Notion AI Prompt Templates

TL;DR: Good Notion AI prompts differ by surface, in-page writing and AI blocks, Autofill on a database property, and Enterprise Search or Research Mode across your workspace all take different inputs. Below are copy-paste Notion AI prompt templates for each, plus the admin-level controls an individual can't see or override.

What Does "Notion AI" Actually Cover in 2026?

"Notion AI" isn't one feature you prompt one way. Notion's own product page, titled "Meet your AI team" as of this writing, groups it into three jobs: Capture, Find, and Automate. That framing is worth noticing on its own: the positioning has visibly shifted from a single writing assistant toward a set of staffed roles (an agent, a search tool, a meeting-notes tool), and each of those roles takes a different kind of prompt. Underneath the framing sit distinct tools with distinct inputs, and a prompt that works well for one does nothing useful pasted into another.

Five surfaces are worth separating:

Plan pricing, for context: Free is $0, Plus is $10 per seat per month, Business is $20 per seat per month, and Enterprise is quote-only, per Notion's own pricing page. Notion AI proper, meaning the core features in that table, only ships on Business and Enterprise. A Plus workspace can try Notion AI, but it isn't buying ongoing access at that tier.

This matters for anyone searching "Notion AI prompts" hoping for a universal syntax to copy: there isn't one, because the surfaces don't share an input shape. A prompt written for a page (a wall of text) doesn't map onto a prompt written for a database property (one row's fields in, one value out), and neither maps onto a scheduled agent (a trigger plus a job description). Treating all three the same way is the most common reason a Notion AI prompt "doesn't work" when the real issue is that it was written for the wrong surface.

The templates below are grouped by which of these you're actually prompting, because "give me a Notion AI prompt" is really several different questions wearing one name.

How Do You Prompt Notion AI Inline (In-Page Writing)?

Notion AI inline is the one most people have already used: highlight text, or hit space on an empty line, and Notion offers to edit, extend, translate, or rewrite in place. AI blocks are the same idea made persistent. Typed as /AI Block into a page, you write the instruction once, and clicking Generate re-runs it against the page's current content, so the block updates rather than needing to be re-typed.

Because AI blocks can pull from "Specified context" (content you point it at) and an optional web search, the instruction is really two things: what to produce, and where to look. A template that only specifies the first half tends to produce generic filler.

Summarize this page in 5 bullet points aimed at someone who has never
seen it before. Pull specifics, names, dates, numbers, rather than
restating headings. If a section has no clear conclusion yet, say so
instead of inventing one.

Rewrite the "Next steps" section below in a more direct, less hedged
tone. Keep every action item and owner name. Cut phrases like "we
might consider" and "it could be worth," replacing them with a plain
verb and a deadline.

Draft a first-draft outline for [topic], using @[page] as the source
of truth for what we've already decided. Flag anywhere the outline
would contradict that page instead of silently picking a side.

Translation runs through the same surface: select ••• on a page, choose Translate, and pick a target language, or ask the Agent directly to "Translate this page." There's no separate translation prompt syntax; it's a menu action, not something you write a template for.

How Do You Write an Autofill Prompt for a Database Property?

Autofill runs one instruction against every row in a database column: think "summary," "priority," or "next action," computed automatically as pages are added or edited. Notion's help center splits Autofill into two tiers. Basic Autofill is included on Business and Enterprise plans at no extra cost. Custom Agent Autofill uses richer context and reasoning and draws on paid Notion credits, because it runs with full Custom Agent capabilities.

The instruction behaves like a template applied per row, so it needs to name which fields on that row it should read, not just describe the output:

Read the "Meeting notes" and "Attendees" properties on this page.
Write a one-sentence summary of the decision made, starting with the
decision itself rather than the discussion that led to it. If no
decision was reached, write "No decision, follow-up needed" instead
of summarizing the conversation.

Based on the "Description" and "Due date" properties, assign a
priority of High, Medium, or Low. High means due within 7 days or the
description contains "blocker". Medium means due within 30 days. Low
covers everything else. Return only the single word.

Based on the "Description" property, assign exactly one category from
this list: Bug, Feature request, Documentation, Question. If none of
the four clearly fits, return "Needs triage" instead of guessing.

