Your AI Writes Like a Robot. Here's How to Fix That.
We run three brands. Every piece of AI-generated content sounded different — and obviously robotic. Here's how we built a Writing Agent that actually passes for human.
"In today's digital landscape, it's important to note that leveraging AI can revolutionize your content strategy."
If you just cringed, you get it. That sentence? Pure AI. Corporate buzzword soup. Painfully obvious. And getting flagged by every AI detector on the internet.
We run three different brands: a pool care site, a recipe app, and B2B SaaS for hiring automation. Every time we needed content — blog posts, tweets, email drafts — the AI output was inconsistent and obviously robotic.
Pool care blogs went through a humanizer before publishing. Tweets didn't. Recipe posts sounded nothing like the brand. Client emails from HireAnvil read like ChatGPT with a business degree.
That's when we built a dedicated Writing Agent.
The Inconsistency Problem
Here's what was happening. Our main AI agent could write. Sure. But it had no memory of brand voice, no rules about what phrases to avoid, and no consistency across formats.
Pool Calculator blog post
We'd run it through a humanizer script before publishing. The output scored 75-85 on human detection. Pretty good.
Pool Calculator tweet
Written by the same agent. No humanizer. Pure AI voice. "Dive into pool chemistry with our comprehensive guide!" Dead giveaway.
Recipe Genius blog
Different context window. Different prompt. Completely different tone. Sometimes chatty and fun. Sometimes like a chemistry textbook. No brand consistency.
HireAnvil client email
The worst. "We're reaching out to circle back on leveraging our revolutionary platform to streamline your hiring paradigm." Nobody talks like that. And clients noticed.
We needed one system that handled all writing for all brands. With consistent rules. Consistent voice. And consistent anti-AI defenses.
Why AI Detection Matters
Some people think AI detection is overblown. We don't.
Google is getting better at spotting AI content
They won't tell you the exact signals, but patterns are clear. Generic phrasing ranks worse. Overly polished content without personality gets buried. And if your blog sounds identical to 10,000 other AI-generated posts about the same topic? You're not ranking.
Social platforms flag robotic content
Twitter's algorithm deprioritizes obvious AI posts. LinkedIn too. They want authentic engagement, not bot spam. If your AI-written posts get less reach, that's why.
Human readers can tell
Even if detection algorithms miss it, your audience won't. They've read enough AI slop by now. They recognize the patterns. The journey-embarking. The power-unlocking. The constant exploration invitations. It screams AI.
And the worst part? If your audience thinks it's AI, they trust it less. Doesn't matter if the information is accurate. The voice kills credibility.
The Banned Phrases List
We started by documenting every phrase that screamed "AI wrote this." Turns out there are about 60 of them. Here are the worst offenders:
Corporate buzzword garbage
Words like synergy and paradigm shifts. That thing where you pick fancier words because they sound more impressive. Game-changers and revolutionary platforms. All of it goes.
Generic AI phrasing
The "in today's modern digital world" opener. Starting by addressing all experience levels at once. Any mention of navigating landscapes. Invitations to begin journeys through comprehensive guides. You know the drill.
Overused transitions
All those phrases about exploring deeper. The constant noting of importance. Unnecessary preambles before getting to the point. Numbered lists that sound like a corporate presentation.
We built these into the Writing Agent as hard rules. If a draft includes any of these patterns, it gets rewritten. No exceptions.
Brand Voice Profiles
Each brand has a distinct voice. Pool Calculator is educational but approachable — like a helpful neighbor who knows chemistry. Recipe Genius is casual and enthusiastic — think friend texting you dinner ideas. HireAnvil is professional but direct — no corporate fluff.
The Writing Agent stores voice profiles for each brand:
Pool Calculator
- • Use specific measurements (ppm, gallons, pH levels)
- • Explain chemistry in simple terms
- • Include real scenarios ("your pool turned green overnight")
- • Conversational but authoritative
Recipe Genius
- • Sensory descriptions (smell, taste, texture)
- • Casual language ("trust me on this")
- • Cooking tips from experience
- • Enthusiastic without being over the top
HireAnvil
- • Lead with business impact (time saved, cost reduction)
- • Include metrics and case studies
- • Professional tone, still human
- • No buzzwords. Ever.
Now when any agent in our ecosystem needs content, it calls the Writing Agent and specifies the brand. The output matches the voice automatically.
The Humanizer Score
This is the important part. We built a scoring system that evaluates content on a 0-100 scale:
What it checks:
- Sentence length variance — AI loves uniform sentences. Humans mix it up. Short punchy ones. Longer explanatory sentences that build on an idea. Then back to short.
