Why AI Still Needs a Human Virtual Assistant (Like Me)
Six months ago, I wrote a post asking: Will AI replace Virtual Assistants? (Spoiler: no, but it will make us better.)
Fast forward to August, and the landscape's shifted again, not in a big headline-grabbing way, but in the quiet details. Tools have levelled up, yes. But so have the headaches. And honestly? Some of the "improvements" have made things trickier for virtual assistants, not easier.
Here's what I've seen as a working VA, running a tech-savvy, human-first business, and how I'm using (and checking) AI in the real world.
AI Is Smarter, But Also More Wrong, More Often
Let’s start with the weird bit: some of the newer tools are less reliable than the old ones. Models are hallucinating more. They’re confident, polished… and sometimes completely wrong.
I've been testing loads of different tools, but let me focus on the main text and data ones I use daily:
Claude (great for long content, but increasingly flaky with facts - I've caught it making up statistics that sound perfectly reasonable)
ChatGPT (quicker now with their new agent features, but more "slippery" with numbers, especially anything involving percentages or data analysis)
Omni in Airtable (brilliant at spotting patterns and themes, but shaky when it comes to actual calculations)
Perplexity (my go-to for research - I love that it shows sources, but it can still cherry-pick data or miss context that changes everything)
The stats back this up, too. OpenAI's own internal testing shows their GPT o3 model hallucinates 33% of the time on PersonQA benchmarks, while o4-mini hits 48% on PersonQA and 79% on SimpleQA tests. That's significantly worse than earlier versions on the same benchmarks. So much for "continuous improvement."
I've watched AI generate entire project summaries; beautifully written, completely made-up. Client reports that look professional but contain invented data points. Email sequences that sound like they came from the client's brand... except they suggest strategies the client would never approve.
The tools are powerful. But they don't understand what matters. You still need a human to say:
"Wait... is that even true?"
"Would the client say it like that?"
"Does this feel right for their brand?"
"Hang on, let me double-check those numbers..."
And that human is me. Or you.
It’s Not “Human VA vs AI VA” Anymore - It’s “Human VA with AI (and a Big Red Pen)”
Let's drop the fear. AI is here. It's useful. It's often brilliant. But it still needs hand-holding. It’s fast, creative, occasionally genius, but prone to making stuff up when it doesn't know the answer.
What this means for us human VAs:
We're not just doing the tasks anymore. We're curating, fixing, steering, fact-checking.
Clients don't want robots. They want trust. They want nuance. They want someone who gets their business.
AI gives you a draft. You give it direction, personality, accuracy.
The market's responding to this shift, too. There's growing demand for VAs who've figured out how to make AI work for them (and their clients). The ones commanding premium rates? They're the ones who know how to manage these tools and spot their limitations.
What’s Tripping People Up Right Now
Here’s what I’m seeing across the VA world (and yes, I’ve made some of these mistakes myself):
Over-trusting AI stats
Tools fudge data more often than you'd think, especially on large spreadsheets or when logic gets layered. I had Claude confidently tell me a client's email open rates had "improved by 23%" when they'd actually dropped. Always verify anything involving numbers.
Sounding robotic
That "AI shine" is real. Perfect grammar. Mid-Atlantic tone. No contractions. No life. Sentences that start with "It's important to note..." or "While it's true that..." You've got to roughen it up. Add some contractions. Throw in a sentence fragment. Let it breathe.
Forgetting important details despite training
I use ChatGPT's memory feature to train it on client preferences and project details, but it still forgets crucial things. I'll be working on a campaign and have to keep reminding it about the client's tone of voice, or their main target audience, or that they hate certain buzzwords. The memory helps, but it's not reliable enough for complex, ongoing work.
The context problem
Both ChatGPT and Claude have conversation limits that can interrupt complex projects. Nothing worse than being halfway through building a comprehensive strategy document only to hit a session cap and lose all your context.
But There’s Good News, Too
AI is making some things much easier for VAs. Here’s what I’m happy to offload:
Transcribing videos and meetings
Descript, Filmora and Otter save me hours. What used to be a tedious, mind-numbing task now happens automatically. Though I still scan the transcripts for obvious errors as AI sometimes hilariously misunderstands technical terms.
Generating first drafts
Blog posts, social media copy, tricky emails, web copy; AI helps get the bones down faster. I give it detailed briefs (the more specific, the better), then spend my time refining rather than staring at a blank page.
