Episode Details

Most companies waste hours each week searching for information they already have. The shocking part? With Microsoft 365, that doesn’t need to happen. Viva Topics can automatically detect and connect your organization's knowledge, but only if you know how to guide it. Today, I’ll expose the curation and customization steps most admins skip — the ones that turn a vague AI guess into the exact insight your team needs in the moment they need it. This could be the difference between people using Search… and people actually trusting it.AI Topic Detection: Friend or Frenemy?You switch on Viva Topics for the first time and the AI starts surfacing topic cards everywhere. At first, it’s exciting. Then you notice a few that make you question if it’s been reading the wrong SharePoint sites entirely. The names seem familiar, but the content? Half of it feels like it’s been stitched together by someone who skimmed your library once and guessed the rest. That’s the core reality with AI-based topic detection — it can be brilliant in one moment and baffling in the next.Under the hood, Viva Topics is constantly scanning your Microsoft 365 environment. It’s looking at files in SharePoint, conversations in Teams, and even list items or pages you haven’t touched in a while. It pieces together people, files, and keywords into what it believes is a coherent “topic.” With no human intervention, it can connect the dots on projects or acronyms your team tosses around daily. That’s where it shines: picking up patterns you didn’t explicitly tell it to look for. But it’s also why it will sometimes lump unrelated content together. The AI doesn’t understand nuance — it matches on signals and relationships, not business meaning.Take “Project Falcon” as a real-world example. For your team, it’s an internal initiative tied to a specific rollout. The AI detects documentation from SharePoint, meeting notes from Teams, and timelines buried in old Excel files. So far, so good. But then it notices a pile of onboarding training documents from a completely separate “Falcon Security” course in HR’s library. Since both sets of content share a project name and some overlapping terms, the AI assumes they’re related. Now the topic card is split between two very different worlds, and users searching for one get smacked with the other.Mechanically, it’s not magic. Microsoft’s AI is looking at content titles, file metadata, authorship patterns, link structures, and how often certain terms appear together. In Teams, it can watch how a phrase is used in chat threads and who it’s associated with. When enough signals line up, it generates a topic. The process works well for straightforward, unique points of reference. It’s less reliable when names or concepts overlap, or when documents are poorly tagged.Leaving that process completely unchecked is like telling your email service to auto-file every message by its subject line. Sure, some will land where you want them. But you’ll also have meeting invites mixed with marketing newsletters, and bank alerts sitting next to cat meme threads. Without oversight, the order turns into noise faster than you think.And that noise isn’t just an annoyance — it’s a trust problem. If employees click into a topic card and 4 out of 10 times it’s irrelevant, they start ignoring it entirely. That drops usage of Search features and erodes any return you hoped to get from activating Viva Topics in the first place. AI can scale faster than any human could manage, but the minute accuracy dips, the credibility gap widens.The good news is you don’t have to choose between full automation and full manual control. There are ways to shape the AI’s output so it remains a net positive. In the first week of running Viva Topics, it’s worth doing a quick sweep based on clear criteria. If a topic is tied to active projects, repeatedly accessed documents, or critical business terms, it’s a candidate to keep. If it’s vague, built on outdated files, or has obvious con