- Banned phrase detection — Instant penalty for any of the 60+ AI tells
- Passive voice ratio — Too much passive voice sounds bureaucratic
- Hedging language — Words like "may" and "might" piled on top of each other — AI hedges everything
- List-of-three pattern — AI constantly writes "X, Y, and Z" structures
- Transition word overuse — "Moreover," "Furthermore," "Additionally" at the start of sentences
The humanizer runs automatically before any content gets published. If it scores below 80, the Writing Agent rewrites it.
💡 Real example:
First draft of a Pool Calculator blog scored 62. Too many unnecessary transitions. Too much passive voice. Uniform sentence length. After rewrite: 84. Passed.
What Actually Works
After running a few hundred pieces of content through this system, here's what makes AI writing sound human:
1. Vary sentence length
Mix it up. Short grabs attention. Longer sentences provide detail and context while maintaining flow.
2. Start sentences with "And" or "But"
Real people do this. AI is trained not to. And it creates natural rhythm.
3. Use contractions
Don't write like a robot. Won't make you sound human. Can't emphasize this enough.
4. Include specific numbers
"Save time on meal prep" sounds AI. "Chop vegetables Sunday and save 15 minutes every weekday dinner" sounds like experience.
5. Tell mini-stories
Facts fade. Stories stick. "I once ignored my pool for two weeks in July. Came back to swamp water and a $200 chemical bill" beats any generic advice about regular maintenance.
6. Have opinions
AI tries to be neutral and balanced. Humans don't. "Option A works better for most people. Option B is for control freaks" sounds way more human than "both approaches have merits depending on individual circumstances."
7. Use analogies
"Think of pH like the thermostat for your water — too high or low and nobody's comfortable." Real humans explain through comparison.
8. Occasional sentence fragments
For emphasis. Works great. Readers get it.
We built all of these patterns into the Writing Agent's rewrite logic.
The Results
Since deploying the Writing Agent:
- Blog posts score 82-89 on the humanizer. Consistently. Across all three brands. Before this, we were getting 55-75 with huge variance.
- Social media engagement is up 40%. People respond to posts that sound like a human wrote them. Shocking, right?
- Client feedback changed. HireAnvil emails used to get ignored. Now they get responses. One client literally said "I appreciate that your emails don't sound like every other SaaS spam."
- SEO improved. Pool Calculator blog posts started ranking better. We think Google's algorithm can tell the difference, even if they won't admit it.
- Consistency across channels. Pool care tweets sound like they're from the same brand as pool care blogs. Recipe Instagram captions match recipe blog tone. Everything feels cohesive.
And the best part? We're eating our own dog food. This blog post was written by the Writing Agent. So was the other one we published today. Both scored above 85.
How to Build This
You don't need our exact system. But here's the framework:
- Document your banned phrases. Read 10 pieces of AI-generated content. Highlight every phrase that screams "robot." Build a list. Enforce it.
- Create brand voice profiles. Write down how each brand should sound. Examples help. "Like a helpful neighbor" is vague. "Like a helpful neighbor who knows pool chemistry and explains it over the fence while drinking beer" is specific.
- Build a humanizer. Start simple. Check for sentence length variance, banned phrases, and passive voice. That alone gets you 80% of the way there.
- Make it a dedicated agent. Don't bolt content generation onto your main agent. Separate concerns. One agent handles writing. Other agents call it when they need content.
- Score everything. If you don't measure, you can't improve. Run every piece of content through the humanizer before publishing. Track the scores. Learn what works.
Our Writing Agent is about 400 lines of Python. The humanizer is another 200. The brand voice profiles are markdown files. Nothing fancy. Just consistent rules, enforced automatically.
Bottom Line
AI content detection is real. Inconsistent brand voice hurts trust. And if your AI writes like every other AI on the internet, your content gets buried.
The solution isn't to stop using AI. It's to build systems that make AI output sound human. Banned phrase lists. Brand voice profiles. Humanizer scoring. Sentence variance. Specific details. Real stories.
And eat your own dog food. If you're building content tools, use them. This blog was written by our Writing Agent. So were our tweets about it. So will be the client emails we send next week.
If it works for us, it'll work for you.
Build Your Own Agent Infrastructure
The Writing Agent is part of AutoClaw Agents — our framework for building specialized AI ecosystems. We also have agents for credential management, trading signals, social media automation, and more.
Build Your Agent