Let’s talk about “delve.”

If you’ve used ChatGPT for more than 10 minutes, you’ve seen it. “Let’s delve into this topic.” “We’ll delve deeper.” “To delve into the nuances...”

Nobody talks like this. Nobody has ever said “delve” in a conversation. Your friend has never texted you “let’s delve into dinner options.” And yet AI uses it constantly.

Why does this happen?

Language models learn from training data. And that training data is disproportionately filled with academic papers, Wikipedia articles, and formal writing — all genres where “delve” appears far more frequently than in everyday speech.

AI doesn’t know that “delve” sounds weird. It just knows that in the patterns it learned, “delve” is a statistically likely word to use when introducing a topic. It’s AI’s comfort blanket.

The same problem, bigger scale

“Delve” is just the most obvious symptom. The same pattern produces: leverage, synergy, innovative, cutting-edge, game-changer, tapestry, embark, foster, harness, navigate, unlock, empower.

These are all words that AI defaults to because they appear frequently in formal writing. They make everything sound corporate, generic, and immediately identifiable as AI-generated.

The one-line fix

Add this to any prompt:

Do NOT use the following words: delve, leverage, synergy, innovative, cutting-edge, game-changer, tapestry, embark, foster, harness, navigate, unlock, empower, revolutionize, streamline.

That’s it. One line. The output will immediately sound more human.

Why this works

You’re overriding AI’s default word choices. When you ban “delve,” AI has to find an alternative — and the alternatives are almost always simpler, more natural words like “explore,” “look at,” or “examine.”

The banned words list is the single highest-ROI prompt technique I’ve found. It takes 10 seconds to add and transforms every output.

Build your own banned list

My full database has 60+ words. You can get it free in the 5 Prompt Psychology Frameworks guide. But start with the line above — it’ll fix 80% of the “this sounds like AI wrote it” problem immediately.

Your prompts deserve better than “delve.”

Aisha was paying $4,000 a month for a copywriting agency.

The copy was fine. Professional. Clean. But the back-and-forth was killing her. Every project took 3 rounds of revisions. Every email started with “as per our previous discussion.” Every deliverable sounded like it was written by a committee.

Then she learned to prompt.

What she was doing before

Like most people, Aisha was typing things like:

“Write me a marketing email for my product launch.”

And getting exactly what you’d expect: generic, corporate, buzzword-filled copy that sounded like every other marketing email in existence.

What she changed

Three things. That’s it.

1. She gave AI her actual voice. Not “professional tone.” She pasted 3 paragraphs of her own writing and said “match this voice exactly.”

2. She banned specific words. Not just vibes. A list: “Do NOT use innovative, excited, thrilled, game-changer, or leverage.”

3. She iterated instead of restarting. Instead of saying “try again” (which resets everything), she said: “The structure is great. Keep it. Make the tone more conversational and add a joke in paragraph 2.”

The result

Within 2 weeks, Aisha’s AI-generated copy was better than the agency’s. Her clients noticed. One said: “Did you hire someone new? The writing feels more... you.”

She cancelled the agency. Annual savings: $48,000.

The unhinged part

Here’s the thing nobody talks about: the agency was also using AI. They were just charging Aisha $4K/month to do it badly.

The skill that saved her $48K wasn’t some secret AI tool. It was learning how to communicate with AI. Three frameworks. One afternoon of learning. Forty-eight thousand dollars in savings.

The ROI on learning to prompt properly is genuinely insane.

Imagine walking into a restaurant.

You sit down. The waiter comes over. You look them dead in the eye and say: “Food please.”

The waiter, bless them, brings you a boiled egg.

You technically got what you asked for. You just asked terribly.

This is what you’re doing with AI

Every time you type “write me a blog post” or “create a marketing plan” or “help me with my resume,” you’re saying “food please” to a very capable chef.

And you’re getting a boiled egg. Every. Single. Time.

The Specificity Ladder

Here’s how to stop ordering boiled eggs:

Level 1 (boiled egg): “Write me a blog post.”

Level 2 (microwaved lasagna): “Write me a blog post about marketing.”

Level 3 (decent restaurant): “Write a 600-word blog post about email marketing for solopreneurs.”

Level 4 (Michelin star): “You are a senior content strategist. Write a 600-word blog post about why solopreneurs fail at email marketing. Audience: founders doing $10-50K/month. Include 3 mistakes with fixes. Tone: conversational, slightly provocative. Do NOT use: leverage, innovative, game-changer. End with a clear CTA.”

The difference

Level 1 and Level 4 use the same AI. Same subscription. Same technology. The difference is 30 seconds of thinking before you type.

30 seconds of thinking saves 30 minutes of re-prompting.

That’s a 60x return on effort. And most people are too impatient to take it.

Stop ordering boiled eggs.

I’m going to say something that might upset the “passive income prompt pack” crowd:

Generic prompt templates are useless.

Not “less effective.” Not “could be better.” Useless.