Basic Autofill is the right tier for a rule you could describe to a new hire in one sentence, a fixed mapping like the priority or category examples above. Custom Agent Autofill is for a property that genuinely needs judgment across several other fields, and it's worth knowing which one you're configuring before a credits bill surprises you. The same underlying feature, Notion AI for databases, can also build the database structure itself (views, properties, relations to other databases) and write or edit formula properties, which is a separate action from Autofill but lives in the same part of the product.

How Do You Prompt Notion AI to Search Your Whole Workspace?

Enterprise Search answers a question using your Notion pages, plus any connected apps (Slack, Google Drive, GitHub, Jira, and others) an admin has enabled, plus the open web if that's turned on. It's the closest thing Notion has to retrieval over your own data rather than a single page's content, and it cites the sources it used.

Because it's retrieval-based, a vague question gets a vague, broadly sourced answer. A question that names the document type, the app, or the time window gets a narrower, more citable one:

Find the most recent decision about [topic] across Notion and Slack.
Quote the specific message or page where it was decided, and note the
date. If two sources disagree, show both instead of picking one.

Search connected apps for open items assigned to me that mention
[project name]. Group by app (Slack, Jira, GitHub) and skip anything
older than 30 days.

Answer using only [database name] as a source. If the answer isn't in
that database, say so explicitly instead of pulling from elsewhere in
the workspace.

Research Mode, a toggle inside Notion AI in the sidebar on Business and Enterprise plans, is built for the opposite kind of question, open-ended and exploratory rather than a single lookup. Use it for "build a report on X using our workspace and the web," not for "what did we decide on Tuesday." Whichever surface answers, check the citation before trusting the summary; Notion's own connector documentation notes new content can take up to 72 hours to be indexed after a connector is first set up, so a very recent Slack message may not show up yet even when the connector itself is active.

Enterprise Search also lets a workspace switch which underlying model answers a question, without losing the context already gathered, which matters if a team wants to compare how two models summarize the same retrieved sources rather than treating the answer as tied to one vendor. That model-agnostic framing is one of the few places Notion is explicit that "Notion AI" is a layer on top of several providers' models, not a single model of its own.

Can You Write a Prompt for a Custom Agent?

Yes, and it's the surface with the most vendor-documented structure. A Custom Agent's instructions run automatically on a trigger, so the prompt has to specify not just what to do, but when to do it and what it's allowed to touch. Notion's own guidance is to start with the job and outcome, then add concrete steps, inputs, and outputs, and it offers three ways to start: describing the job in chat and letting Notion draft the instructions, starting from a template, or writing the instructions field manually from a blank agent.

Real trigger types, from Notion's own Custom Agents documentation: a recurring schedule ("every day at midnight PST"), a Notion event (a comment added, a page added to a database, a property updated, a page removed, an AI Meeting Note finished), or a Slack event (a message posted, an emoji reaction, the agent mentioned).

Trigger: Recurring, every Monday at 9:00 AM.
Job: Post a weekly summary to the #team-updates Slack channel.
Read from: the "Sprint tasks" database, filtered to items updated in
the last 7 days.
Output: group by status (Done / In progress / Blocked), list task
name and owner under each, and skip posting if nothing changed.

Trigger: Notion event, property "Status" changed to "Blocked" in the
"Sprint tasks" database.
Job: Post to #eng-triage tagging the assignee, quoting the "Blocker
notes" property. Do not close or reassign the task yourself.

This is structurally the same discipline as writing an agent instruction file for a coding tool: name the trigger, the scope, and the boundary of what it's allowed to do on its own. If you already write AGENTS.md-style instructions for a coding agent, the shape will look familiar, and our AI agent prompt templates for autonomous workflows cover the same trigger-scope-boundary pattern outside Notion. Teams running comparable automations in Make.com will recognize the same job-then-steps structure in our Make.com AI prompt templates, with a slightly different vocabulary.

Custom Agents are Business/Enterprise only. The free period runs through May 3, 2026 at the time of writing; after that, Notion's product page prices ongoing use simply as "$10 per 1,000 credits." On Enterprise, key configuration and access changes to a Custom Agent (instruction updates, permission changes, integration additions) are recorded in Notion's audit log, which is one of the few places an admin gets a paper trail on what an agent's instructions used to say.

