The Day My Robot Wrote Better Than Me (And Why I Didn't Quit)
The email pinged in at 7:42 AM. Subject line: “Can you review this before we send it to the client?”
I opened the draft. It was a two-page proposal for a SaaS onboarding campaign. The tone was polished. The grammar was flawless. The structure was so clean it could have been printed on a surgical tray. And I felt my stomach drop into my shoes.
Because I didn’t write it. My intern did—with ChatGPT.
She’d fed it the client brief, a few bullet points about their product, and a prompt about “professional but friendly” tone. The output was… fine. More than fine. It was good. It was arguably better than the first draft I would have written, because it didn’t carry my hesitation, my weird comma habits, or my tendency to over-explain.
I sat there with my coffee going cold, staring at the screen. The classic existential dread washed over me. Not the “robots are taking our jobs” panic—I’d made peace with that years ago. No, this was worse. The dread was: What if my voice doesn’t matter anymore?
Here’s the thing about the AI content wave. Everyone is panicking about quantity—how much content we can produce, how fast, how cheap. But the real crisis is quality in the most human sense of the word: distinctiveness. If every SaaS company uses the same LLM with the same prompts, we’re all going to sound like the same slightly-hallucinating, vaguely-optimistic, bullet-point-loving robot.
I’ve spent the last six months working through this problem. Not just for my own writing, but for the startups I advise. And I’m here to tell you: the fix isn’t to use less AI. The fix is to use it more surgically—and to re-assert your humanity on purpose.
Let me show you how.
The “Smoothie” Problem: Why Blending Everything Makes You Tasteless
There’s a common mistake I see in every early-stage SaaS team. They treat AI like a blender. They throw in their product specs, their competitor analysis, their founder’s LinkedIn bio, and three blog posts from industry leaders. They hit “puree” and expect a smoothie.
What they get is a gray, watery sludge.
Why? Because a smoothie homogenizes everything. When you ask ChatGPT to “write a blog post about our CRM integration,” you’re not asking it to think. You’re asking it to average. It averages every CRM integration article ever written. It averages the tone of every SaaS blog. It averages the structure of every “5 Ways to Improve” listicle.
The result is statistically perfect and personally meaningless.
I saw this firsthand with a founder named Priya. She runs a fintech SaaS for freelancers—think invoicing, tax tracking, payment reminders. She came to me frustrated because her blog traffic was flatlining. She was publishing three times a week. She was using AI for everything: outlines, drafts, meta descriptions, even email newsletters.
“I’m doing everything right,” she said. “But people aren’t engaging. The comments are empty. The shares are zero.”
I read one of her posts. It was about “Simplifying Quarterly Tax Payments.” It was… fine. It had statistics. It had a table comparing traditional accounting vs. her software. It had a conclusion that said “In conclusion, simplifying your quarterly tax payments is essential for freelancers.”
I asked her one question: “When did you last miss a tax payment?”
She laughed. “Literally last April. I forgot to set aside the money, and I had to scramble to sell some stock to cover it. My accountant yelled at me.”
“Why isn’t that in the blog post?”
Silence.
“Because ChatGPT didn’t know about it,” she said.
That’s the whole problem in one sentence. Your AI tool doesn’t know about your failures, your embarrassing moments, your weird quirks, or the specific way your heart races when you click “submit” on a payment you can’t afford. And if you don’t inject those into your content, you’re just another cog in the content machine.
The “Director” Method: Treat AI Like Your Best Employee, Not Your Replacement
Let me reframe the mental model. Stop thinking of AI as a writer. Think of it as a brilliant, hyper-literate, slightly sycophantic research assistant who has read everything but experienced nothing.
You are the director. The AI is the actor.
When you direct a scene, you don’t just say “say the lines.” You give blocking. You give subtext. You give backstory. You say, “Remember, you’re exhausted here, and you’re lying to your partner, and you just ate a bad burrito.” That’s when the magic happens.
In practice, this means you must provide the raw material of your voice. Not just keywords. Not just topics. But texture.
Here’s my exact workflow for a client project—a cybersecurity startup that was struggling to sound less like a firewall manual and more like a trusted ally.
Step 1: The Voice Vault (Your Secret Weapon)
Before I let the AI write a single word, I make it read me. I built a “Voice Vault” for each client. It’s a Google Doc with:
- Three “golden” pieces of content that perfectly capture the founder’s speaking style. Not their best-performing content—their most authentic content. The one where they swear a little, or tell a story about their dog barking during a board meeting.
- A list of “never say” words. For this cybersecurity client, it was “leverage,” “utilize,” and “synergy.” We replaced them with “use,” “need,” and “team up.”