The prescription glasses analogy

Using someone else’s prompt template is like wearing someone else’s prescription glasses. They were made for a different person, with different eyes, looking at different things. They might technically sit on your face, but they won’t help you see.

Why templates fail

1. No context. A template that says “Act as a marketing expert and write a campaign for [PRODUCT]” has zero information about YOUR product, YOUR audience, YOUR budget, or YOUR goals.

2. No exclusions. Templates never include banned words or things to avoid. So you get generic, buzzword-filled output every time.

3. No voice. Templates produce the same robotic tone for everyone. Your output sounds identical to the 10,000 other people using the same template.

What works instead

Learning why certain prompt structures produce better results. When you understand the psychology, you can write your own prompts from scratch — prompts that are tailored to YOUR specific situation.

You don’t need 500 templates. You need 5 frameworks.

One teaches you to assign roles. One teaches you to set exclusions. One teaches you to iterate instead of restart. One teaches you specificity. One teaches you voice matching.

With those 5 frameworks, you can generate infinite prompts — all customized, all specific, all producing outputs that actually sound like a human wrote them.

That’s the difference between copying someone’s homework and understanding the subject.

I’m going to save you months of frustration in about 60 seconds.

Every good prompt has 5 elements. Miss any one of them and your output will have “boiled egg energy.” Get all 5 right and you’re in the top 1% of AI users. Here they are:

1. Role

Tell AI who to be. Not “be professional.” Tell it exactly who.

Bad: “Write me an email.”

Good: “You are a senior copywriter with 15 years experience in SaaS.”

In plain English: You’re giving AI a costume to wear. Without it, AI shows up in pajamas.

2. Context

Give background info. Your situation, your audience, your goal.

Bad: “Write a marketing plan.”

Good: “I run a 1-person SaaS company selling a $29/mo tool to remote teams of 5-20 people. Budget: $500/mo.”

In plain English: You’re telling the GPS your actual address instead of just saying “take me somewhere.”

3. Task

State clearly what you need. One sentence. No ambiguity.

Bad: “Help me with marketing.”

Good: “Create a 90-day organic marketing plan with 3 channels ranked by ROI.”

4. Format

Specify the structure and length of the output.

Bad: (nothing)

Good: “Format as a table with columns: Channel, Action, Timeline, Expected Result. Under 500 words.”

In plain English: Without this, AI writes a 47-paragraph essay when you wanted 3 bullet points.

5. Exclusions

Tell AI what NOT to do. This is the secret weapon most people skip.

Bad: (nothing)

Good: “Do NOT use buzzwords. Do NOT recommend paid ads. Do NOT start with ‘In today’s fast-paced world.’”

In plain English: You’re telling the chef “no onions.” Simple, but makes the meal way better.

That’s it.

5 elements. Role, Context, Task, Format, Exclusions. Get all 5 right and your AI outputs will be in the top 1%.

Most people get 1-2 right and wonder why AI “doesn’t work.”

Now you know what they don’t.

I need to tell you something that might ruin your day.

Every time you type “try again” into ChatGPT, you’re resetting the entire conversation context. AI doesn’t remember what worked. It doesn’t remember what you liked. It starts from scratch.

You’re not iterating. You’re amnesia-ing.

Why “try again” is the worst prompt

When you say “try again,” AI interprets it as: “Everything I just did was wrong. Generate something completely different.” So it throws away the parts that were working and starts over.

That paragraph structure you liked? Gone. The tone that was almost right? Deleted. The examples that were relevant? Nowhere to be found.

What to say instead

Be a coach, not a reset button. Here are the magic words:

“The structure is great — keep it. But make the tone more conversational and replace the third example with something about freelancers.”

Three sentences. You told AI what worked (structure), what to change (tone), and what to swap (example). AI keeps the good parts and fixes the bad parts.

The coaching formula

1. What worked: “The structure/format/length is good.”

2. What to change: “Make the tone/voice/angle more [specific].”

3. What to swap: “Replace [specific part] with [specific alternative].”

This takes the same amount of time as typing “try again” but produces dramatically better results.

You’re welcome.

Here’s a question: why does everything AI writes sound like the same person?

That person is a mid-level corporate employee who uses “leverage” unironically, starts every email with “I hope this finds you well,” and has never made a joke in their professional life.

That’s AI’s default voice. And it’s not your voice.

The Voice Card technique

A Voice Card is a paragraph that describes YOUR writing style. You paste it at the start of any conversation, and AI matches your voice for everything that follows.

Here’s how to build one:

Step 1: Find 3 things you’ve written that sound like “you.” Emails, social posts, anything.

Step 2: Describe your style in specific terms. Not “professional” — that means nothing. Instead:

“Short punchy sentences. Max 12 words per sentence. Slightly irreverent. Uses analogies from everyday life. Backs every claim with a specific number. Never uses jargon — if a simpler word exists, use it.”

