Day in the Life of an AI-Powered VA: My Real Week

This week, from Thursday to Saturday, I made two YouTube videos, two Shorts, and a 15-post blog plan, and Claude fact-checked 295 claims along the way. And I didn't film a single frame.

This is what a day in the life of an AI virtual assistant actually looks like for me. It's not a morning routine with a perfect latte. It's the real list of what I did, what Claude did, and what went wrong.

A Day in the Life of an AI Virtual Assistant: My Setup

Everything ran through Claude Code inside the Claude desktop app. Claude Code is the version of Claude that can use tools on your computer, like opening files, searching the web and running programs, instead of only chatting.

The videos are for my small test channel, The 1% Mindset. I pick the topics, approve every script, and say no when something looks off. Claude does the heavy lifting in between.

First: Research Before Anything Else

Both videos started with research, and this is the part I'd show any VA first.

For each topic, Claude sent several research helpers out at the same time, each looking at a different angle: the company's own pages, posts on X, and other YouTube videos. Then a separate set of checkers opened every source again and marked each claim as verified, partly right, unverified, or wrong.

The first topic was MiniMax M3.1 Flash, a new AI model. That ledger, which is just a list of every claim with its verdict, ended up with 194 claims. 155 checked out, 25 were partly right, 10 couldn't be confirmed, and 4 were wrong.

The second was Kimi K3.1, from the Chinese company Moonshot. That one had 101 claims: 76 verified, 19 partly right, 5 unconfirmed and 1 wrong.

That's 295 claims, and only the checked ones were allowed into a script.

Building the Videos

Once the facts were in, Claude wrote each script and I read and approved it. The voice-over comes from Kokoro, a free voice model that runs on my laptop. Most of the visuals are real screenshots of the sources, with red circles and underlines added by Claude.

The MiniMax video came out at 4 minutes 12 seconds. Its topic: a new AI model that guesses an answer 93% of the time when it doesn't know.

The Kimi video is 3 minutes 55 seconds, about what different AI models build when you ask them to build themselves in Minecraft.

Each long video took most of a day, with me reviewing at every step. On top of that came two Shorts, at 27.3 and 25.6 seconds. Claude also made the thumbnails with Nano Banana 2, Google's image model, and wrote the YouTube titles, descriptions and tags after checking what people actually type into YouTube search.

The Plot Twist: News Broke Mid-Project

Halfway through the Kimi project, Moonshot released Kimi K3.1 Pro (October 23, 17:15 UTC). Our script was suddenly out of date.

So we stopped and checked Moonshot's own launch post and its blog before changing a word. The script was rewritten the same day, using only what Moonshot itself had said. If you've ever had a client change the brief mid-project, you know that feeling.

The Part That Caught Real Mistakes

After each render, Claude pulled a frame from every 2 seconds of the video and checked them all, like a very patient editor.

It found a red circle sitting on the wrong bar of a chart, and an empty 8-second stretch where nothing was on screen. On the Kimi video, it spotted two screenshots that showed numbers we hadn't verified, and those were taken out.

None of these were big disasters. But each one is exactly the kind of thing a client would notice and I'd be embarrassed by.

Saturday: Planning This Blog

The last job was this blog. Claude ran the same research-then-verify process on what VAs search for: 61 claims checked (46 verified, 13 partly right, 2 wrong), 104 ideas, and a 15-post plan for this week.

The first post went live on Saturday: Can Claude Make YouTube Videos? I Made a 4-Minute One.

What This Means for VAs

Notice what I spent my time on. Not typing scripts or dragging clips around. I spent it deciding: which topic, which facts were good enough, which frame looked wrong, what to do when the news changed mid-project.

That's the job an AI-powered VA does. The tools handle volume, and you handle judgment. Clients can buy volume anywhere. Judgment is the part they keep you for.

If you want to start, copy just one habit from my week: never let a fact reach a client until something, or someone, has opened the source again.

Want the behind-the-scenes clips from weeks like this?

Claude Deep Research for VAs: How I Fact-Check Every Claim

Claude's deep research feature can hand you a sourced report on almost any topic. Checking it is still your job.

The feature is officially just called "Research", and it's on every paid plan. Turn it on, ask one question, and Claude runs search after search, then writes you a report with its sources.

But a report isn't the same as facts you can trust. This week my research setup checked 356 claims for my videos and this blog, and the second check caught real mistakes. Here's how Claude Research works, and the fact-checking method I add on top.

What Claude Research Is (in Plain Words)

Normal web search in Claude looks something up once or twice. Research goes further. Anthropic's help page says Claude "runs multiple searches that build on each other and decides what to look into next based on what it finds."

Anthropic's guide on choosing between features says Research is best for questions that need five or more lookups over one to three minutes, like comparing a client's competitors, and that it pulls the results into a longer report with citations.

How to Use Claude Deep Research

Who gets it: Research is available on paid plans only (Pro, Max, Team and Enterprise), on the web, the desktop app and the phone app.

How to turn it on: web search has to be on first. Then click the "+" button at the bottom left of the chat and choose "Research". A blue marker shows it's on.

What it can read: the web, plus any connectors you've switched on for that chat. A connector is a link between Claude and another app, like Gmail, Google Calendar or Google Docs. On the phone app, Research only searches the web and quietly skips your connectors, so start important research on a computer.

The catch: it uses up your plan's limits faster than a normal chat, because it reads so many sources.

