Your AI already knows Notion. It just doesn’t know you.

Most people connect Claude or ChatGPT to Notion, ask for a summary, and call it a day. A smaller group writes the AI a rulebook first, and their pages come out the same way every time. I just watched a video where the creator, productivity YouTuber Jeff Su, shows exactly how to set that up, and it changed how I think about AI and my tools.

Here’s the frustrating part he points out. Almost any AI can connect to Notion now, and even weak models summarize notes well. So getting a page created isn’t the hard part. The hard part is getting a page created the way you like it, in the right database, with the right template and the right tags, every time.

The Key Idea: Treat Your AI Like a New Assistant

The author explains it with a simple thought experiment. Picture life before AI. You used Notion by hand and built your own habits along the way. Maybe every task name starts with a verb. Maybe every task has three bullet points of context underneath.

Now picture hiring a human assistant to run your workspace. The first thing you’d do is teach them your preferences. If you didn’t, they’d organize everything their own way.

AI works the same way. A Notion connector means the AI knows how Notion works. It doesn’t mean the AI knows how you work. That second part needs a written rulebook, which the expert calls a “Notion Conventions” file.

Old Way vs. New Way

The creator ran a side-by-side test that makes the difference really clear.

The old way: connector only. He pasted the same prompt into two fresh Claude chats, with Notion connected and no files attached. Both chats created the task. But the results didn’t match:

• The first page had no icon and no section headers.
• The second page had an icon and a header, but it stuck an unneeded “Friday” into the task title.
• Neither one followed a predictable layout, so he’d have to fix both by hand.

The new way: connector plus rulebook. He ran the same kind of request in two new chats, this time with a folder holding his Notion Conventions file mounted. The wording inside each page was a little different, but the structure was the same every time:

• The task name starts with a verb.
• A pinned callout gives a short, clear description.
• The topics to cover sit underneath as bullet points.

That’s the whole point. You don’t need identical sentences. You need a predictable shape, so you can trust the output without double-checking it. The expert says this “consistency in output” gives him peace of mind, and I totally get that.

It’s not just a Claude thing either. He repeated the setup in ChatGPT to show it works across tools, and says Gemini works too. The rulebook is the magic, not the model.

🛠️ How to Build Your Own Notion Rulebook

Here’s the process the original poster walks through, step by step.

1. Start with a conventions file. He shares a starter Notion Conventions doc, which you download as a markdown (.md) file. It already includes basic rules, like “confirm the page or database before starting any work” and “if a database has templates, use the closest match instead of a blank page.”
2. Create a dedicated folder. Make a folder (he calls his “Notion Ops”) and drop the conventions file inside. Then create a project in ChatGPT or Claude and mount that folder so the AI can read and update it.
3. Show the AI what good looks like. Find one page in Notion that matches your ideal structure. For tasks, his has a checkmark icon, a verb-first name, a pinned callout, and context bullets. Paste that page’s link along with his teaching prompt. The AI studies it and writes your task preferences into the conventions file.
4. Test it right away. Open a fresh chat with the same folder mounted and ask for a task. His fun example was a reminder to inspect “three Terminators” a week before they travel back in time. The AI nailed the due date, icon, callout, and layout.
5. Level up to whole databases. For his notes database, he pasted the database link instead of a single page. The prompt tells the AI to study every property and template. One tip that matters here: explain what each tag means. For example, “research” applies when there’s a source like a website or PDF. Without that, the AI is just guessing. The payoff was great. He asked for a newsletter summary and a meeting prep page for a chat with Jensen Huang. The AI tagged the first as “research” and the second as “notes,” and both used his key takeaways, action items, and notes layout.
6. Scale it with an interview prompt. For the rest of your databases, the process stays the same. Paste the database link into his template prompt. The AI reviews properties, templates, and recent entries, explains how it thinks you use them, then asks about anything unclear. In his case, Claude asked him to explain the difference between “pending review,” “confirmed,” and “published.” Once you answer, it saves the rules.

🚀 The Advanced Move: Whole Workflows

This is where I was blown away. The author has his full project management workflow written into his conventions file. So he asked Claude to create a Notion project for a live Q&A workshop.

Claude filled in the project properties, the North Star, and the win conditions. It also created a notes thread, the five deliverables every workshop needs, and linked tasks, all in his preferred format. Doing that by hand would mean writing each piece and linking everything back together yourself.

Why Not Just Use Notion AI?

Fair question, and the creator answers it. Notion AI can do a lot of this. But his setup puts Claude, ChatGPT, or Gemini at the center, with a personal folder of context about how he works. Notion becomes one of many tools that main assistant manages. He still uses Notion AI for things like reviewing copy and bulk database migrations. Since his work goes way beyond Notion, a central AI with portable rulebooks just makes more sense.

A Few Things to Watch Out For

• Keep the folder mounted. Every demo depends on the AI having access to the conventions file. Start a chat without it and you’re back to random results.
• Do the upfront work. Vague tag definitions lead to wrong tags. Spend the extra minute explaining your choices.
• Expect small differences in wording. The structure stays fixed, but the sentences won’t match word for word.

The big lesson goes way beyond Notion. Spend a little time teaching your AI your preferences once, save them in a rulebook, and you get consistent results from then on. The prompts and starter file are linked in the original video, so check out the full walkthrough and set up your own rulebook today.