Notion is using AI to automatically write your blog posts, job descriptions, and poetry

Someday soon, Notion will be able to write your notes for you. The company is starting to test a new feature today called Notion AI, which Notion CEO Ivan Zhao tells me could ultimately change the way people use the app — and do their jobs. With Notion AI, all you have to do is tell the app what you want (a blog post, a recruiting email template, a list of great business books to read), and the app actually creates the content right in front of your eyes.

The first thing Notion AI is for, Zhao says, is writing. Lots of work requires writing, some of it repetitive (like creating new job descriptions) and some of it creative (like writing blog posts or brainstorming new ideas). Zhao thinks Notion AI can help in both cases. “For many users in this space, it solves the cold start problem. And even more than that, it saves time.” Notion AI won’t write you a publishable blog post, but it can help you skip the crappy first draft phase and go straight into improving and polishing.

Notion is making use of the same large language models that companies like Google are starting to bake into search results and startups like Mem are using to structure your personal information. There are plenty of AI-powered writing tools out there, of course, like Grammarly and Jasper and Google’s new Wordcraft, and all are improving rapidly thanks to large language models. These models provide software with vastly more context so they can comprehend sentences and paragraphs rather than individual words, which makes them better at both understanding and writing text of all kinds.

It’s all very early, the tech and Notion’s implementation, and Zhao reminds me of this fact several times. What’s launching today isn’t so much a new product as a private alpha of a new product: the company plans to roll it out slowly and carefully but wants to start getting it out into users’ hands now.

I was the first person outside of Notion to get to try Notion AI. Here’s how it works so far: when you create a new page, you get a bunch of new options. In addition to “Empty page,” you get “Blog post,” “Pros and cons list,” “Outline,” “To-do list,” “Meeting agenda,” “Sales email,” and many others. Click on one, and it opens a prompt: “What should the blog post be about?” or “What do you want to make a pros and cons list about?” Type in your prompt, hit the blue “Generate” button, and Notion begins to fill out the page for you. If you like it, keep it. If not, try a new prompt, or just hit “Try Again” and watch it repopulate.

Here’s what Notion AI came up with automatically when I asked it to make a pros and cons list for running a marathon:

Pros

Cons

Not exactly groundbreaking, but not bad! Here are the first three paragraphs it spit back when I asked for a blog post about “the state of the smartwatch industry”:

In this case, Notion AI gets it directionally correct — it is indeed a blog post about the smartwatch industry — but it gets a lot of its facts wrong. This is where Zhao would remind me again that this is a private alpha with no set release date and that Notion is still learning as much as anyone. But clearly, the tech is a long way from finished.

One more: I asked Notion AI to write a sales email about why everyone should read The Verge. Here’s what I got:

I’m sold. 

In addition to automatic first drafts, Zhao says Notion AI can be a useful ideation machine. You can start brainstorming and then tell Notion AI to continue writing and have it fill out more possibilities. Or you can prompt it with something like “the three best documentaries about Formula One” and have it spit back a list you can work with. (In case you’re wondering: Senna, Drive to Survive, and 1. A pretty good list.)

Most of the data in Notion AI’s model comes from the broader internet. Zhao says the company is partnering with AI providers but declines to say who. Ultimately, though, the company plans to personalize the model to companies, even individual pages, so the AI can write in a company’s voice and make use of its internal information. This raises some serious privacy questions, which, again, is why it’s a private alpha.

Notion also imagines using AI for much more than just writing. “It can help you answer questions,” he says. “A lot of Notion users pay us because we’re a good knowledge base, and this is going to be a killer use case for knowledge bases.” You could just search for anything and Notion AI would paste the right information in place. Beyond even that, he says, “Connecting to your calendar, connecting to your task management — there’s crazy things we can do with this.”

Zhao has often talked about modeling Notion after Douglas Engelbart, who worked for years on ways to use technology to augment human capabilities. “For us,” Zhao says, “this is almost the purest version of that. Software is awakened a little bit, to understand you better, to truly augment you. It’s no longer just words on the page, where you save some to-do lists — it’s no longer text, it’s meaning.”

Zhao and his team are still figuring out what Notion AI can be good for and how to help users understand how to use the new tech. (One challenge of generative AI is that an empty text box can be overwhelming, but so can a thousand tips and pop-ups.) They’re opening the Notion AI wait list today, with plans to open Notion AI to some of the product ambassadors first, then more users over time. Zhao says he’s trying hard not to be in a hurry. But he’s convinced that now is the moment that generative AI goes from a nifty science project to a useful consumer tool. “I think this has on the order of internet or web-scale leverage, that you can feel in your life.” He’s been tracking the space for years, and the underlying tech is finally ready, he says. “And then just wait until next year.”