Everyone on a chat team has a favourite follow-up line. Usually it's the one they wrote. Fans don't care who wrote it. When he's gone quiet and a message lands, he either answers or he doesn't, and that's the most honest feedback you'll ever get on your copy.
So we treat each send as a vote.
How the vote works
Her follow-up lines go out exactly as written. That part matters. If the AI rewrote each one, you couldn't tell which words did the work. Every line keeps a count of how often it was sent and how often the fan answered.
In EmpireChat, the Follow-ups page ranks every line best first by that answer rate, inside its category. Lines that do well get picked more often. No fan gets the same line twice, and when he's heard everything that fits, the AI writes him one instead. You can switch any line off and add your own, and yours get scored like all the others.
The result is that your copy improves while you do something else. Weak lines sink and good ones rise, with no meeting about it.
One vote isn't a win
The first trap is small numbers. A brand new line goes out once, the guy answers, and it now has a 100% answer rate. It would sit above every line you've got.
So we pull each line's score toward the average until it's had a fair number of sends. Say line A has 1 answer from 1 send and line B has 30 from 100. B ranks higher, because we actually know something about B. As A collects sends, its own numbers take over. Good new lines still climb. They just have to earn it.
Compare lines that do the same job
A nudge sent minutes after a lively chat will always beat a line sent to a guy who's been silent for weeks. That's about the situation and has nothing to do with the words. That's why the ranking works inside a category. Comparing across them just tells you recent chats are easier.
The same page also compares kinds of follow-up: her lines, AI lines about him, longer ones, flirtier ones, ones that start a script, with or without a picture. That's often the more useful question. If one kind beats another across lots of lines, you know what to write more of.
We also run a proper split test on the biggest version of that question. Every guy stays in one group for good. One group gets follow-ups written by the AI about him, the other gets the usual mix of her lines. We score whether he answers within a day. No result to quote here, but it's how we'd settle it rather than by taste.
Keep the ballot clean
Bad sends lose votes for the wrong reasons, and they cost you a fan's patience as well.
- Wrong time of day. We caught lines about morning yoga going out at 1 a.m. his time. A fine line, sent at the wrong hour. Time-specific lines now only go out when they fit his clock, and not at all when we don't know his time zone.
- Same subject again. One fan kept getting asked about his cat's medicine. It's a sweet question the first time. Now she doesn't repeat a subject.
- Stop means stop. When he's said no or stop, follow-ups end. Those guys shouldn't be in the count at all.
Replies aren't money
A reply is a great first score because it's fast. You get an answer in a day, while a purchase can take a week. But a reply isn't the goal. A cheeky line can pull lots of answers and no sales. A line can get a reply that says "stop". That's why we track unsubscribes per line as well as answers.
Follow-ups do earn real money. In our first 8 days, follow-up resends averaged $22.95 a sale, on our own account, in a small sample. So once a line has a high answer rate, look at what those chats did next. A line that gets answers and leads nowhere goes back on the bench.
What to do with it
Once a month, open each category. Switch off the bottom lines that have had a fair number of sends. Add a few new ones. Copy what the top lines do, not their words: a question he can answer in one word, a detail from her day that fits his time. Then leave it alone and let the guys vote.