Notion AI Export — What's Saved, What's Lost, and How to Archive AI-Assisted Content

The three Notion AI surfaces and what each stores

Notion AI is not a single product — it's three distinct features with different storage and export behavior. Understanding which surface you're using determines what you can recover.

1. Inline AI writing (Ask AI, improve, summarize)

The most commonly used Notion AI surface. Triggered by pressing the Space bar in an empty block, or by selecting text and pressing Ctrl+J (Cmd+J on Mac). You type a prompt ("Draft a summary of this page", "Improve the following paragraph", "Write a decision record for choosing PostgreSQL"), Notion AI generates text, and you click Accept or Discard.

What's stored: The accepted output becomes a regular Notion block — a paragraph, heading, bullet list, or table. It is stored as part of the page, fully durable, and included in any page or workspace export.

What's not stored: The prompt you typed. The alternative drafts generated before you accepted one. The "Try again" regenerations. Once you accept or discard, only the final accepted text remains in Notion's storage. There is no prompt history sidebar, no session log, no way to retrieve what you asked for.

Export behavior: Exports with the page via Settings → Export, the Notion API, or any page share. The AI-generated text is indistinguishable from human-written text at the block level — there is no AI-block metadata in the export that marks content as AI-generated.

2. Notion AI Q&A (sidebar chat interface)

The second surface: a chat panel accessed via the lightning bolt icon or the "Ask AI" button in the sidebar. Unlike inline AI writing, this operates as a question-answer interface where you ask a question about your workspace ("What did we decide about the authentication system?" or "Find all pages mentioning the Redis migration") and Notion AI searches your workspace and responds with a summary and links.

What's stored: The questions and answers in the current browser session. Notion AI Q&A is session-based by design — it surfaces information from your pages on demand but does not maintain a persistent thread history. When you close the panel, the conversation is gone.

What's not stored: The conversation history. The next time you open Notion AI Q&A, it starts fresh with no memory of previous questions or answers. There is no Q&A history in the workspace export, no API endpoint for Q&A sessions, and no GDPR export category for Q&A thread content.

Export behavior: Nothing to export. The Q&A thread is ephemeral by design. If a Q&A response surfaces a useful summary you want to preserve, copy-paste it into a Notion page before closing the panel.

3. Notion AI meeting notes and templates

The third surface: Notion AI can generate structured content from meeting transcripts, audio recordings, or meeting templates — summarizing key points, action items, and decisions from a meeting input. This is functionally the same as inline AI writing: the output is placed in a page as regular blocks.

What's stored: The generated summary, action items, and decisions — as regular page content. The input (meeting transcript or audio) is processed and discarded; Notion does not store meeting recording files.

Export behavior: Same as inline AI writing — exports with the page. The source transcript may or may not be present depending on whether you pasted it into the page before running the summarization.

Exporting Notion workspace content

The standard Notion workspace export path exports all page content including AI-generated text as regular content:

1. Go to Settings → Settings and Members (or just Settings in the sidebar).
2. Click the Settings tab (or scroll to the Export section).
3. Select Export all workspace content.
4. Choose format: Markdown & CSV (best for Markdown files, retains heading structure), HTML (retains more formatting, including nested page structure), or PDF (print-quality but not machine-readable).
5. Enable Include subpages to ensure nested pages are included in the export.
6. Notion generates a ZIP file and emails a download link to the account owner.

What the export includes

• All page content, including AI-generated text accepted into pages
• Page hierarchy (subpages as subdirectories in the ZIP)
• Database records and their properties
• Inline images and file attachments uploaded to pages
• Table of contents structure

What the export does not include

• Notion AI prompt history (the questions you typed to generate content)
• Notion AI Q&A conversation history
• Draft content that was not accepted (discarded AI responses)
• Comments and discussion threads on pages (exported separately in some plan tiers)
• Integration data (Notion databases connected to external tools via integrations)
• Version history (page edit history is not in the export ZIP)

Plan-tier differences

The Notion API as a programmatic export path

For teams that want programmatic or automated access to Notion content — including AI-generated text — the Notion API (https://api.notion.com/v1/) provides direct access to page and database content.

Fetching page content

Page content is accessed via the Blocks endpoint:

curl -H "Authorization: Bearer $NOTION_TOKEN" \
     -H "Notion-Version: 2022-06-28" \
     https://api.notion.com/v1/blocks/{page_id}/children

This returns a paginated list of block objects. Each block has a type field (paragraph, heading_1, heading_2, bulleted_list_item, etc.) and a content object with the rich text array. For AI-generated text that was accepted into a page, the block type is the same as any other block of that type — there is no ai_generated: true flag or equivalent in the Notion API response.

Recursing subpages

Pages that contain subpages require recursive fetching: each child_page block gives you a page ID, which you then fetch with another /v1/blocks/{page_id}/children call. A simple recursive export script:

import { Client } from "@notionhq/client";

const notion = new Client({ auth: process.env.NOTION_TOKEN });

async function exportPage(pageId, depth = 0) {
  const blocks = await notion.blocks.children.list({ block_id: pageId });
  for (const block of blocks.results) {
    // process block content...
    if (block.type === "child_page") {
      await exportPage(block.id, depth + 1);
    }
    // handle has_children for nested lists, toggles, etc.
    if (block.has_children && block.type !== "child_page") {
      await exportPage(block.id, depth + 1);
    }
  }
}

The Notion API rate limit is 3 requests per second per integration. Large workspace exports via the API require backoff logic to avoid hitting the rate limit.