- A set of “signature phrases.” The founder, a former white-hat hacker, always said “Assume the breach.” We made that the anchor phrase for every blog post.
Then, in the prompt, I don’t say “write in a professional tone.” I say: “You are writing in the style of a former hacker who is now a CEO. You are direct, slightly paranoid, but empathetic to small business owners. You use the phrase ‘Assume the breach’ at least twice. You never use the word ‘leverage.’ You prefer short sentences. Start with a story about a time a client almost lost their data.”
That prompt is 100 words. It produces a 1,500-word post that sounds like the founder. Not like every other cybersecurity blog.
Step 2: The “Dirty Draft” Technique
Most people use AI for a clean first draft. I use it for a dirty draft.
I write the first 200 words myself. I get the blood flowing. I establish the scene, the voice, the point of view. Then I let the AI continue, but I give it a very specific instruction: “Continue this draft in the same tone. Do not add any new facts. Do not add any statistics unless I provided them. You may only use the anecdotes I have shared in the conversation. If you need to fill space, ask a rhetorical question instead of inventing a case study.”
Why? Because AI hallucinates details. It will invent a client named “John from Ohio” who saved 40% on his cloud bill. That’s dangerous. That’s a liability. By restricting the AI to my facts and my anecdotes, I keep the content true. And truth is the ultimate human differentiator.
The “Rough Edge” Principle: Why You Should Ruin the AI’s Perfect Prose
Here’s a confession. When I review AI-generated copy, I’m not looking for errors. I’m looking for perfection to ruin.
AI writes in a perfect, balanced, symmetrical rhythm. Every paragraph is the same length. Every sentence has a subject-verb-object. Every transition is buttery smooth. That’s not how humans write. Humans get tangled. Humans have run-on sentences. Humans interrupt themselves.
I call this the Rough Edge Principle. You need to intentionally introduce friction into the AI’s output.
Concretely, I do three things:
- Delete the first sentence. AI loves to start with a sweeping generality like “In today’s fast-paced digital landscape…” That sentence is garbage. Delete it. Start with the second sentence, which is usually more specific.
- Cut one adjective per paragraph. AI is adjective-happy. “Seamless integration,” “robust solution,” “intuitive interface.” Cut them. The text gets leaner and more confident.
- Add a parenthetical aside. If the AI wrote a serious paragraph, add a joke in parentheses. If it wrote a list, interrupt it with a personal anecdote. For example: “There are three main reasons to migrate to the cloud. (The fourth reason is that your server room smells like burnt toast, and you’re tired of it.)”
That last one is gold. It breaks the spell. It reminds the reader that a human was here.
I remember editing a draft for a project management SaaS. The AI had written a bullet list of features. It was accurate. It was boring. I added a single line after the list:
“We also have a dark mode. Because we know you’re hiding in a closet during a family gathering, pretending to work, just to get five minutes of peace.”
That post got the most comments in their company history. Why? Because everyone has hidden in a closet during a family gathering.
The “Voice Check” Ritual (Read It Out Loud)
Here’s the most low-tech, high-impact trick I know. After you’ve edited the AI draft, read it out loud. Actually speak the words.
Your ears catch what your eyes miss. When you read silently, your brain autocorrects weird phrasing. When you read aloud, you stumble. You hear the clunky rhythm. You notice when a sentence is too long to breathe.
If you stumble, rewrite that part in your own words. Don’t edit the AI’s text. Rewrite it. Type it from scratch. This is the key difference between editing and owning.
Editing is passive. Owning is active.
When you rewrite the sentence, you’re pulling it into your body. You’re making it yours. This is what separates a founder who uses AI as a crutch from a founder who uses AI as a lever.
I do this with every piece of content I publish on my own site, https://www.harishapc.com, where I write about AI strategy and content operations. I literally read the final draft out loud to my dog. He doesn’t care. But the process forces me to slow down and hear the voice. If my dog doesn’t perk up at a sentence, it’s probably too corporate.
The “Boring Business” Exception: When You Actually Should Sound Like a Robot
Now, I need to be honest. There are times when the human voice is not the goal. There are times when you want the smooth, homogenized, robotic output.
Think of your FAQ page. Think of your API documentation. Think of your legal terms and conditions. These are not places for personality. You don’t want your privacy policy to say, “We collect your data because we’re nosy, but in a cute way.” No.
For these, let the robot be a robot. Use AI to generate clear, concise, unambiguous copy. This is a feature, not a bug. The human voice is precious. Don’t dilute it by trying to make your error messages charming.
But for marketing, thought leadership, sales emails, and onboarding flows—anything that requires connection—you need the human voice.
