What are the different Copilots and which one am I using?

Microsoft ships several things called Copilot. The big three for file workflows: Microsoft 365 Copilot (the paid enterprise version that plugs into Word, Excel, Outlook, Teams, and SharePoint for organizations), Copilot Pro (the consumer paid add-on for individuals with Microsoft 365 Personal or Family), and the free Copilot chat at copilot.microsoft.com. They share the same model family but have very different data access.

Does Microsoft 365 Copilot really see my SharePoint and OneDrive?

Yes, when configured. In a Microsoft 365 Copilot deployment, the Copilot can ground answers in documents from SharePoint sites, OneDrive folders, Teams chats, and email you have permission to access. It respects the permissions already set on those files — it won't surface documents you can't already see. This is the feature that justifies the per-seat price.

Why does Copilot give me a weird answer about my Word document?

Three common causes. First, your doc uses images of text (like pasted screenshots) instead of actual text — Copilot struggles with those. Second, tables in the doc are actually text boxes or non-standard tables — Word supports several kinds, and Copilot handles the standard ones better. Third, the file is very long and Copilot is picking up the wrong sections. For important documents, a clean markdown version often produces better results than fighting with Word's formatting.

Can I upload PDFs to Copilot?

Yes, in both the consumer Copilot app and in Microsoft 365 Copilot through the chat interface. Quality varies the same way as other tools — clean PDFs are fine, scanned or heavily-formatted ones come through unevenly. Converting to markdown first is a reliable fix.

Will Copilot summarize a whole SharePoint site?

It can synthesize across documents you point it at, but it's not going to read thousands of files in a single prompt. For large SharePoint corpora, scoped questions work better than broad ones — ask about specific folders, document types, or recent changes rather than "what does our SharePoint say about X."

How to Upload Files to Microsoft Copilot: Types & Limits

The short version

Microsoft Copilot is in a different category from ChatGPT or Claude because its killer feature isn't the chat — it's the integration. Microsoft 365 Copilot can answer questions using the actual documents in your company's SharePoint, files in your OneDrive, conversations from your Teams, and messages from your Outlook. You don't upload anything; the Copilot is already inside the place your files live.

That changes the file-upload question. The interesting questions for Copilot aren't usually "how do I upload this?" — they're "how do I get better answers out of documents that are already there?" And the answer to that is often the same as for other AI tools: if the source document is messy, the answer will be messy. Clean documents in, clean answers out.

Which Copilot are you actually using?

Microsoft calls a lot of things Copilot. The ones that matter for file workflows:

Product	Who it's for	File access
Microsoft 365 Copilot	Enterprise / Business seats	SharePoint, OneDrive, Teams, Outlook, plus uploads
Copilot Pro	Individual / Family users with Microsoft 365	Works inside Word/Excel/PowerPoint on your documents; chat supports uploads
Copilot (free)	Anyone at copilot.microsoft.com	Chat uploads only — no Microsoft 365 grounding
Copilot in Windows	Windows 11 users	Chat interface; file handling matches the free Copilot

The per-seat Microsoft 365 Copilot is where the integrations live. If you're trying to get value from the SharePoint and Teams context, you need that tier. If you're an individual and you just want AI help inside Word, Excel, and PowerPoint, Copilot Pro is the right product.

How Microsoft 365 Copilot grounds answers

When you ask Microsoft 365 Copilot a question, it does two things in parallel:

Semantic search across your graph. Copilot searches Microsoft Graph — SharePoint sites, OneDrive folders, Teams chats, emails — for content you have permission to see that seems relevant to the question. This is the "work context" that makes the product different.
Grounded generation. The retrieved passages go into the prompt as context. The model generates an answer grounded in those passages and cites the source documents (with clickable links back to the original).

Two consequences worth noting. First, Copilot respects your existing permissions — it won't surface a document you can't open yourself. Second, retrieval quality depends heavily on your SharePoint hygiene. If your org has every document named "Untitled.docx" across ten sites, Copilot is going to have a rough time figuring out which one you mean.

What gets retrieved well

Documents with clear titles and section headings.
Recently edited files in active sites.
Word, PowerPoint, and Excel files with real (not image-based) text.
SharePoint pages with clean HTML content.
Teams messages and Outlook threads you participated in.

What gets retrieved poorly

Scanned PDFs and image-based documents — text isn't indexed.
Documents buried in inactive sites that nobody opens.
Files with generic names ("Report.docx", "v2 final FINAL.docx").
Attachments deep in old emails (sometimes surfaced, often not).
Content inside images or complex diagrams embedded in slides.

Inside Word, Excel, and PowerPoint

Copilot inside Office apps is different from chat-with-upload. It's operating on the document you have open, with the full file in context.