What the API cannot access

The Notion API cannot retrieve Notion AI Q&A conversation history (it is not stored server-side after the session ends), AI prompt history for inline writing, or discarded AI drafts. These are not exposed via any current API endpoint.

GDPR data portability for Notion AI content

Notion's GDPR data portability request (Article 20, right to data portability) is processed via Settings → My account → Privacy settings → Request data export, or by submitting a request at Notion's privacy portal.

The GDPR export includes:

• All workspace content in the same format as the Settings → Export workspace export
• Account metadata (email, account creation date, plan history)
• AI-generated text that was accepted into pages (as regular page content)

The GDPR export does not include:

• Notion AI prompt history (not retained server-side per Notion's data practices)
• Notion AI Q&A thread history (ephemeral, not stored)
• Model training data derived from your prompts (this is a separate GDPR Article 17 deletion right, not a portability request)

For enterprise organizations, GDPR data subject access requests (DSARs) are typically processed through the organization's admin or DPO rather than as self-serve requests. Notion's enterprise admin console provides member data export capabilities that allow admins to export content on behalf of members for GDPR compliance purposes.

AI training data opt-out

Notion uses interaction data to improve its AI features by default. The opt-out is in Settings → My account → Privacy settings → AI privacy settings. Opting out prevents future prompts from being used for training but does not retroactively remove prompts that were already used. This is relevant to teams with confidentiality requirements — the opt-out should be configured before using Notion AI for sensitive architecture discussions.

Notion AI vs ChatGPT and Claude for decision capture

The fundamental difference between Notion AI and dedicated AI chat platforms — relevant to any team using AI for architecture deliberation — is the interaction model.

The core issue for decision documentation: an ADR's most valuable sections — Alternatives Considered and the specific rejection reasoning — require understanding what options were evaluated and why each was rejected. Notion AI generates plausible-sounding output from a single prompt, but the reasoning that produced that output is inaccessible. ChatGPT and Claude conversations preserve the full deliberation: the alternatives the engineer considered, the constraints that drove rejections, the concerns that were raised and addressed. That conversation is exactly the raw material the ADR needs.

This is not a criticism of Notion AI — it is excellent for what it's designed for (inline writing assistance, summarization, formatting, first-draft generation). It is a specific limitation when the goal is to capture architectural reasoning for an ADR record. Use the right tool for each job: Notion AI for polishing and formatting decisions you've already made, ChatGPT or Claude for working through the decision itself.

Durability risks specific to Notion AI content

Notion users often assume page content is durable because Notion is their primary knowledge base. There are three durability risks that are higher for Notion AI content than for human-written content:

Accidental deletion during cleanup

AI-generated draft content that was accepted into a page but not identified as "a decision" is at risk during page cleanup sessions. Engineers editing a Notion page may delete AI-generated prose that was actually the only record of a design rationale — it looks like a rough draft or auto-fill, not a decision record. The lack of any AI-generated-content marker in the export makes it impossible to distinguish AI-generated blocks from human-written ones without reading the content carefully.

Free plan trash window

Free plan Notion workspaces have a 7-day trash window. A page deleted in error is permanently lost after 7 days with no recovery option. For engineering teams using Notion AI to draft architecture notes, this is a meaningful risk — the draft is visible in the editor, feels durable, and then disappears permanently if the page is deleted during a workspace reorganization.

Workspace deletion on member departure

When an engineer leaves a company, their personal Notion workspace (if separate from the team workspace) is eventually deleted. Any AI-assisted content they created in their personal workspace — architecture notes, draft ADRs, design rationale — is lost unless they exported it before their account was closed or it was accessible from the team workspace.

Using Notion as the ADR destination, not the deliberation surface

The workflow that preserves both the deliberation and the final record:

1. Deliberate in ChatGPT or Claude. Use a multi-turn conversation to work through the architecture decision — state the constraints, ask the AI to surface alternatives, evaluate each, identify the one that best fits. This conversation preserves the full reasoning chain.
2. Export the conversation. Export your ChatGPT history or Claude conversations. The WhyChose extractor processes the export and surfaces the decision-dense sessions, producing structured output (decision summary, alternatives, constraints, consequences) that maps onto ADR fields.
3. Write the ADR in Notion. Use the extractor output as the raw material for the ADR's Context and Alternatives Considered sections. Use Notion AI to polish the prose — improving readability, reformatting, filling gaps in the Consequences section. The output of the Notion AI polishing step is saved as regular page content.
4. Link the ADR page to the conversation export. Add a "Source" or "Related" property to the Notion database row pointing to the conversation export or the WhyChose reference ID. This closes the audit trail loop: the Notion ADR page is the clean final record; the conversation export is the evidence for the Alternatives Considered section.

This workflow uses each tool for what it does best. Notion is excellent as an ADR store: it supports databases with custom properties (Status, Deciders, Date, Tags), bidirectional links between ADRs via Notion's block reference system, and the shared workspace model that makes ADRs discoverable across the engineering team. The Notion ADR template page covers the specific database properties and template setup.