How I Research a Topic (My Real Method)

Here's where I'm honest about my own setup. This week I didn't use the Research button. I used Claude Code, the version of Claude that can run tools on my computer, and I built the research and checking into separate steps. But the method works with the Research button too.

Step 1: Several helpers, different angles. Claude sent out research helpers at the same time, each with one job: the company's own pages, posts on X, other YouTube videos.

Step 2: Separate checkers. A different set of checkers opened every source again, without trusting the first set's notes.

Step 3: A ledger. Every claim went into one list with a verdict: ✅ verified, 🟡 partly right (with the corrected wording), ❓ couldn't confirm, ❌ wrong.

Step 4: Only checked facts move on. Nothing reaches a script or a post unless it's ✅, or 🟡 with the corrected wording.

The numbers from this week: 194 claims for my MiniMax video, 101 for my Kimi video, and 61 for this blog's plan.

What the Fact-Check Actually Caught

Three examples, so you can see why the second step matters.

The first research pass wrote that freelancers' earnings from AI skills grew 109%. Upwork's own release says something different: skills that mention AI grew 109% in demand, not earnings.

A careers site said the US government expects admin assistant jobs to drop 10%. The US Bureau of Labor Statistics projection for 2024 to 2034 was actually 0%, "little or no change". Its newer projection, for 2025 to 2035, is a 2% decline.

And the first pass decided that Space Bunny, a mystery AI model later revealed as MiniMax M3.1 Flash, was "not #1" on OpenRouter, a site that ranks AI models by use. Half right. It was #2 over the month, but #1 on the daily chart for 10 straight days.

Two of those came from a first pass reading a real source too quickly. One came from a website misquoting a government page. Every one of them looked believable, and that's exactly what a client report can't afford.

Do This With the Research Button

You don't need my setup to copy the method. Run Research on your question. Then open a new chat and paste in the report with this prompt:

"For each factual claim in this report, open the source and quote its exact wording. Mark each claim ✅ verified, 🟡 partly right (give the corrected wording), ❓ couldn't confirm, or ❌ wrong."

Then only use the ✅ claims, and the 🟡 ones in their corrected form.

What This Means for VAs

Research is one of the most common things clients hand to a VA: a competitor list, a supplier check, a "find out what's new in our industry". Claude Research does the searching for you.

What it can't do alone is take responsibility for the answer. A ledger with ticks and crosses is something you can show a client. It says "I checked this", and that's worth more than any fast summary.

Try it on your next research task.

Can Claude Make YouTube Videos? I Made a 4-Minute One

Yes, it can. This week I made a full 4-minute YouTube video with Claude: the research, the script, the voice, every scene, and the thumbnail. But it didn't happen the way most people imagine, and the part that surprised me most wasn't the video at all.

Here's exactly what Claude did, what I did, and what I'd tell any VA who wants to try it.

First, What "Making a Video" Meant Here

Claude didn't press a magic "make video" button. There's no camera and no AI-generated footage in this video.

I used Claude Code, the version of Claude that can run tools on your computer, not just chat. It wrote the video as a set of animated web pages, then recorded them frame by frame into an MP4. Think of it like building slides that move, then filming them.

The video is for my small test channel, The 1% Mindset. The topic: a new AI model that guesses an answer 93% of the time when it doesn't know.

How It Went, Step by Step

Research. I asked Claude to dig into the topic. It sent several research helpers out at once, then a separate set of fact-checkers who opened every source again. They checked 194 claims. Four were wrong, and none of those made it into the script.

Script. Claude wrote a 4-minute script in plain English, then checked its own opening against a retention checklist and rewrote the first lines. I read it and signed it off before anything else happened. That's the same job I'd do for a human writer.

Voice. The voice-over comes from Kokoro, a free voice model that runs on my laptop. Claude split the script into sections, voiced each one, then transcribed the audio back to check that every word came out right.

Visuals. Every chart and post you see is a real screenshot of the source. Claude added the red hand-drawn circles and underlines, timed to the voice. The only made-up art is the cartoon test sheets and bunny from the thumbnail, plus a few simple bar graphics.

Thumbnail. Claude wrote the prompt for Nano Banana 2 (Google's image model), then added the "100% SURE" text on top.

The final file: 4 minutes 12 seconds, full HD.

What Surprised Me

The checking. After the render, Claude pulled a frame from every 2 seconds of the video and looked at all of them, like a very patient editor.

It caught real mistakes. One red circle sat on the wrong bar of a chart, because the website had shifted its layout after the screenshot. Another section sat empty for 8 seconds. Both got fixed before I ever saw the video.

On a second video, it caught something worse: a screenshot that showed numbers we hadn't been able to verify. It cropped them out. As someone whose job is getting details right for a client, that's the part that won me over.

Is It Free?

Not completely. Claude Code needs a paid Claude plan. The voice model is free, and the thumbnail image costs a little through Google.

Time is the bigger cost. This took most of a day, with me reviewing at every step. It's still a fraction of what one person would need to research, write, voice, animate, and edit the same video by hand.

What This Means for VAs

If you manage YouTube for a client, this changes what you can offer. You don't need to be a video editor to turn a script and real sources into a clean explainer, a Short, or a slide-style video.

But notice who made the decisions. Claude did the heavy lifting, while I picked the topic, approved the script, chose the look, and said no when something felt off. That's still VA work. It's just a bigger version of it.

Start small. Ask Claude to turn one of your client's blog posts into a 30-second Short script, and see how much of the rest you'd want to hand over.

If you try it, tell me how it goes on the comments.