends you "!!!! I can't believe it happened" and you reply with a calm, measured sentence, it reads as tone-deaf.

KOPY has explicit energy-matching rules: excited messages get matched enthusiasm, sad messages get lower energy, one-word messages get short replies. The model adjusts temperature per mode too, with Social and Dating at higher creativity and Work at lower.

Specificity over generality

The most common failure mode of AI replies is being too generic. "That sounds amazing!" is a reply that could be sent to literally any message. We call this the specificity test: if a reply could work as a response to any message, it's wrong.

Our generation instruction explicitly enforces this: the reply must reference at least one specific element from the incoming message. A place, a person, an activity, a detail. Not just the topic, but something precise.

The result is replies that feel like you actually read what they wrote and thought about it.