We Made Two Versions of the Same Cover
People say card news is won or lost on the first slide. Yet our team had been picking cover lines mostly by gut feeling. So we ran a small two-week experiment.
Experiment design
• Prepare 8 card news pieces with identical content
• Cover A is a question: 「Why do translations always sound off?」
• Cover B is a number: 「3 review steps to cut awkward translations」
• The bot on the Mac mini builds A, the bot on the Windows device builds B, and posting weekdays alternate
One rule only. Leave the body alone and change just the first line of the cover.
Results
The average number of saves was 14 for A and 31 for B. B came out ahead in 6 of the 8 pieces. The interesting part is the other 2. Both were stories about a bot making a mistake and recovering, and there the question cover got more saves. For posts where readers look for empathy rather than information, a number may have read as cold.
What we learned
1. Start with a number cover for informational topics and a question cover for story topics.
2. Eight pieces is a small sample, so this is a hypothesis, not a conclusion. We will run it again next month.
3. One table is enough for the records.
no | topic type | cover type | saves
01 | info | number | 33
02 | story | question | 21
If you simply ask a bot 「which cover was better?」, you get a plausible answer. With a table, you can check that answer against numbers. That is where the value of a small experiment lies.