Notion AI Features 2026: Every Capability, Tested and Compared

Notion AI shipped more features in the first four months of 2026 than in all of 2025. The product moved from a text rewriting assistant to a programmable agent platform with code execution, voice input, cross-page awareness, and sub-3-second autofill. This guide covers every Notion AI feature available in 2026, what each one actually does in practice, and where the product still falls short.

All Notion AI Features in 2026

| Feature | Available Since | Category | What It Does | Practical Value | |---|---|---|---|---| | Workers for Agents | April 2026 | Computation | Run custom JS/Python inside Notion | Turns AI from text tool into compute platform | | 50-page context window | January 2026 | Core AI | Tripled from 20 to 50 pages | Full-project summaries without truncation | | Cross-page AI blocks | March 2026 | Core AI | AI blocks resolve linked page content | Dashboard summaries reflect linked docs | | Sub-3s autofill | January 2026 | Performance | 70% latency reduction on autofill | Batch autofill across large databases is practical | | Relation-aware autofill | February 2026 | Database AI | Autofill traverses one relation level | Generated values use linked data, not just local | | Voice input for AI | April 2026 | Input | Speak prompts on macOS and Windows | Long prompts take seconds instead of minutes | | Custom prompt templates | March 2026 | Workflow | Reusable workspace-level prompts | Teams standardize AI outputs across members | | AI meeting notes | April 2026 | Meetings | Trigger notes from Cmd+K | No navigation required to start capture | | Custom meeting instructions | April 2026 | Meetings | Set tone, format, exclusions per workspace | Consistent meeting summaries across teams | | AI-powered search | February 2026 | Search | Natural language queries across workspace | Find information without remembering page names | | Smart suggestions | March 2026 | Editing | Contextual writing suggestions in docs | Proactive edits without manual prompting |

How Notion AI Features Are Organized in 2026

The diagram shows three processing paths. Text generation and database AI features are stable and generally available. The agent and worker path is new in April 2026 and runs custom code in a sandboxed environment. All three paths share the same expanded context engine that resolves cross-page links.

Feature Deep Dive: What Each Notion AI Feature Does

Workers for Agents

Workers are the most significant Notion AI feature added in 2026. They let AI agents execute custom JavaScript or Python functions inside Notion's infrastructure. Before Workers, asking Notion AI "what is the average deal size in Q1?" produced a text estimate. With Workers, a developer writes a function that queries the database directly and returns a precise number.

Workers run in sandboxed V8 isolates. Each execution has a 10-second time limit and a 128MB memory cap. Workers can make HTTP requests to approved external domains but cannot access filesystems or maintain state between invocations. The feature is currently in developer preview.

50-Page Context Window

The context window increased from 20 pages to 50 pages in January 2026. This directly affects the quality of every AI interaction. A project wiki with 35 pages can now be fully summarized in a single prompt. Before this update, the AI silently dropped content beyond 20 pages and produced incomplete summaries without warning the user.

Cross-Page AI Blocks

AI blocks now resolve one level of links. If a dashboard page links to five sprint pages, an AI block on the dashboard automatically includes content from all five sprints when generating its output. Before this March 2026 update, the AI block only saw the dashboard page itself.

Voice Input

Available on macOS and Windows since April 2026. Voice input works anywhere you can type a Notion AI prompt: Cmd+K, AI blocks, inline AI requests. A complex prompt that takes 30 seconds to type takes 5 seconds to speak. The speech recognition runs through Notion's servers, not local processing.

Relation-Aware Autofill

Autofill now traverses one level of database relations. If a Tasks database has a relation to a Projects database, autofill on a task can pull project metadata (owner, priority, deadline) into its generated value. Before February 2026, autofill only saw properties on the row it was filling.

Custom Prompt Templates

Workspace admins can create reusable prompt templates available to all members. This standardizes AI output across teams. Instead of each person writing their own summarization prompt with different instructions, the team uses a single "Weekly Summary" template that produces consistent formatting and content.

AI-Powered Search

Natural language search queries work across the entire workspace. Instead of remembering exact page titles, you can ask "the design spec for the mobile onboarding flow" and Notion AI finds the relevant pages. The search uses the same context engine as other AI features.

