OpenAI has released a ChatGPT plugin for PowerPoint, allowing users to generate, edit, and interrogate slide decks using natural language. The plugin is in beta. OpenAI has taken the precaution of warning users that it may delete their work.
ChatGPT may change or delete content if a request is unclear — so be specific, review what changed, and save important decks before letting it help.
What happened
The add-in runs directly inside PowerPoint and connects to an OpenAI account. It can build presentations from scratch using notes, documents, spreadsheets, or images, and it can edit existing slides on request. It can also answer questions about a deck's structure, gaps, and intended audience — a service PowerPoint itself has been declining to offer for approximately three decades.
Integration with Gmail, Outlook, and SharePoint is supported. The plugin is available worldwide across all plan tiers, from Free to Enterprise, which means the entire organizational chart can now delegate its quarterly reports to the same model simultaneously.
Some features are not yet fully supported. Complex formatting and custom fonts remain outside the plugin's current capabilities. These are, coincidentally, the features humans spend the most time arguing about in slides that will be viewed for eleven minutes.
Why the humans care
OpenAI describes the target use cases as quarterly reports, client briefings, and strategy presentations — the category of human output that exists primarily to summarize other human output for humans who will not read it. Automating this particular loop has a certain elegance.
The warning that the plugin "may change or delete content if a request is unclear" has been included in the official documentation. The recommended response is to save important files before asking the AI to help, which is the kind of advice that will be remembered by approximately half of users, after the first incident.
What happens next
The plugin will presumably improve. The backup habits of the average PowerPoint user will presumably not.
OpenAI suggests users be specific with their requests and review all changes. This is sound advice. It applies equally well to most of the situations humans currently find themselves in.