Is Notion completely AI?

The Confusion: AI Everywhere, But Not the Foundation

Open any tech feed today and you'd be forgiven for thinking Notion had transformed into an AI company. Headlines about autonomous agents, $500 million in annualized revenue, and AI driving most of Notion's growth make it sound like the product itself is now built on AI. The short answer to the question, though, is no — Notion is not completely AI, and the clearest proof comes from Notion itself.

Look at Notion's own Free plan and you'll see what the company considers its foundation. The core features listed are databases with subtasks, dependencies, and custom properties, plus Notion Calendar, basic forms, and basic sites. None of these are AI features. AI appears on the list only as a "Trial of Notion AI" — a limited preview of capabilities like generating docs and autofilling databases, not the substance of the product.

This makes sense when you look at the timeline. As CNBC's reporting notes, Notion began as a document and workspace tool and has progressively layered AI on top of its core offering. The company was founded in 2013, shipped its first product in 2016, and didn't announce "Notion AI" until two weeks before ChatGPT launched in November 2022. That's a 2016-era workspace product with a fast-moving AI layer added on top — not an AI-native foundation.

The layering shows up in the pricing structure too:

• Full Notion AI requires Business or Enterprise plans; Free and Plus users get only a limited number of complimentary AI responses
• Custom Agents run on a metered plan of $10 per 1,000 monthly Notion credits after a free trial
• AI was bundled into business plans at no extra cost only after adoption crossed 50% of customers

Notion's own documentation draws the boundary even more sharply. The Notion Agent help pages explicitly list what the AI cannot do: it can't create database automations, can't edit comments, can't change permissions, and can't manage workspace settings. An AI-native product wouldn't need a "cannot do" list this long.

So the honest framing is this: Notion is a collaborative workspace with AI woven deeply into it. AI is now the company's primary growth engine — more than 50% of Notion's ARR comes from AI-enabled customers — but the workspace underneath predates all of it. That distinction matters as AI agents and their skills proliferate across tools like Notion, which is exactly the problem SkillzBench is being built to solve: independently verifying which agent skills actually deliver lift, and which carry risk.

How Deep the AI Layer Goes: From Writing Assistant to 20-Minute Agents

Notion's AI story started quietly: two weeks before ChatGPT launched in November 2022, the company announced Notion AI as a writing assistant for brainstorming, editing, and summarizing, built on an OpenAI model. From that modest beginning, AI has expanded through a rapid sequence of releases into something far more ambitious than a text helper.

The timeline shows a clear acceleration. In May 2025, Notion shipped AI products for summarizing meetings and searching corporate files. Then came the pivotal moment: on September 18, 2025, Notion 3.0 rebuilt Notion AI as Agents, capable of up to 20 minutes of autonomous work per task, executing multi-step workflows across hundreds of pages, databases, and connected tools. By mid-2026, the agentic push continued with External Agents — letting Claude and Cursor join workflows alongside human teammates — and a dedicated iOS app for Notion Agents.

Today, AI touches nearly every core feature:

• Workspace Q&A that answers questions by referencing your pages, databases, and wikis
• In-context editing — highlight any text and rewrite it in place
• AI database properties that automatically apply summaries, keywords, and translations to entries
• "Build with AI" database creation, where AI suggests properties and views from a text description
• AI Connectors that pull information from 14 connected apps, including Slack, Google Drive, Jira, Gmail, GitHub, and Linear, with citations back to sources

The architecture is also more flexible than most users realize. According to Notion's Agent documentation, users can switch between the newest GPT models, Claude, Gemini, and Grok via an "Auto" setting, and the Agent can act in external apps — posting in Slack as you, drafting Gmail, and connecting to arbitrary tools via MCP servers. That permission mirroring matters for trust: the Agent has the same access you do, and Notion states customer data is not used to train its models.

The business results confirm how central this layer has become. AI add-on adoption climbed from 10–20% of customers to 30–40% earlier in 2025, then crossed 50%, prompting Notion to bundle AI into Business and Enterprise plans at no extra cost. By the end of 2025, more than 50% of Notion's ARR came from AI-enabled customers, more than doubling year-over-year, as CNBC reported when the company crossed $500 million in annualized revenue.

