Notion AI

What Notion AI is

Most AI assistants live beside your work and ask to be fed context. This one is built into the workspace instead: Notion AI operates across everything a team already keeps there — pages, documents, tasks, databases — so the context arrives without being requested.

The vendor describes three movements — capture work in one place, find answers with references, let agents push work forward unasked — and the commercial fact follows directly: it is sold with the workspace, not as a separate subscription.

What you can do with it

Inside a page, the assistant improves writing, produces drafts, fills database properties, writes formulas, and includes a research mode returning a longer report on a topic.

Agents accept a task stated in conversation and act with the workspace plus linked applications as their backdrop, creating and editing pages and databases; custom ones fire on a schedule or a trigger, so recurring work happens without being requested. Search goes beyond the workspace into linked tools, returning answers with references and a verification marker on the pages used, so any answer can be traced home. Meeting notes can be taken with no bot entering the call, then transcribed, summarised and turned into follow-ups; the platform is described as model-agnostic, meaning a workflow can switch providers and keep its context.

Who it is for

Teams already running their work in the workspace — small businesses up to large enterprises — are the users, on the argument that AI helps most where the work sits.

Governance includes permissions over what the AI may see and do, an activity overview and usage dashboards — which shows the buyer shifting toward IT and operations teams inside larger organisations. Admins tracking what AI does across a workspace are another case, as are teams expanding their agent usage.

What to watch out for

Agents left to create and edit pages unattended can rewrite the record a team depends on, which turns the permissions defining their reach, plus review before anything consequential, into the settings that matter first.

On data, the policy is explicit and deserves a full read: customer data is stated not to be used for training, subcontractors are contractually forbidden from training on it, recognised security certifications apply, and enterprise customers can opt for zero retention together with health-data compliance, while lower plans hold content for a time. The commercial side is the part to watch: the paid workspace tiers include the core AI features, custom ones are credit-metered with a stated date when charging begins, and an agent halts once credits are exhausted — so a workflow depending on one should be monitored, not left alone. Beyond the metering arrangement, this page lists no prices — and the point is that AI spending here is tied to the workspace plan, so compare against the whole subscription rather than an add-on.