Why Your AI Copy Sounds Like Everyone Else's
You open ChatGPT or Claude, type something like "write a product description for my SaaS app", and get back three paragraphs that start with "Introducing" and end with "Try it free today." It's grammatically correct, professionally hollow, and utterly forgettable.
The problem isn't the AI. The problem is that vague prompts produce vague outputs. Product copy requires specificity — about the user, the pain, the tone, the constraints. When you give the model nothing to work with, it falls back on the average of everything it has ever seen. And the average of the internet is beige.
This tutorial walks through a five-step prompt engineering framework I use when writing copy for landing pages, feature announcements, onboarding flows, and microcopy. Each step builds on the last, and by the end you'll have a repeatable process you can apply to any product.
Step 1: Define the One Reader
Before you write a single prompt, answer this: who is the one person reading this copy right now, and what are they trying to decide?
Not a persona. Not a segment. One person, one moment, one decision.
Then bake that directly into your prompt. Don't leave it implicit.
Weak prompt:
Write a hero section headline for a project management tool.Stronger prompt:
Write a hero section headline for a project management tool.
The reader is a freelance web developer juggling 4–6 client projects at once.
They are landing on this page after clicking a Google ad.
They are skeptical — they've tried Trello and Notion and abandoned both.
They need to feel understood, not sold to.
The headline should address the chaos of context-switching, not generic "productivity."The second prompt gives the model a human being to write for. The output will reflect that.
Step 2: Assign a Voice with Evidence
"Friendly but professional" means nothing to a language model — or to a human copywriter, for that matter. Instead of adjectives, give examples.
Collect 3–5 real sentences from your brand, a competitor you admire, or a product whose voice resonates with your audience. Feed those as reference.
Match this tone and voice. Here are examples of copy I like:
1. "Your invoice is paid. Go do something you actually enjoy."
2. "We handle the boring parts so you can focus on the work that matters."
3. "No setup wizards. No 47-step onboarding. Just open it and go."
The voice is: direct, slightly dry, confident without being arrogant.
Avoid: exclamation marks, the word "seamless", phrases like "unlock your potential".That Avoid: list is underrated. Explicitly banning clichés forces the model to find fresher language.
Step 3: Constrain the Output Format
AI models default to whatever length feels natural given the prompt — which is usually too long. Product copy lives inside constraints: character limits, layout grids, button labels. Specify exactly what you need.
Deliverables:
- 1 headline: max 8 words
- 1 subheadline: max 20 words, expands on the headline
- 3 bullet points: each under 12 words, lead with a verb
- 1 CTA button label: max 4 words, not "Get Started" or "Sign Up"
Return only the copy. No explanations, no alternatives, no preamble.That last line — no explanations, no preamble — saves you from wading through three paragraphs of "Sure! Here's some copy for you..." before reaching the actual output.
Step 4: Use the "Steel Man" Revision Loop
Don't accept the first output. Instead, use a follow-up prompt that forces the model to critique its own work and then improve it.
Now play devil's advocate. You are a skeptical user who just read that copy.
List 3 specific objections or moments of confusion you'd have.
Then rewrite the copy to address each objection without making it longer.This is a surprisingly effective technique. The model will often identify real weaknesses — vague claims, assumed knowledge, missing proof — and the rewrite that follows is noticeably sharper.
You can also run a variant loop:
Give me 3 alternative headlines. Each should take a completely different angle:
- One leads with the outcome
- One leads with the pain
- One leads with the mechanism (how it works)
All must stay under 8 words.Seeing three angles side by side helps you make a real creative decision rather than just accepting the first plausible option.
Step 5: Ground It in Real User Language
The highest-converting copy often uses the exact words real users use to describe their problems. If you have customer interviews, support tickets, app store reviews, or survey responses, paste them directly into your prompt.
Here are 5 things real users have said about their frustration before finding us:
"I was spending more time updating my project tracker than actually working."
"I could never remember where I'd left things after the weekend."
"My clients kept asking for updates and I had no quick answer."
"Everything was spread across Slack, email, and my head."
"I tried every tool and they all made me feel like I was managing the tool, not my work."
Use the language patterns, the specific frustrations, and the emotional texture of these quotes.
Do not quote them directly. Distill them into copy.This technique — sometimes called voice of customer (VoC) seeding — produces copy that resonates because it's built from real human language, not synthetic marketing-speak.
Putting It All Together: A Full Prompt Template
Here's a reusable template that combines all five steps. Drop your specifics into each section.
## CONTEXT
Product: [what it is in one sentence]
Page/placement: [landing page hero / onboarding screen / feature callout / etc.]
## THE READER
[Describe one specific person, their situation, what they're deciding right now]
## VOICE & TONE
Match this voice: [paste 3–5 reference sentences]
Avoid: [list banned words and phrases]
## USER LANGUAGE
Real things users have said:
[paste 3–5 quotes from interviews, reviews, or support tickets]
## DELIVERABLES
[List exactly what you need with character/word limits]
Return only the copy. No preamble.
## AFTER FIRST DRAFT
Play devil's advocate: list 3 user objections to this copy, then rewrite to address them.What AI Can't Do (And What That Means for Your Process)
Prompt engineering gets you dramatically better first drafts, but there are things the model genuinely cannot do without you:
- It doesn't know your actual metrics. It can't tell you whether "free trial" or "start free" converts better for your audience. You need to test that.
- It doesn't feel your brand. It can approximate a voice from examples, but the judgment of whether something is off is still yours.
- It will hallucinate specifics. If you ask it to incorporate feature details, verify every claim it makes. AI is confident about things it's making up.
Think of it as a very fast junior copywriter who works at 3am and never gets tired. Your job is still to brief well, review carefully, and make the final call.
The Compounding Benefit
The real payoff of this framework isn't any single piece of copy — it's the prompts themselves. Once you've written a strong prompt for your hero section, you can reuse that structure for every new feature launch, every A/B variant, every seasonal update. The prompt becomes an asset.
Save your best prompts somewhere. Version them like you version code. Refine them as your product evolves. Over time, you end up with a prompt library that encodes your brand voice, your customer understanding, and your copy standards in a form that any team member — or any AI — can execute against.
That's the difference between using AI as a novelty and using it as infrastructure.
