Copilot can now Analyse Images in Word, PowerPoint, and PDF Files

Microsoft 365 Copilot has long been able to draw on information from charts and images in its responses. With a recent update, it can now go a step further: analysing visual content directly from images embedded in Word, PowerPoint, and PDF files. Rather than processing text alone, Copilot now takes the visual content of these documents into account when forming its answers. That pays off in particular for reports, sales presentations, and project documentation — the kind of files where the most important points are often locked inside a chart. Here’s what the feature does and how to make it work for you day-to-day.  
Info Value: Analyse embedded images directly
UseCase Use Case: Analysing reports
Zeit Read time:
4 minutes
Schwierigkeit Difficulty: Intermediate

The feature appears in Microsoft’s current release notes as part of the ongoing development of declarative agents in Copilot. It marks a clear shift in direction: Copilot is moving beyond pure text processing toward understanding complete documents in context — visual content included.

  1. Open the Microsoft 365 Copilot app via the Start menu or in your browser.
  2. Switch to Chat and attach a file — Word, PowerPoint, or PDF — using Add content or by dragging and dropping it directly into the chat, for example.
  3. Ask a question about the document — for example, about a specific chart, screenshot, or other visual element. Copilot can analyse images embedded within the file, not just the text.
  4. Optional: For recurring or large knowledge sources, use Agents to connect documents on a persistent basis and query them directly whenever you need them.

In our own tests, Copilot proved reliably capable of processing visual content in documents, provided the files are added directly in chat — either via upload or the Add work content button. Both charts and structured graphics were interpreted meaningfully and integrated accurately into the response.

One pattern that emerged: Copilot prioritises semantically relevant content. Larger structures — charts, labelled axes, annotated process diagrams — are picked up consistently. Smaller details like icons or decorative visual elements can go unnoticed, especially if you don’t ask about them explicitly.

Our testing also suggests that re-uploading a document under the same filename doesn’t always trigger a fresh visual analysis. If you’re running tests across different versions of a file, use distinct filenames to make sure each upload is processed independently.

Overall, the feature is well-suited for extracting meaning from visual content in documents. It’s not a substitute for careful manual review of complex or high-stakes material, but as a first-pass tool it works well.

  • Watch what you upload: Uploaded files stay within the Microsoft 365 environment, but it’s still worth being selective about which documents containing sensitive content you share with Copilot.
  • Verify complex interpretations: For intricate charts or data-heavy visuals, it’s worth double-checking Copilot’s reading before acting on it.
  • Supported formats only: The feature currently works with .docx, .pptx, and .pdf files — other formats aren’t supported.
  • Room for improvement: If small labels or annotations matter, enlarge the relevant graphics before uploading — fine text in images can be missed.
  • Ask specific questions: Rather than “what does this document say?”, ask directly about a chart or a trend. Targeted questions produce noticeably sharper answers.
  • Use screenshots strategically: Images of UIs or process flows work well for quick explanations of tools or workflows.
  • Follow up: A second, more precise question often unlocks more depth in the analysis than the initial response provides.
  • Put existing reports to work: Presentations and PDF reports you already have become instant knowledge sources — no reformatting needed.

Image analysis in Word, PowerPoint, and PDF documents makes Microsoft 365 Copilot genuinely more useful with real working files. The clearest gains are in reports and presentations — content that would otherwise require manually reviewing slide after slide can now be queried directly. Upload a presentation or PDF report and ask for the key takeaway from a specific chart: that’s where the feature earns its keep.

There’s still room to grow, particularly around detail accuracy in complex visuals. The best way to find out where it adds value for your team is to start small — run it with a pilot group, collect honest feedback from real tasks, and go from there.