Why Your AI Sounds Like a Corporate HR Department (And How to Fix It)
Note: This is an excerpt from my new book, THE ART OF WHISPERING TO MACHINES, available now on Amazon.
Why Your AI Sounds Like a Corporate HR Department (And How to Fix It)
Note: This is an excerpt from my new book, THE ART OF WHISPERING TO MACHINES, available now on Amazon.
If you ask ChatGPT to write an email turning down a wedding invitation, it will probably sound like a polite HR manager letting someone go. The message will be overly formal, long-winded, and lack any real feeling.
If you ask it to explain a coding concept, it often comes across like a nervous intern reading straight from a textbook.
This is what I call The Polymath’s Curse.
The AI has read every book, email, and blog post in existence. It has read Shakespeare, and it has read YouTube comments. It has read the US Tax Code, and it has read 4chan.
When you ask a general question, the AI tries to be everything at once. It averages all its training data.
To solve this, we should stop thinking of AI as just a generic smart computer and start using what I call the Spotlight Theory.
The Spotlight Theory
Picture the AI’s mind as a huge, dark theater stage. On this stage, millions of actors stand in for every bit of knowledge the model holds.
There’s a doctor, a lawyer, a surfer, a 19th-century poet, a software architect, and even a toddler.
When you type a simple prompt like “Explain how photosynthesis works,” it’s like turning on the house lights. The whole stage lights up a little. The doctor starts talking, but so do the surfer and the poet. The AI blends their voices, so you get an answer that’s accurate but dull.
Persona prompting means choosing who gets the spotlight.
If you start your prompt with “Act as a cynical biology professor who hates early mornings,” you turn off the house lights and shine a bright spotlight on just one actor.
Everyone else on stage goes quiet. The “Cynical Professor” steps forward, and the response changes right away.
- Generic Prompt: “Explain photosynthesis.”
- AI: “Photosynthesis is the process by which green plants and some other organisms use sunlight to synthesize foods…” (Yawn).
- Persona Prompt: “Act as a cynical biology professor.”
- AI: “Look, plants eat light. That’s it. They take a deep breath of CO2, drink some water, stand in the sun, and vomit out oxygen. Try doing that before your first cup of coffee.”
Real Life Example: The Salary Negotiation
Imagine a friend asks you to helpuse AI to negotiate a salary offer.
First, you paste in the job offer and ask: “Write a counteroffer email asking for 10% more.”
The result sounds nervous and apologetic. It uses phrases like “I was wondering if there might be wiggle room” and “I totally understand if not.” It is the average of every timid employee asking for a raise.
You delete it.
Then you apply the Spotlight. You tell the AI: “You are an FBI Hostage Negotiator. Your goal is to get the best outcome while keeping the other side calm and feeling in control. Rewrite this email.”
The difference is striking. The new email is short, firm, and empathetic. It makes a proposal instead of asking. It justified instead of apologizing. Your friend sends the email and gets the 10%.
The Technical Reality: The “Director’s Note”
Why does this work so well? It’s not just about roleplay—it’s about probability.
Behind the scenes, AI is always asking, “Given the words I just wrote, what’s the most likely word to come next?”
When you leave the prompt open, the “likely” next words are drawn from the entire universe of English text. The probability is flat and wide. When you assign a persona, you are telling the model: “Ignore 99% of your training data. Only look at the cluster of data associated with ‘FBI Negotiation Manuals’ and ‘Psychology Textbooks’.”
The “System Prompt”
In software, there’s a tool for this called a System Prompt.
Think of your conversation with AI as a play.
- The user prompt is like the audience shouting at the stage, “Tell me a joke!”
- The system prompt is the quiet instruction the director gives the actor before the show starts.
The director quietly says, “You are a grumpy 1920s coal miner.” When the curtain rises and the audience shouts, “Tell me a joke!” the actor doesn’t tell a “Knock Knock” joke. Instead, he shares a dark joke about canary birds.
You can do this yourself. Most AI tools now have “Custom Instructions” or “Memory” settings. This is your chance to be the director. If you want concise answers, put it in the System Prompt. If you want code without explanations, put it in the System Prompt. Set the stage before the play begins.
Conclusion: Don’t Hire the Generalist
The biggest mistake people make with AI is treating it like a search engine. A search engine finds information. An AI acts like a speaker.
If you don’t give it a role, it tries to be everyone at once. And that’s boring.
So next time you open a chat window, don’t just ask a question. Choose who you want to hear from. Let everyone else on the stage fade away.
By Joshua McDonald on January 18, 2026.
Exported from Medium on August 26, 2026.
Reader discussion