What to fill in
| Placeholder | What goes there |
|---|---|
| {{SAMPLES}} | three posts you have actually published |
| {{TOPIC}} | what we learned migrating off a vendor |
Why it works
The reason AI-written LinkedIn posts are recognisable is not the model. It is that people ask for “a post about X” and get the average of every post ever written about X.
Three samples of your own writing changes the output more than any other single thing. The explicit bans matter too: left to itself, a model will reliably open with a throat-clear and close with “What do you think?”, because that is what the training data does.
Delete the first line anyway. It is almost always still a run-up.