Imagine hiring a sharp new assistant and handing them a sticky note every morning. “Write five hooks.” “Fix this intro.” “Summarize that thread.” They’d do every job well. But they’d never get a chance to tell you what’s actually working.
Most of us use AI exactly like that, and I’m guilty of it too. This week I found a LinkedIn post that made me rethink the whole setup. It starts with one number: 41 saves per post became 57 in four weeks.
The author didn’t write a better prompt. They stopped giving the AI tasks altogether.
📋 The chatbot trap
For a long time, the original poster used AI on LinkedIn the way most people do. They sent short requests and got quick answers. That was useful, but they were still the one deciding what to post and why. The AI was a faster pair of hands, not a thinking partner.
So the creator tried something different. They set up an AI agent inside Slack and gave it no task list at all. It got one goal: more saves per post.
Why saves? A save tells you a post was worth keeping. It’s a stronger signal than a like, and it’s hard to fake.
🔍 What the AI did with a goal
This is where the story turns. The author stepped back and watched what the agent decided to do on its own:
- Found the pattern: It looked at the last 60 posts and saw that posts ending with one practical step got saved far more than posts ending on an opinion.
- Proposed a rule: Every post on the calendar should end with a step people can try that same day.
- Planned ahead: It drafted outlines for the next two weeks and flagged 3 older posts worth updating.
- Reported back: Every Friday, it posted a short note in Slack covering what worked, what didn’t, and what to try next.
The person never asked for any of those things. The AI started from the goal and figured out the tasks itself.
✋ The human stayed in charge
This part matters. Nothing went live without the author’s yes. They rewrote plenty and threw out one of the outlines completely.
So this isn’t a story about handing your brand to a bot. It’s about changing what you hand off. You keep the final call. The AI does the analysis and planning you probably never had time for.
💡 The lesson
A chatbot answers the question you ask. An AI employee asks what you’re actually trying to achieve. Give it the goal, and it works out the tasks.
Four weeks in, average saves per post went from 41 to 57. That’s about a 39% jump. It’s one person over one month, so treat it as a promising experiment, not proof. Still, the logic holds up.
The closing line from the post stuck with me: “Chatbots answer. Employees own outcomes.”
🛠️ How to try this yourself
You don’t need a special setup to test this. Any capable AI assistant can play the role if you change how you brief it:
- Pick one goal you can measure, like saves, replies, or click-throughs.
- Paste in your last 30 to 60 posts along with their numbers.
- Ask the AI to find what separates your top posts from your bottom ones before it suggests anything.
- Have it propose one rule for your content calendar, not ten.
- Set up a weekly check-in where it tells you what worked, what didn’t, and what to test next.
- Approve, edit, or reject everything before it goes live.
Here’s a simple prompt to start with: “My goal is more saves per LinkedIn post. Here are my last 60 posts with their save counts. Don’t write anything yet. Tell me what separates the top 10 from the bottom 10, then propose one rule I should test for the next two weeks.”
You can also try the author’s own finding right away. End your next post with one practical step people can try today. That’s the pattern that moved their numbers.
I love this idea because it’s so simple. The biggest gain didn’t come from a smarter tool. It came from a smarter question. Check out the full LinkedIn post to read the experiment in the creator’s own words.