Not everything worth teaching the AI deserves a help article. These three sources exist for the rest.
A snippet is text written for the AI and published nowhere. This is the right home for anything true but not presentable: how you word a known issue while a fix is pending, the exception to your refund policy, the fact that a feature is coming next quarter.
They're the fastest fix available when an answer is wrong. Two or three sentences take a minute, whereas a proper article takes half an hour and can wait until you know the question keeps coming.
Snippets aren't limited to short facts. One snippet can hold a whole checklist, runbook or troubleshooting guide, with Markdown headings and lists, and you edit it in Gleap whenever it changes. If you were about to split a long document into many small snippets, or upload it as a file just to keep it out of your help center, write it as one snippet instead. Tag it internal if customer-facing answers shouldn't use it.
If you already have this material in a spreadsheet — an FAQ list, a table of known issues — you can import snippets from CSV instead of typing them in one at a time.
Upload a PDF, Word document, PowerPoint deck, Excel sheet or Markdown file and it's converted to text for the AI to use. Useful for manuals, spec sheets and onboarding decks that nobody is ever going to rewrite as articles.
An upload is a copy. You can open it under Files and edit the converted text, but changes to the original document don't flow through. For material that keeps changing, maintain it as a snippet, or connect Notion with Sync content.
Two things convert less well than you'd hope. Slides often carry their meaning in the layout, so a deck can arrive as a list of disconnected phrases. Spreadsheets lose the relationship between a value and the row and column that gave it meaning.
Documents written as prose — manuals, policies, guides — come through well. For the other two, check what the AI actually answers before relying on them.
Paste a video URL, or a playlist, and the content is analysed and made answerable. If your onboarding is a video series, this turns it into something the AI can quote in text.
It works best on videos where someone explains something out loud. A screen recording with music and no narration contains nothing to extract, however clear it is to watch.
Use a snippet when you know the answer and just need the AI to know it too, or when you want to write and maintain a document inside Gleap. Use a file when the answer already exists as a document. Use YouTube when it only exists as someone explaining it.
And when a snippet keeps proving useful, that's a strong signal it should become a help article — the question is evidently common, and customers who search for it directly would find an article too.
