Most people think asking AI for feedback makes their work better. It feels smart. You get a second opinion in seconds, and it never gets tired of your questions.
I used to think that too. Then I saw a LinkedIn post that flipped the idea on its head, and I haven’t stopped thinking about it.
🎲 The number 7 experiment
The author opens with a simple game. Pick a number between 1 and 10. Picked?
Odds are you picked 7. According to the original poster, about one human out of three does.
Now ask Claude. Or ChatGPT, Gemini, Grok. The creator says the answer is always 7.
Try it yourself. It takes ten seconds, and it’s a little unsettling the first time you see it. People are biased. But as the post’s author puts it, AI is much more biased.
🤔 Why this matters more than it seems
With numbers, it’s funny. It gets less funny when you think about how often we ask AI for its opinion on real decisions. The author lists the questions most of us type every day:
- “Is this post good?”
- “Which logo is better?”
- “What should I call my product?”
Here’s where the common wisdom breaks down. If millions of people ask the same models the same questions, they all get nudged toward the same answer. Everyone ends up with the same 7.
And when everyone has the same 7, it stops being tasteful. The original poster sums it up in one sharp line: “If everyone can do it, no one wants it.”
Language models are trained to give the most likely answer. That’s great for facts and grammar. It’s terrible for taste, because taste is about choosing what’s not the obvious pick.
🔄 The contrarian rule: never ask AI for its opinion
So the advice is blunt. Never ask AI for its opinion. Use it to find the 7. Then go somewhere else.
I love this framing because it doesn’t tell you to stop using AI. It changes what you use it for. The model becomes a map of the average, so you know exactly where not to land. The creator adds that even the people who build Claude say so.
🛠️ How to put this into practice
The author points to six skills that make up being tasteful with AI. Here’s how I’d apply each one based on the post:
- Stop the flattery: Don’t ask “Is this good?” Ask Claude to list the three weakest parts of your draft and explain why a skeptical reader would bounce.
- See the 7 before you pick it: Ask, “What would most people write here?” Now you know the default answer, and you can steer away from it.
- Ban the clichés: Have the model list the phrases and ideas everyone in your niche is posting right now. Then tell it to avoid every one of them.
- Go deeper, not “better”: “Make it better” just gets you a shinier 7. Ask for a more specific example, a stranger angle, or a counterargument instead.
- Teach Claude your taste: Feed it examples of work you love and explain what makes them yours. Give it a reference point that isn’t the average of the internet.
- Notice when your taste expires: What felt fresh six months ago might be everyone’s 7 today. Check your go-to styles regularly.
💡 A quick example
Say you’re naming a new product. The old way is typing “Give me a name for my AI writing tool” and picking the favorite. The better way is asking for the 20 most predictable names first. You’ll probably see a lot of “Write”, “AI”, and “Genius” mashups. Now you know the crowded zone. Your real brainstorm starts outside it, with your own instincts and maybe a human friend or two.
The same trick works for logos, headlines, and LinkedIn hooks. Find the 7. Then go somewhere else.
🧭 My takeaway
This post changed how I think about AI feedback. The tool isn’t a judge of quality. It’s a mirror of the most common answer. Once you see that, you stop asking it to validate you and start using it to show you where the crowd is standing.
Save this before your next “is this good?” moment, and check out the full LinkedIn post to see the original take from the author.