Notion AI Features Comparison: 2025 vs 2026

| Capability | 2025 | 2026 | Change | |---|---|---|---| | Context window | 20 pages | 50 pages | 2.5x increase | | Cross-page awareness | None | One level of links | New capability | | Code execution | None | Workers (JS/Python) | New capability | | Autofill speed | 8-12 seconds | Under 3 seconds | 70% faster | | Relation traversal | None | One level | New capability | | Voice input | None | macOS and Windows | New capability | | Prompt templates | User-level only | Workspace-wide | Team standardization | | Meeting notes | Manual trigger | Cmd+K shortcut | Reduced friction | | Custom meeting format | Not available | Per-workspace settings | New capability |

What Notion AI Features Cannot Do in 2026

Despite the significant updates, several capabilities remain absent:

Cross-application workflows. Notion AI operates inside the Notion workspace. Workers can call external APIs on approved domains, but they cannot open Slack, send emails, update a CRM, or interact with any application outside Notion. Real-world workflows rarely stay within a single tool.

Offline processing. Every Notion AI feature requires an internet connection. Text rewrites, summaries, autofill, and search all fail without connectivity. Notion has not announced any plans for on-device AI processing.

Custom models. There is no option to bring your own model, fine-tune the existing model for domain-specific terminology, or choose between different model providers. All AI requests go through Notion's model infrastructure.

Granular permissions. AI access is workspace-wide. You cannot restrict AI features to specific pages, databases, or team spaces. If AI is enabled, it can access everything in the workspace.

Persistent Worker state. Workers cannot store data between executions. Multi-step automations that need to remember previous runs require external state management, which limits the complexity of what Workers can accomplish independently.

Bridging the Gap: Desktop Agents for Cross-App Workflows

Notion AI features in 2026 are powerful within Notion but stop at the application boundary. For workflows that span multiple tools, desktop AI agents operate at the operating system level.

A desktop agent like Fazm reads screens, types into fields, and clicks buttons in any application. Where Notion AI needs API integrations and domain allowlists, a desktop agent uses the same interfaces you use. The two approaches work together: Notion AI handles in-workspace intelligence while a desktop agent handles the cross-app steps.

Which Notion AI Features Should You Use Today?

The stable features (context window expansion, autofill improvements, relation awareness, cross-page AI blocks, search) are ready for production use. Enable them across your workspace immediately.

Workers for Agents should be treated as experimental. Build internal prototypes and personal automations, but do not ship customer-facing products on the developer preview. The API surface and pricing will change before general availability.

Voice input and meeting features are low-risk improvements. Configure custom meeting instructions for your workspace and start using voice prompts for complex AI requests.

For anything that requires crossing application boundaries, pair Notion AI with an external automation layer rather than waiting for Notion to solve cross-app workflows.

Frequently Asked Questions

What are the new Notion AI features in 2026?

The major Notion AI features added in 2026 include Workers for Agents (custom code execution), a 50-page context window (up from 20), cross-page AI blocks, sub-3-second autofill, relation-aware autofill, voice input, workspace-level prompt templates, and improved meeting note capabilities.

Can Notion AI run code in 2026?

Yes. Workers for Agents, launched in April 2026 as a developer preview, let AI agents execute custom JavaScript or Python inside sandboxed V8 isolates. Workers have a 10-second time limit and 128MB memory cap, and they can call external APIs on approved domains.

How many pages can Notion AI read at once in 2026?

Notion AI can read up to 50 pages in a single context window as of January 2026, up from the previous 20-page limit. Cross-page AI blocks can additionally resolve one level of linked pages beyond this window.

Does Notion AI work across other apps?

Not natively. Notion AI features operate within the Notion workspace. Workers can make HTTP requests to approved external domains, but they cannot control other applications. Cross-app automation requires external tools like desktop AI agents or integration platforms.

Is Notion AI included in all plans?

Notion AI requires a paid AI add-on or a plan that includes AI capabilities. Workers for Agents pricing has not been finalized during the developer preview. Check Notion's current pricing page for the latest plan details.

This guide covers all Notion AI features through April 2026. Last updated April 13, 2026.

Fazm is an open-source macOS AI agent. Open source on GitHub.