What Are Good Prompts for Meeting Notes and Page Translation?

Two more surfaces get lumped into "Notion AI prompts" even though neither one is really a prompt you write from scratch. AI Meeting Notes, currently in beta, transcribes a meeting and generates a searchable summary automatically; the "prompt" you control is really a follow-up instruction on the generated notes, not the transcription itself:

Using the AI Meeting Note above, write three follow-up emails: one to
the client summarizing next steps, one to engineering listing action
items with owners, and one to my manager as a two-sentence status
update. Keep each under 120 words.

Page translation, as noted above, is a menu action rather than a typed prompt, but you can still steer it through the Agent chat instead of the ••• menu, which is useful when you want the translation combined with something else in one step:

Translate this page into Spanish, and flag any product names or
proper nouns you left untranslated so I can check them manually.

Image generation is worth a brief mention here too, since it's technically a Notion AI prompt surface: while in beta, it's capped at 10 image generations or edits per 24 hours and 30 per 30 days per user, available on Business and Enterprise, and workspace owners can enable or disable it entirely.

Who Controls What Your Notion AI Prompts Can Actually See?

This is the part an individual prompt-writer usually can't see or override, and it's worth knowing before assuming a prompt is "just talking to an AI." A workspace owner, not you, decides:

• Whether AI Connectors exist at all. Connecting a third-party app to Notion AI requires being the workspace owner and having admin rights in that app. Notion documents connectors for Slack, Google Drive, Jira, Gmail, Microsoft Teams, SharePoint and OneDrive, GitHub, Microsoft Outlook, Linear, and Google Calendar; Notion Mail and Notion Calendar connect free on every plan, the rest require Business or Enterprise. An individual contributor can't add one alone.
• Whether Notion AI can search the web, and whether it needs confirmation before it does. Both are workspace-wide toggles in Settings, Notion AI.
• Whether image generation is enabled, and who's allowed to create Custom Agents. Both are admin-level switches on Business and Enterprise plans.
• How long the underlying data is retained. Notion states there is zero data retention with model providers on Enterprise plans, versus 30 days on other plans, and that customer data isn't used to train Notion's or its subprocessors' models unless you opt in.

A connected AI Connector "acts as you" in the app it's connected to. It can only surface what the person asking could already see there, and Notion says it honors existing permissions rather than granting an end run around them. But the connector's existence, its scope, and how far back it reads (about one year from the setup date, per Notion's own FAQ) are admin decisions, not prompt-level ones. If a connector is disconnected, the content it surfaced becomes unsearchable within about an hour and is deleted from Notion's side within a day, per the same documentation. If your workspace handles anything sensitive, confirm the connector setup with whoever owns the workspace before writing a prompt that assumes broader access than you actually have.

Why Don't These Prompts Travel Outside Notion?

Every template above leans on something Notion-specific: an @-mention of a page, a database property Autofill can read, a Custom Agent's own trigger vocabulary, a Connector only Notion exposes. Copy one into ChatGPT or Claude and the mechanics fall away, even though the underlying instruction, summarize this, prioritize that, post this on a schedule, is portable in spirit.

That's the real gap: the thinking behind a good Notion AI prompt (name your inputs, define the output shape, state the fallback for the edge case) works anywhere, and our prompt engineering cheat sheet covers those fundamentals in more depth. The prompt itself doesn't travel. If you want to keep the exact wording you refined for a property or an Agent and reuse it the next time you're prompting ChatGPT, Claude, or Gemini directly, that's a save-and-sync problem rather than a Notion problem.

It's also worth being honest about what "portable" would even mean here. A Custom Agent instruction that references a specific database and a specific Slack channel was never meant to run anywhere else; it's automation glued to one workspace's shape. What's actually reusable is the pattern underneath it: the trigger, the read, the output contract. Save that pattern in plain language once, outside Notion, and rebuilding the Notion-specific version the next time takes minutes instead of a rewrite from a blank page.

Prompt Architects keeps a personal prompt library that syncs across ChatGPT, Claude, and Gemini, so an instruction you finally got right for an Autofill property or a page rewrite doesn't have to be reconstructed from memory next time. Saving prompts is on the Pro plan and above; the free plan gives 5 prompt enhancements a day if you just want to tighten wording before pasting it anywhere.