Why Notion AI does not solve this
Notion AI answers questions inside Notion. That is genuinely useful, and it is also the limit. It cannot hand your workspace to Claude when you want a second opinion from a different model, and it cannot see the repo, the PDFs, or the client emails that live outside Notion. The moment your question spans more than one system, you need the content somewhere portable.
The fix is to take the pages out as text. Once your Notion content is a structured file, every model can read it, and you decide which model that is.
Getting the pages out of Notion
There are two routes. You can connect your workspace to Riflet with a Notion connection token and pick pages from a live tree, which pulls current content every time you rebuild. Or you can export pages as Markdown and open the folder, which needs no setup but gives you a frozen snapshot.
Setting it up: the step-by-step for both routes, including where to find your Notion connection token and the Content Access step most people miss, lives in how to connect Notion to Riflet. The rest of this guide assumes the pages are in front of you and asks the harder question: which ones should go in?
Which pages are worth including
Include the project you are actually asking about, and nothing else. If the question is about a client redesign, that means the brief, the meeting notes, the decision log, and the deliverables tracker. It does not mean your reading list, an unrelated team's standups, or last year's archive. Every irrelevant page you add is a page the model has to rule out before it can answer.
Structured pages carry the most weight per token. PRDs, specs, decision logs, and meeting notes with named owners and dates give a model the kind of detail it can cite back to you. Loose brainstorming pages, half-finished drafts, and pages that are mostly embeds rarely earn their place.
One useful test before you add a page: could you say in one sentence what question this page helps answer? If not, leave it out.
How much of a workspace actually fits
A typical Notion page runs 500 to 2,000 tokens. Against a 1,000,000-token context window on Claude Opus 5, or 1,050,000 on GPT-6 Astra, that is several hundred pages in theory. Most workspaces fit with room to spare, which is exactly why people over-stuff them.
Fitting is not the goal. Models get less reliable at picking the right detail as the surrounding noise grows, so a tight 30,000-token bundle of the right pages routinely beats a 400,000-token dump. The budget matters most when you combine sources: a Notion export plus a repo plus a scraped docs site adds up quickly. Riflet shows a running token count against the model you pick, so you can see the total before you commit to it.
Notion export quirks to watch for
- Linked database views export as empty files. The data lives in the source database, not the view, so the view comes out as a shell. Uncheck them.
- Inline database references flatten into bare links. If a page's value is the table it embeds, export the database itself rather than the page that points at it.
- Sub-pages become nested folders. This is helpful, not a bug: the folder structure is your page hierarchy, and it is how you find the branch you actually want.
- Extra blank lines and stray formatting are common. They cost a few tokens and no accuracy. Not worth cleaning by hand.
- File names carry the page ID. Notion appends a long hash to each file name. It looks untidy and does no harm, since the heading inside the file is what the model reads.
A connected workspace sidesteps most of this, because content comes from Notion's API as blocks rather than through the export pipeline.
What to ask once the workspace is loaded
The questions worth asking are the ones that were expensive before, the ones that need someone to hold twelve documents in their head at once and spot where they disagree.
I have given you the meeting notes, the project brief, and the technical spec for the Acme redesign. List every decision made about the navigation structure, who made it, and flag any decision that conflicts with what the technical spec describes.
Cross-referencing, contradiction-hunting, and timeline reconstruction are where a merged bundle pays for itself. Asking the model to summarize a single page is not.
Keeping the context current
Rebuild when the facts change, not on a calendar. A context bundle should hold what is stable about a project, so a decision reversal or a new spec is a reason to rebuild and a typo fix is not.
On a connected workspace, rebuilding pulls current page content, so the bundle is only ever as stale as your last export. On the export route you re-export from Notion first, which takes a couple of minutes and has one genuine advantage: you keep the old files, so you always know which version of your workspace a past conversation was working from.
Turn your Notion workspace into AI context
Riflet connects to Notion, or reads your export, and bundles the pages you choose into one portable file for any AI.
Download Riflet freeNotion context: common questions
Should I export Notion as Markdown or PDF?
Markdown, in almost every case. Models read Markdown natively and use its headings, lists, and tables to navigate long context. PDF export bakes your content into page layout, and the text an AI recovers from it is flatter and often out of order. Use PDF only when the visual layout is the point.
How many Notion pages can I give an AI at once?
A typical Notion page runs 500 to 2,000 tokens, so a 1,000,000-token context window holds several hundred pages on paper. In practice you want far fewer. Relevance beats volume: 20 well-chosen pages usually produce better answers than 300 pages of everything you have ever written.
Is it better to export the whole workspace or select pages?
Select pages. A whole-workspace dump buries the pages that matter for your question under unrelated projects, personal notes, and archived material. Pick the project you are actually asking about, then add adjacent sources like a repo or a docs site if the question needs them.
Why do some pages from my Notion export come out empty?
Linked database views usually export as empty or near-empty files, because the underlying data lives in the source database rather than the view. Inline database references can also come through as bare links. Skim the export and uncheck the empty shells before you build the bundle.
How often should I rebuild the bundle?
When the facts change, not on a schedule. A context bundle should hold the stable knowledge about a project. If you use a connected workspace, a rebuild pulls current page content automatically. If you work from an export, re-export when key pages have moved on.
Can I combine Notion pages with code and documents?
Yes, and it is usually the point. Notion carries the strategic context, the brief, the decisions, the meeting notes. A GitHub repo and a docs site carry the technical detail. Riflet's Pro plan puts all of them in one bundle so the AI reasons across the whole project instead of half of it.