The distinction is: Transactional vs. Relational.
- Transactional: How do I reset my password? (Robot is fine.)
- Relational: Why should I trust you with my business? (Human required.)
So many SaaS founders make the mistake of applying a “brand voice” to transactional content and a “generic AI voice” to relational content. It’s backwards.
The Startup Founder’s AI Content Workflow (Step-by-Step)
Let’s get practical. Here’s the exact workflow I recommend to the startups I work with. It’s a blend of speed and humanity.
- Brainstorm with AI, not without it. Don’t sit in a blank room. Open ChatGPT and ask for 20 blog post ideas on your topic. Pick the one that makes you feel something—fear, excitement, or anger. If you’re not slightly scared to publish it, it’s too safe.
- Write the “Soul Paragraph.” Before you prompt the AI, write a single paragraph (100-150 words) about why this topic matters to you personally. Include a failure, a doubt, or a specific customer story. This is your anchor.
- Prompt with Context, Not Just Topic. Give the AI your Soul Paragraph, your Voice Vault, and your “never say” list. Explicitly instruct it to extend your paragraph, not rewrite it.
- Generate 3 Versions. Ask for three different angles on the same brief. The first will be safe. The second will be weird. The third will be surprisingly good. Take the best parts of each.
- The Rough Edge Pass. Do the three things I mentioned: delete the first sentence, cut one adjective per paragraph, add one parenthetical aside.
- Read It Out Loud. Do the dog test. If you stumble, rewrite.
- Add the “Human Proof.” Screenshot a real tweet. Reference a real conversation. Link to a real case study. AI can’t do this. It’s your secret sauce.
This process takes me about 45 minutes per post. That’s down from 3 hours. But the output sounds more human than my old drafts, not less. Why? Because the AI handles the heavy lifting of structure and grammar, freeing me to focus entirely on the emotional core.
The “Harish Test” for AI Content
I have a personal benchmark. I call it the “Harish Test” (named after my site, https://www.harishapc.com, because I’m not humble).
The test is simple: If you covered up the author’s name, the logo, and the product, could you still tell who wrote it?
If the answer is no—if it could be any SaaS blog—you’ve failed. You’ve let the AI win.
I read a blog post last week from a Series B startup. It was about “The Future of Remote Work.” It was well-written. It had data from Gartner. It had a nice infographic. But it could have been written by any of 500 companies. There was no soul. There was no point of view. There was no risk.
Contrast that with a post I read from a solo founder who builds a note-taking app. He wrote about how he accidentally deleted his own company’s database during a demo. He described the panic, the sweat, the phone call to his co-founder at 2 AM. He tied it to the importance of backups. It was raw, it was funny, and it was terrifying.
That post got 10,000 views. The Series B post got 200.
The difference wasn’t AI. The difference was vulnerability.
AI can’t be vulnerable. AI can’t say “I messed up.” AI can’t admit that you’re scared. AI can only project confidence. And confidence is boring. Vulnerability is magnetic.
A Final Warning: The “Echo Chamber of One”
There’s one more trap I want to warn you about. It’s the “Echo Chamber of One.”
When you use AI to write in your voice, you’re training the AI to mimic you. That’s good. But it also means you’re slowly removing the influence of other voices. Your writing becomes a closed loop. You stop evolving.
To prevent this, I force myself to do the opposite. Once a month, I ask the AI to write in the voice of someone I disagree with. I ask it to argue against my own product. I ask it to write a blog post that criticizes my SaaS’s pricing.
This sounds counterintuitive, but it keeps my own voice sharp. It reminds me that there are other perspectives. It injects a little bit of chaos into my echo chamber.
Then, I take that opposing argument and address it in my next piece of content. That’s how you build trust. That’s how you show you’re not a bot. You show you’ve heard the other side.
The Final Word: You Are the Algorithm
Look, I’m not going to tell you to stop using AI. That’s Luddite nonsense. AI is the most powerful writing tool since the printing press. It’s a gift. It levels the playing field. It lets a solo founder publish like a content team of ten.
But the gift comes with a caveat: The algorithm can generate text, but it cannot generate perspective.
Perspective is earned. It comes from late nights, failed launches, angry customers, and accidental successes. It comes from the specific, messy, contradictory experiences that make you you.
So, the next time you sit down to write with AI, don’t ask it “What should I say?”
Ask yourself, “What did I learn that the AI couldn’t possibly know?”
That’s your voice. That’s your edge. That’s the thing that no model can replicate.
Use the AI to be faster. Use the AI to be clearer. Use the AI to be more structured.
But never, ever let it be braver than you.
Now, go write something that makes your dog perk up.
Top comments (0)