Word Copilot

Drafts, rewrites, summaries, and Q&A about the current document. It generally does well on well-structured Word docs — documents that use actual Heading 1 / Heading 2 styles, real bullet lists, and standard tables. Docs that are styled with manual bold-and-bigger-font headings or pasted-in screenshots for tables trip it up.

If you're getting bad summaries from a long Word document, check two things: are the headings real heading styles, and are the tables actual Word tables (not screenshots or text boxes)? If the answer is no, the fastest fix is often to export to markdown, clean up in any editor, and re-import — or just keep working in markdown. Our deep dive on DOCX to markdown covers the practical conversion patterns.

Excel Copilot

Analysis, formulas, pivots, and plain-English queries over your data. Works best when your data is in a proper Excel Table (Insert → Table, with defined column names) rather than a raw range. Copilot uses the table structure to understand your columns; without it, it's guessing at column meanings from cell values.

PowerPoint Copilot

Drafts decks from prompts or from a source document (usually Word). Can rewrite existing slides. More mixed results than Word/Excel in our experience — slide layouts and visual design are things the model has less native skill with.

Outlook Copilot

Summarizes email threads, drafts replies, and coaches tone. File handling is minimal — it works on email content, not attachments directly. For attached documents, download and use Word or chat upload.

A real before-and-after

A product manager uses Microsoft 365 Copilot to ask: "summarize the risks our legal team has flagged on the Q4 launch." Two scenarios:

Poor SharePoint hygiene

Files named "Legal Review v3 FINAL FINAL (Joe edits).docx"
Half the risk memos are PDFs pasted into email threads
Some are Loop components, some are Word docs, some are OneNote
Tables are screenshots in Word docs

Copilot gives a generic answer citing one document, misses two others entirely.

Cleaned up

Risk memos all live in one SharePoint folder with consistent names
Each memo uses real headings and tables (or markdown)
Scanned contracts converted to searchable text
A "Q4 Launch" SharePoint page links to all of them

Copilot produces a structured summary citing all six memos, with clickable links.

When converting to markdown helps with Copilot

Unlike the other AI tools in this series, Copilot is living inside a Microsoft-native world where DOCX, XLSX, and PPTX are the dominant formats. You won't typically be uploading markdown directly into Copilot. Where markdown still matters:

Cleaning up messy source material before it lands in SharePoint. If you're building a knowledge base for Copilot to ground on, convert scanned PDFs, old formats, and web-clipped content to markdown first, then save as Word or publish as SharePoint pages. Whatever Copilot indexes is only as good as what you put in.
Working with AI tools in parallel. Many people use Copilot for email and Office work, Claude for deep document analysis, ChatGPT for data work. Having your reference material in markdown means it ports across all three without format-specific problems.
Archiving for long-term AI access. Proprietary formats can rot. Markdown won't. If the knowledge in your SharePoint is going to matter for years, a parallel markdown archive is a good hedge.

Decision tree

Is the file already in SharePoint or OneDrive, and you're on Microsoft 365 Copilot?

Don't upload — just ask. Copilot will ground from the file directly.

Working on a specific document in Word, Excel, or PowerPoint?

Use the Copilot panel inside the app. Make sure the doc uses real styles, tables, and structure — that's what Copilot reads.

Messy PDF or scanned document?

Convert first. Save the cleaned version back to SharePoint so Copilot can ground on real text.

Building a knowledge base for Copilot to use going forward?

Invest in the hygiene: consistent file names, real headings, proper tables, text-based PDFs. Copilot's retrieval quality tracks your source quality directly.

Question needs long-context reasoning across a big document?

Copilot isn't the best tool for that. Convert to markdown and use Claude or Gemini.

Things people get wrong about Copilot

"It knows everything in my company's SharePoint"

It knows what Microsoft Graph has indexed and what you have permission to see. Files that are in obscure corners, haven't been touched in years, or are scanned images rather than text may as well not exist from Copilot's point of view.

"Copilot in Word will fix my messy doc"

Copilot will write. It won't re-structure a document that was built on manual formatting. If your doc needs real styles and tables, you still have to add them — or convert to markdown, clean up, and convert back.

"Turning on Copilot will make our team more productive immediately"

Copilot rewards organizations that have clean document practices, and it struggles in organizations that don't. The biggest predictor of Copilot ROI in practice isn't the tier you bought — it's the state of the SharePoint it's grounding on.

Where to go next

If you're using Copilot alongside other AI tools — which most serious users are — the comparison pieces matter: uploading files to Claude for deep-document work and uploading files to ChatGPT for analysis and experimentation outside the Microsoft stack.

If you're cleaning up documents before they land in SharePoint, the Word to markdown and PDF to markdown converters are what most teams use to get source material into a shape Copilot can actually read.

Using Files with Microsoft Copilot: 365, SharePoint, OneDrive