Notion AI's Q2 2026 memory update changed what the product actually is.
The product we reviewed six months ago was AI assistance built into Notion. The product after the Q2 update is something more interesting: an AI that knows your workspace and can answer questions from it. If you use Notion seriously, this distinction matters.
I want to be specific about the Q2 2026 update because it's the reason this review changed meaningfully from our previous assessment. Before: Notion AI was a capable writing assistant that happened to live in your workspace. After: Notion AI learns from your existing pages, databases, and notes over time and can answer questions from that accumulated knowledge.
Ask "what did we decide about the product roadmap last quarter?" and Notion AI searches your workspace rather than admitting it doesn't know. Ask "which of our clients renewed in Q1?" and if you track clients in a Notion database, it queries that database. These are qualitatively different capabilities from what the product was before.
The memory feature in practice
After three weeks of use with the memory feature enabled, the most practically useful capabilities were: surfacing past decisions when starting new projects ("what did we decide about the mobile strategy?"), finding contradictions across different planning documents ("the Q3 plan says X but the team retrospective says Y"), and identifying relevant past context before meetings.
The memory feature improves with workspace content. After two weeks of normal use in a well-maintained workspace, the retrieval quality was good. In a poorly organised workspace with inconsistent tagging and mixed content types, retrieval was more hit-and-miss. The feature rewards good Notion hygiene.
Database querying: the underrated feature
Plain-English database queries are genuinely useful for teams using Notion as a lightweight CRM, project tracker, or content calendar. "Show me all the blog posts scheduled for next month that don't have a writer assigned" — this works on a well-structured database. "Which tasks are overdue in the marketing project?" — this works. The AI isn't writing SQL; it's interpreting your intent and using Notion's native filtering capabilities, which means it's reliable when your database structure is clear.
Writing quality: context is everything
For writing tasks within Notion — summarising a long document, drafting content that builds on existing pages, expanding an outline — Notion AI is well-suited and fast. For standalone writing quality on tasks where context doesn't matter, Claude and ChatGPT are better. The calculation is: if you're already in Notion and the context is already there, Notion AI is the right tool. If you need the best possible output quality, open Claude.
Pricing
- Basic Notion features
- No AI capabilities
- Limited block count
- 1 guest
- Unlimited blocks
- Unlimited AI usage
- AI memory feature
- Database AI queries
- Priority support
- Everything in Plus
- Advanced analytics
- SAML SSO
- Admin tools
- Audit log
Who should use Notion AI?
- Active Notion users who already maintain a meaningful workspace — the memory feature compounds value with workspace size and quality
- Teams using Notion as a knowledge base or wiki who want stored knowledge to become queryable rather than just searchable
- Project-driven teams who want AI that understands their project history and past decisions, not just their current document
- Non-Notion users — the AI quality doesn't justify switching from a working setup, and standalone Claude or ChatGPT are better general tools
- Teams whose primary AI need is content generation — Notion AI's generation quality is below Claude and ChatGPT
- Small teams with disorganised Notion workspaces where the memory feature won't find what it's looking for reliably