Step 3: Add what you NEVER sound like:

“Never uses: leverage, synergy, innovative, game-changer. Never starts with ‘In today’s fast-paced world.’ Never ends with ‘In conclusion.’”

Step 4: Paste this at the start of every conversation.

Why this works

AI is a pattern matcher. When you give it your Voice Card, you’re giving it a pattern to match. The more specific your description, the closer the match.

“Professional tone” gives AI nothing to work with. “Short punchy sentences, slightly irreverent, max 12 words” gives AI everything.

One client told me: “My clients think I hired a writer. It’s just me and my Voice Card.”

Build yours. It takes 5 minutes and changes every output forever.

Every week someone asks me: “Will AI replace my job?”

Wrong question.

The right question is: “Will someone who uses AI better than me replace my job?”

And the answer is: probably, yeah.

The uncomfortable math

Two people in the same role. Same skills. Same experience. Same salary.

Person A types “write me a marketing plan” and gets generic garbage. Spends 3 hours cleaning it up.

Person B writes a 5-line prompt with role, context, format, tone, and exclusions. Gets a usable first draft in 30 seconds. Spends 20 minutes polishing.

Person B just did in 20 minutes what Person A did in 3 hours. Same quality. Same AI. Same subscription price.

Who do you think gets promoted?

The skill gap is in the prompt

Everyone has access to the same AI. ChatGPT doesn’t have a “premium intelligence” tier. The tool is identical for everyone.

The only variable is how you use it. And right now, 95% of people use it terribly.

That’s not an insult. It’s an opportunity. Because the 5% who learn to prompt well have an absurd competitive advantage — and the barrier to joining them is embarrassingly low.

An afternoon of learning

Learning to prompt well isn’t a 6-month course. It’s an afternoon. Maybe a weekend if you want to get fancy.

5 frameworks. A banned words list. Some practice.

That’s the difference between “AI is overhyped” and “AI is my unfair advantage.”

The clock is ticking. Your competitors are learning this right now.

Before you hit enter on any prompt, take 10 seconds and ask yourself 3 questions.

If any answer is “no,” your prompt isn’t ready.

Question 1: Would a stranger understand what I want?

Read your prompt as if you’re a stranger seeing it for the first time. Does it make sense without any context about your life, your business, or your situation?

If not, add context. AI is that stranger.

Question 2: Did I tell AI what NOT to do?

This is the single most skipped element in prompting. And it’s arguably the most powerful.

Exclusions kill generic filler. “Do NOT use buzzwords. Do NOT write more than 200 words. Do NOT start with a preamble.”

10 words of exclusions save you 30 minutes of editing.

Question 3: Did I specify the format?

Length. Structure. Number of items. Tone.

If you don’t specify, AI defaults to “write a college essay that says nothing in 500 words.” You have to override that manually.

The math

10 seconds of thinking. 3 questions. Fixes 80% of bad prompts.

That’s a 30-minute time savings for a 10-second investment.

The ROI is 180x. I can’t think of any other skill with a 180x return.

Use the rule. Every time. No exceptions.

Not all prompt mistakes are created equal. Some cost you 5 minutes. Others cost you clients. Here’s the definitive tier list.

S-Tier: Career-Ending

Sending AI output to a client without reading it. I’ve seen proposals with placeholder text like “[insert company name here]” sent to real clients. I’ve seen emails signed “Best regards, ChatGPT.” Read. Your. Output.

Using AI for sensitive topics without fact-checking. AI hallucinates. It makes things up with complete confidence. If you’re using AI for legal, medical, or financial content, verify every claim.

A-Tier: Reputation-Damaging

Not removing AI artifacts. “Certainly! I’d be happy to help!” as the opening line of a client deliverable. “In conclusion” at the end of an email. “It’s worth noting that” in a casual Slack message. These scream “I didn’t even read what AI wrote.”

Using the same prompt template as everyone else. When 10,000 people use the same “best LinkedIn post prompt,” the output is immediately recognizable. Your audience can smell template-generated content.

B-Tier: Time-Wasting

Saying “try again” instead of iterating. Resets everything. Wastes the good parts. Costs you 10-30 minutes every time.

Not specifying format. You wanted 3 bullet points. AI wrote 47 paragraphs. Now you’re spending 20 minutes editing instead of the 10 seconds it would have taken to specify.

Forgetting exclusions. Welcome to Leverage City, population: your entire document.

C-Tier: Mildly Annoying

Not setting a tone. Output sounds corporate when you wanted casual. Fixable, but annoying.

Being too vague about context. AI guesses. Sometimes it guesses right. Usually it doesn’t.

Asking for “creative” content. AI interprets “creative” as “add 47 metaphors about journeys and tapestries.” Be specific about WHAT KIND of creative.

The pattern

Notice how every mistake on this list is solved by the same thing: being more specific in your prompt.

Specificity isn’t just a nice-to-have. It’s the difference between S-tier disasters and consistently great output.

Learn the frameworks. Save your career. Or at least save your afternoon.