Most AI copy is terrible. You can spot it from three sentences in. The hedging. The filler. The way it says everything and nothing at once.
But the problem is not AI. The problem is how people use it. They type "write me a landing page for my SaaS product" and expect magic. What they get is the copywriting equivalent of elevator music — technically correct, emotionally dead.
Here is the thing: AI can write copy that converts. Copy that sounds like a human with strong opinions wrote it. Copy that makes people click, read, and buy. But it requires a fundamentally different approach than most people take.
This guide is the approach I use. It is the same process I refined while running marketing across 14 markets — adapted for a world where AI does your first draft.
Why Most AI Copy Fails
Before fixing the problem, you need to understand it. AI copy fails for three specific, fixable reasons.
It Defaults to the Middle
Large language models are trained on the internet. The internet is full of mediocre copy. So when you ask AI to write copy, it gives you the statistical average of all the copy it has ever seen. That average is bland.
You get phrases like "in today's fast-paced world" and "unlock your full potential" and "seamless, end-to-end solution." These phrases exist because thousands of bad copywriters used them. The AI learned from them.
It Hedges Everything
AI is trained to be helpful, harmless, and honest. The "harmless" part means it avoids strong claims. Instead of "this will double your revenue," it writes "this may help potentially increase your revenue over time." Hedging kills conversion.
It Lacks Specificity
Good copy is specific. "We helped 847 e-commerce brands increase AOV by 23% in 90 days" beats "we help businesses grow" every time. AI does not have your data, your case studies, or your customer stories. So it writes generically.
The fix for all three problems is the same: stop treating AI as a copywriter. Start treating it as a draft engine.
The AI Draft to Human Edit Workflow
This is the core workflow. Everything else in this guide builds on it.
Step 1: Brief the AI Like You Would Brief a Human
You would never hand a freelance copywriter a one-line brief and expect great work. Do not do it with AI either.
Your brief should include:
- Who you are writing for (specific persona, not "small business owners")
- What you are selling (specific product, feature, or offer)
- Why they should care (the pain point you solve, in their words)
- Where this copy will live (landing page hero, email subject line, Facebook ad)
- Voice constraints (tone, reading level, words to avoid, words to use)
- Proof points (specific numbers, customer quotes, case studies)
Step 2: Generate Multiple Drafts
Never accept the first output. Generate three to five variations. Ask for different angles:
- Lead with the pain point
- Lead with the outcome
- Lead with social proof
- Lead with a contrarian take
- Lead with a question
Step 3: Frankenstein the Best Parts
Your final copy will rarely come from a single draft. Take the strongest headline from draft two, the best body paragraph from draft four, and the CTA from draft one. Combine and rewrite.
Step 4: Add What AI Cannot
This is the step most people skip, and it is the step that matters most.
Add your real numbers. Add your customer's actual words. Add the specific, ugly, concrete details that make copy believable. Replace every generic claim with a specific one.
Step 5: Read It Aloud
If it sounds like a robot wrote it, edit until it does not. Good copy has rhythm. It has short sentences. And longer ones that build on them, layering detail and nuance until the reader feels something specific. Like that.
Prompt Engineering for Copy That Converts
Generic prompts produce generic copy. Here are the prompt structures I use for different copy types, refined through hundreds of iterations.
Headlines
Headlines are where AI shines — if you constrain it properly. AI is good at generating volume, and headlines are a volume game.
Write 20 headlines for [product/offer].
Audience: [specific persona with specific pain]
Goal: [click, sign up, buy, read more]
Tone: [specific tone — not just "professional"]
Format constraints: Under 10 words. No colons. No "unlock" or "revolutionize."
For each headline, use a different angle:
- Numbers and specificity
- