AI Status Reports: How to Use AI Summaries Without Sending Confident Lies | Sanjana

Introduction: AI Status Reports and the Trust Trap

[0:00]Hi, I'm Sanjana, a project manager at Humanitarians AI. Here's something that's quietly become normal. The AI in your project tools will now write your status report for you. One click and out comes a clean paragraph you can send to your boss or your client. It sounds great. But a status report has your name on it. So, the real question isn't whether the AI can write it. It's whether you can trust what it wrote before you hit send. Here's the whole idea in one breath. It's tempting to treat the AI's status report as the status itself. It isn't. It's a draft built only from what's on your board and a confident paragraph can hide an empty or stale one. Your job is to verify it

Sourced, Current, and Complete: The Three Verification Checks Real Sprint Demo: Verification Pass and Adding Unlogged Risks The Green Trap Demo: 3 Fails and the Danger of Outdated Boards The Prompt Rule: Human Ownership and Verification Prompts

[0:42]Before you sign it. So, here's the situation. Every tool now has a summarize this sprint button and the paragraph it gives you is genuinely good writing, clear, confident, ready to send. That's the trap. The AI sets the status to green by counting closed tickets, but green isn't a fact, it's a judgment about whether you'll actually make it. The AI can't see the vendor who's gone quiet, the engineer who's out, the call that moved the deadline. As a time-saver, it's a gift. As the truth, it's a liability. So, before you forward anything the AI wrote, verify three things. One, is it sourced? Every claim should point to something real on the board, a closed ticket, an actual date, not smooth prose filling a gap. Two, is it current? A status is only as true as the board behind it and the AI will happily summarize a board nobody's touched in 2 weeks.

Three, is it complete? The most important line is usually the one that isn't logged anywhere, the blocker in a Slack thread, the risk from a call. Three checks pass, you sign it. Any check fails, you fix it first. Let's run a real one. Here's the end of a sprint. Watch me ask the AI to draft the status, but only to draft it. Pull what's on the board into a clean update. Don't decide if we're on track, and don't invent anything. And there's the draft date of 10 tickets closed, two in progress, written up cleanly. Now verify it. Sourced? Yes, every line matches a real ticket. Current? Yes, I updated the board this morning. Complete? Almost, and this is the part only I can do. The export feature depends on a vendor API that's been flaky all week, and that's on no card anywhere. I add one line flagging it as a

Risk. Now it's sourced, current, and complete. I sign it, and I send it. Now the trap, because a clean green report is so easy to forward. Same button, different sprint. The AI writes, "Project on track, all green, looks perfect." Verify it. Sourced? No, on track is the AI's guess from counting tickets. The board never says it. Current? No, nobody's touched this board in 10 days, because the team quietly moved to a hot fix. So the AI is summarizing fossil. Complete? No, the real blocker, a key engineer out sick, is nowhere on it. Three fails. Forward that, and you've told your client a confident lie. The test is what catches it before you do. So here's the rule you can carry to any tool. Let the AI draft. It reads the board and writes the prose faster than you ever will, but you sign it. And your signature means you checked three

Things, that every claim is sourced, that the board is current, and that you added what the AI couldn't see. The AI writes the paragraph. You write the one line that actually matters. Your turn. Next time your tool offers to write your status, don't just copy it out. Run this prompt. Have the AI draft the update, but make it tag every single claim with the exact ticket or data it came from, and separately list everything it could not verify from the board alone. Read that second list closely. That's your verification checklist, handed to you. A good answer gives you a source draft plus an honest here's what I couldn't see. A bad one gives a confident paragraph with no sources and now you know that's the one never to forward. AI status reports trust but verify. With Sanjana Rao for humanitarian's AI, let the AI draft it, you sign it. Thanks for watching.