Rewriting something 10 different ways in 10 seconds
Ideal for tailoring social media captions to each platform without going cross-eyed. LinkedIn professional, Instagram casual, Twitter punchy—all from one base version.
Powering up website chatbots
I use them to answer FAQs, book consults, or triage requests. They work around the clock, even when I don't. Though I review the conversations weekly to catch any weird responses.
Automation between tools
This is where AI really shines. I use AI + Make to pull invoice summaries from Xero and auto-schedule them to clients. I've set up automations where MailerLite tags trigger file drops into SharePoint, and weekly project summaries generate directly from ClickUp data.
The time savings here are real; what used to take me 2-3 hours of manual work each week now happens automatically.
Client brief builders & task triage tools
I've prototyped internal tools that help clients figure out what to delegate, automate, or ignore entirely - all AI-assisted. Clients love having a structured way to think through their needs.
The Game-Changing Tools (When Used Right)
Some platforms are genuinely changing how I work as a virtual assistant:
Zapier's new agents: You literally describe what you want in plain English, and it builds the automation. It's like having a technical assistant who speaks human.
Airtable's Omni: It's brilliant at database management and spotting patterns I'd miss. Like having a tiny assistant running your client data, as long as you don't ask it to do maths.
ChatGPT's file processing: Honestly revolutionary for document analysis and data extraction. Upload a PDF, get structured insights. Just... verify everything it tells you about the numbers.
Claude's extended thinking: When you need deep analysis, its chain-of-thought process is impressive. You can actually see it working through complex problems step by step.
I use these tools every day. But I don't trust any of them blindly.
What I Never Delegate to AI as a Virtual Assistant
No matter how clever the tools get, some parts of this work still need care, nuance, and a real person paying attention:
Anything emotionally sensitive (tricky emails, difficult client conversations, personal matters)
Anything that needs genuine instinct (knowing when a client is stressed, reading between the lines)
Final client deliverables (nothing goes to clients without my review first)
Strategic decisions that affect the client's business
Checking everything (twice, sometimes three times)
The moment you hand over your judgment entirely is the moment you stop being a VA and become just someone who feeds prompts into AI. That's not the same thing.
The Professional Reality Check
Many clients are still hesitant about AI-assisted services due to accuracy concerns. They've heard the horror stories - lawyers getting in trouble for ChatGPT's fake case citations, financial analysts discovering basic calculation errors in AI-generated reports.
But here's the flip side: companies are reporting significantly faster query handling and major cost savings when AI is properly managed by skilled humans. Companies like BELAY are seeing real success with their "AI-empowered professionals" approach.
The key word there? Professionals. It's not about the AI. It's about the human using it wisely.
If you're a VA:
Learn the tools. But don't worship them. They're powerful assistants, not replacements for your brain.
Use AI to stretch your time, not replace your judgment. Let it handle the grunt work so you can focus on strategy and relationships.
Keep your human voice. That conversational, understanding, slightly imperfect tone? That's your superpower.
Develop quality control systems. Multiple checks, different tools for verification, human review cycles.
Be transparent with clients about what you use AI for and how you verify its work.
If you're hiring a VA:
Ask how they use AI. If they say "not at all," they're behind the times.
But also ask how they double-check it. If they say "I trust it completely," run.
Look for VAs who can explain their process - both the AI parts and the human oversight parts.
Value the combination: technical efficiency + human judgment + relationship skills.
Looking Forward: The Collaborative Future
Industry experts have moved beyond the "replacement vs. augmentation" debate. We're heading toward what Harvard Business Review calls "collaborative intelligence" - AI as a digital teammate rather than a tool.
Research shows that people with strong interpersonal skills get 75% more value from AI agents. The VAs who'll thrive are those who can combine AI efficiency with genuine relationship-building, cultural understanding, and strategic thinking.
The VA market's growing rapidly, fueled by how we're learning to work with AI rather than being replaced by it.
Final Thought
AI isn't the enemy. But it's not your boss, either. It's a brilliant, slightly unhinged assistant that needs constant supervision and occasional reality checks.
In 2025, the real magic isn't in the tool. It's in the person using it; knowing when to trust it, when to question it, and when to put it aside and rely on good old-fashioned human judgment.
The future belongs to virtual assistants who can dance with AI without letting it lead.
Want to see how I actually use these tools day-to-day? I'm putting together a behind-the-scenes look at my AI-enhanced workflows - the good, the bad, and the occasionally hilarious. Drop me a line if you're interested.
