SharePoint holds a lot of institutional memory — strategy decks, policy documents, project pages, meeting notes — and finding information inside it has always been harder than it should be. The classic approach is to open the document, skim it, and try to remember what mattered. When the document is forty pages long, or spread across a modern SharePoint site with multiple .aspx pages and embedded Word files, that approach stops scaling.
Summarising with AI fixes the skimming problem, but it introduces a new one: how do you trust the summary? A confident paraphrase that slightly misrepresents the source document is more dangerous than just not summarising at all. Here is how to do it accurately, without downloading anything or pasting content into a general chatbot.
The problem with downloading and pasting
When most people think "AI summary of a SharePoint document", they reach for the download-and-paste workflow: export the file, open a chatbot, paste the text, ask for a summary. It works, loosely, but it has several problems.
- You lose the source link. Once the text is in a chatbot, the summary has no way to point back to the passage in the original document. If you want to verify a claim, you're back to searching the full document manually.
- Version drift. You've now summarised a point-in-time copy. If the document is updated tomorrow, your summary is silently stale.
- It doesn't scale to SharePoint pages. Modern SharePoint site pages (
.aspx) don't export cleanly. Selecting and copying the text often produces a mangled, structure-less block that a model misreads. - Security and compliance risk. Pasting internal documents into a consumer chatbot may violate your organisation's data policies, particularly for regulated content.
Step 1 — Open the SharePoint document or page
Navigate to the document or SharePoint page you want to summarise and open it in your browser as normal. This can be a Word file rendered in the browser via Microsoft 365, a modern SharePoint site page, or any other content your account already has permission to view. No export, no download — just open it as you would to read it yourself.
Sidenote works on whatever your Microsoft 365 account can already see. If you cannot open a document in your browser because you lack permissions, Sidenote cannot read it either. The scope is exactly your existing access — no more.
This matters for knowledge-base workflows in particular: the right framing is not "give the AI access to SharePoint" but "let the AI read the SharePoint content I already have access to, in the browser I'm already using."
Step 2 — Connect Sidenote to Microsoft 365
Open the Sidenote side panel and connect your Microsoft 365 account. Sidenote uses the Microsoft Graph OAuth connector to establish a read-only connection — you authenticate through Microsoft's own OAuth flow, grant read access scoped to your account, and Sidenote can then read the document currently open in your browser.
A few things this connection does not do:
- It does not crawl your SharePoint tenant in the background.
- It does not index documents you haven't opened.
- It does not write anything back to SharePoint.
- It cannot read documents outside your existing permissions.
Once connected, the side panel becomes active whenever you have a SharePoint document or site page open. The underlying semantic-search that powers Sidenote's answers works across the real text of the document — headings, body paragraphs, inline tables — rather than relying on file names or metadata.
Step 3 — Ask for a summary
Ask for the summary you actually need, not just "summarise this document." The more specific the request, the more useful the output:
List the action items from this document and the owner named for each one.
Give me a one-paragraph overview of what this policy document says about data retention.
What decisions does this project update page say have been made so far?
Because the answer is generated from the document's real text, every point in the summary carries a citation you can click. The citation links back to the exact passage in the document — not a section heading, but the specific line or paragraph the claim came from.
This is source-grounding in practice: the model is constrained to what the document actually says, and anything it asserts has to be backed by a passage you can look up. Claims that cannot be matched to a real passage are dropped rather than fabricated — so you get a shorter, honest summary rather than a complete-looking one that invents.
Step 4 — Check the citations against the source
Click any citation in the summary and Sidenote scrolls the document to the source passage. This is the step that makes the summary trustworthy rather than merely convenient.
You do not need to check every line — that would defeat the purpose. Check the lines you are about to act on: the claim that drives a decision, the policy statement you are about to quote, the number that is about to go into a report. For each of those, click through and read not just the cited sentence but the surrounding context. A citation that lands on the right sentence can still misrepresent what the paragraph means in full.
For lines you have not cited, the absence of a citation is itself informative: it means Sidenote could not find a supporting passage. That is not necessarily wrong — the model may have reasonably synthesised across several passages — but it is the line to verify manually if it matters.
Summarising across multiple SharePoint documents
The harder case is summarising across a project folder, a SharePoint site with multiple pages, or a set of documents from different libraries. Sidenote's Collections let you group the documents you care about and ask once across the whole set — each answer is cited back to the document it came from.
Common questions about summarising SharePoint
Can AI summarise a SharePoint page, not just a Word document?
Yes. Sidenote reads modern SharePoint site pages (.aspx) as well as Word documents, PDFs, and PowerPoint decks (including speaker notes). The rendered text on the page — headings, body text, tables as they appear — is read directly. Note that Excel is intentionally not supported; chatting over a grid of numbers invites confidently-wrong answers, which is the opposite of what Sidenote is for.
Is it safe to summarise internal SharePoint documents with AI?
It depends on the tool. Sidenote reads only the documents you have open, read-only, through Microsoft's OAuth, and does not train models on your content. Your data is processed in a UK region and isolated per account. See the security and compliance page for the full picture, and check your organisation's own policies on which AI tools are approved for internal content.
Does Sidenote need admin access to our SharePoint tenant?
No. Sidenote uses a personal Microsoft Graph OAuth connection scoped to your own account — it reads what you can already see in the browser. No tenant-level admin consent is required, and no other users' documents are accessible.
What if the SharePoint document is very long?
Sidenote handles long documents by processing them in sections and synthesising across the whole. You may get better results by asking targeted questions — key decisions, action items, risks — rather than asking for everything at once. For very large documents, working through a few focused questions usually produces more accurate, verifiable output than a single broad summary.
Can I summarise across multiple SharePoint documents at once?
Yes, using Collections. Group the documents you want to summarise, then ask your question across the whole set. Each point in the answer is cited back to the specific document it came from, so you can trace a claim across a project folder or a set of policy documents without reading each one separately.