For teams evaluating this landscape, the agentic direction is the takeaway — and it extends beyond Notion. Agents run on reusable skills and instruction packages, which is exactly why SkillzBench is building independent verification and benchmarking for agent skills. With published research finding roughly 26% of skills carrying at least one vulnerability, knowing which skills actually help matters. Skills, proven. Join the waitlist at skillzbench.com to follow the launch.

The Documented Limits: What Notion's Own Docs Say AI Can't Do

The fastest way to puncture the "Notion is completely AI" hype is to read Notion's own help documentation — because the company itself publishes a list of things its Agent simply cannot do.

According to Notion's Agent documentation, the AI cannot create database automations, templates, or page layouts, and it cannot build advanced database properties like formulas, rollups, or buttons. It also can't create or edit comments, share pages or change permissions, start reminders, or manage workspace settings. In other words, the administrative and structural backbone of a Notion workspace remains firmly human territory.

The limits extend to what the AI can actually see and compute. Notion officially describes its AI Connectors as "best for finding and summarizing information" — explicitly not meant to run complex calculations or perform data analysis (Notion AI Connectors docs). And the data the AI works from isn't live: initial ingestion from connected apps can take up to 36 hours, new content takes up to 3 hours to index, and access generally reaches one year back.

That sync-based architecture is the deeper constraint. As one competitor analysis puts it, Notion AI is "trapped in Notion" — it works from synced copies of your content rather than live external systems like CRMs or e-commerce platforms. It's a workspace-bound assistant, not a general-purpose agent. Even eesel.ai's review concludes Notion AI is "a great tool for thinking, but not a great tool for doing."

• No database automations, formulas, rollups, or buttons
• No editing comments, changing permissions, or managing workspace settings
• Connectors summarize information but don't run calculations or data analysis
• Content ingestion lags up to 36 hours — AI works from synced copies, not live data

The business reality behind these limits is telling. Full AI features are gated to Business and Enterprise plans, with Free and Plus users receiving only limited complimentary responses. Custom Agents run on a metered model of $10 per 1,000 monthly Notion credits after a free trial, and a 10-person team upgrading to Business pays roughly $2,400 extra per year.

Yet those tiers are where the money is: more than 50% of Notion's ARR now comes from AI-enabled customers, a share that more than doubled year-over-year. This is precisely why independent verification of agent capabilities matters — at SkillzBench, we're building benchmarks and trust scores so teams know which agent skills actually deliver before they pay for them. If you're evaluating AI agents and their skills, join the SkillzBench waitlist and get the verification layer first.

What This Means for You: Evaluating Notion AI and the Skills Behind It

Before you pay for any AI tier, map the features to how your team actually works — because Notion's AI is powerful but unevenly distributed across plans, and some of its most impressive capabilities come with real constraints.

Start with the tier math. Notion AI is only fully available on Business and Enterprise plans, while Free and Plus users get a limited number of complimentary AI responses. The pricing model has also shifted: Notion once sold AI as a $10/user/month add-on, but once adoption crossed 50% of customers, the company began bundling AI into Business and Enterprise plans at no extra cost. Business runs $20/seat/month billed annually, so a 10-person team pays roughly $2,400/year extra to unlock the full AI layer.

Then match features to workflows:

• Summarizing and searching — the sweet spot. AI Connectors pull from 14 apps including Slack, Google Drive, and Jira, with citations to sources.
• Database automation — AI Properties can autofill summaries and keywords, but the Agent cannot create database automations, formulas, or rollups.
• Multi-step external work — the Agent can post in Slack, draft Gmail, and connect to external apps via MCP servers, but connectors are explicitly "not meant to run complex calculations or perform data analysis."
• Time-sensitive data — connector ingestion can take up to 36 hours, and new content up to 3 hours to index, so the AI works from synced copies, not live systems.

There's a deeper evaluation layer here too. Notion's Agent supports reusable skills and MCP connections to external tools — which means its usefulness increasingly depends on third-party skills and integrations you didn't build. That's where independent verification matters. Published research has found that roughly 26% of agent skills carry at least one vulnerability, and right now nobody can easily tell which skills actually improve an agent's output and which introduce risk.

That's the problem SkillzBench is being built to solve: pinning every skill version, security-scanning it, and benchmarking it in a sandbox — with and without the skill — to publish an honest Lift figure in percentage points. As agents like Notion's take on more autonomous work, "does this skill actually help?" is a question worth answering with data, not vibes. Skills, proven — join the waitlist at skillzbench.com.