The Prompt Psychologist

Why AI Keeps Saying "Delve" (And Other Crimes Against Language)

8 min read

If I see the word "delve" one more time, I'm going to delve right into the ocean.

You know the words I'm talking about. "Leverage." "Synergy." "Robust." "Paradigm shift." AI loves these words like a teenager loves saying "literally."

But here's the thing: it's not the AI's fault. It's ours.

Where This Garbage Comes From

AI models are trained on the internet. And the internet is full of:

So when you give ChatGPT a vague prompt, it pattern-matches to... that. It thinks that's what "good writing" looks like because that's what it's seen the most.

The AI isn't trying to sound like a corporate drone. It's just doing what we trained it to do: mimic the most common patterns in its training data. And unfortunately, the internet is full of corporate droning.

The Worst Offenders

1. "Delve"

Nobody says this in real life. Nobody. If you walked into a coffee shop and said "Let's delve into the menu," people would ask if you're okay.

Why AI loves it: It appears in academic papers and "thought leadership" content. AI thinks it sounds sophisticated.

The fix: Use "explore," "dig into," "look at," or literally anything else.

Here's a test: Would you say this word out loud to a friend? No? Then don't let AI use it either.

2. "Leverage"

Unless you're talking about actual physics or financial instruments, stop.

Why AI loves it: Business blogs use it to mean "use" but fancier.

The fix: Just say "use." You're not impressing anyone.

Real talk: Every time someone says "leverage our core competencies," an English teacher dies inside. Don't be that person. Don't let your AI be that person.

3. "Robust"

What does this even mean anymore? Robust framework. Robust solution. Robust strategy. It's lost all meaning.

Why AI loves it: Sounds technical and important.

The fix: Be specific. Strong? Comprehensive? Reliable? Say that.

"Robust" has become a nothing word. It's what you say when you want to sound smart but don't want to commit to what you actually mean. Is it strong? Flexible? Complete? Just say the actual thing.

4. "Synergy"

If you use this word unironically, we can't be friends.

Why AI loves it: Peak 2000s corporate speak. Still haunting the training data.

The fix: "Work together." "Complement each other." Normal human words.

Fun fact: "Synergy" was voted one of the most annoying business buzzwords multiple years in a row. Yet here we are, still hearing it in 2026. Make it stop.

5. "Paradigm shift"

Unless you're literally talking about Thomas Kuhn's philosophy of science, you mean "big change."

Why AI loves it: Sounds dramatic and important.

The fix: "Major change." "New approach." "Complete rethink."

A paradigm shift is when Copernicus said "actually the Earth revolves around the sun." Your new CRM implementation is not a paradigm shift. Sorry.

6. "Utilize"

This is just "use" in a tuxedo. And it's not even a nice tuxedo. It's one of those rental ones that smells weird.

Why AI loves it: Longer = smarter, right? (Wrong.)

The fix: Just say "use."

7. "Facilitate"

"I'll facilitate a discussion" = "I'll lead a discussion."
"This will facilitate growth" = "This will help growth."

See? You don't need it.

The fix: Use the simpler verb hiding inside.

How to Stop This

Option 1: Explicitly Ban Words in Your Prompt

"Write this. Do NOT use: delve, leverage, robust, synergy, paradigm, utilize, facilitate, or any other corporate zombie words."

Sometimes you gotta be direct. AI is literal. If you say "don't use delve," it won't use delve.

Option 2: Give It Better Examples

"Write like [author/publication you actually like]. Not like a LinkedIn influencer who discovered a thesaurus."

Try: "Write like you're explaining this to a friend over beer. Use the words you'd actually say out loud."

Option 3: Set a Reading Level

"Write at an 8th-grade reading level. Short sentences. Common words. No jargon unless absolutely necessary — and if you use jargon, explain it."

This isn't about dumbing things down. It's about clarity. Hemingway wrote at an 8th-grade level. So did Orwell. They weren't stupid. They were clear.

Option 4: The "Would You Say This Out Loud?" Test

Add this to your prompt: "Write like you're speaking out loud to a smart person. If you wouldn't say a word in conversation, don't use it."

Nobody says "utilize" when they're talking to another human. They say "use." So should your AI.

The Real Problem

Here's the uncomfortable truth: AI writes like this because humans write like this.

We've been churning out soulless corporate content for so long that the AI thinks it's normal. Every "thought leadership" post about "leveraging synergies" trains the model to think that's acceptable professional writing.

Every time you let "delve" slide in your prompt, you're training the next generation of AI to think that's acceptable. Every time you don't specify tone, you're getting the average of all the bland business writing on the internet.

We're in a feedback loop: Humans write corporate slop → AI learns from it → AI generates corporate slop → Humans don't correct it → More training data for future models.

Break the cycle.

What Good Writing Actually Looks Like

Good writing is:

Bad writing hides behind big words because the ideas are small. Good writing uses small words because the ideas are big.

Einstein explained relativity using examples of trains and clocks. He didn't need to "leverage robust frameworks to facilitate paradigm shifts." He just... explained it clearly.

The Solution

Write prompts like you're directing a human writer. Because in a way, you are. You're directing a very literal, very pattern-matching, very "I learned English from Reddit and corporate blogs" writer.

So be specific. Be opinionated. Tell it what you hate. Tell it what you love. Give it examples. Set constraints.

Say things like:

And for the love of all that is holy, ban the word "delve."

A Challenge

For your next 10 prompts, add this line:

"Write like a human, not a business textbook. Ban: delve, leverage, robust, synergy, utilize, facilitate."

Watch how much better your output gets.

You're welcome.