Do you think you can spot AI writing? What if I add a 🚀? How about a “Here’s the thing”? Even better, the dreaded “—” (that’s an em dash, btw). Do you think AI wrote the previous sentences? Yup. So how do you get AI to avoid using AI’isms? I get asked this a lot. People assume there’s some trick, some secret prompt formula I’m using.

There’s no trick. I just write long, detailed prompts. Then I go back and forth with the AI until it sounds right. That’s the whole thing. No magic. But it matters because most people are doing the opposite—typing quick, conversational requests and expecting the AI to read their mind.

Take email drafts. Ask any AI to “Write an email to my customers about our new business hours” and you’ll get something that starts with “I hope this message finds you well” and includes phrases like “I am pleased to inform you” and “please do not hesitate to reach out.”

The AI wasn’t broken. When you give it a vague prompt, it’s basically calculating “what words most commonly appear together in emails in my training data”—and those corporate clichés won the statistical lottery. They appear in millions of generic business emails, so when the AI has no other direction, it reaches for them.

It’s not trying to sound robotic. It’s doing what it was trained to do: predict the most probable next words. And “I hope this message finds you well” is statistically very probable in the “opening of a business email” category.

Those phrases aren’t just bad writing. They’re actually telling you something. When you see them, the AI is basically saying your prompt was too vague. It’s filling in gaps with the most common patterns it knows.

The fix? Add real context.

Instead of “Write an email about new hours,” try: “You’re the owner of a family bakery writing to loyal customers. Draft a warm, casual email announcing our new Saturday hours (now open 7 AM instead of 8 AM). Mention we’re doing this because customers kept asking for earlier weekend service. Keep it under three short paragraphs and end with genuine appreciation, not corporate speak.”

Same request, but a completely different output. The second version tells the AI who you are, who you’re talking to, what tone to use, and what to avoid. You’re not being picky. You’re being clear.

I learned this the hard way. Early on, I’d get frustrated when AI spit out something generic, like it wasn’t even trying. Took me a while to realize the problem wasn’t the AI—it was my instructions. I was asking for “a social media post” when I should’ve been saying “a short LinkedIn post for small business owners in the Gulf Coast, conversational tone, leading with a question about their biggest operational headache.”

This works for everything. Social media posts, project summaries, meeting notes, whatever you’re creating. The more specific you are about context, audience, and tone, the less your output sounds like a customer service bot.

The chat interface wants you to be casual. The AI needs you to be clear. And once you understand that gap, the “secret” stops being secret. You just have to be willing to write a longer prompt.

—————————

I encourage everyone to join our Discord server. It’s free and is a great resource for people looking to stay connected to the local tech community. You can join the LA-AI Discord server here.

// community highlights

Know someone who’d enjoy this?

Don’t miss the next issue.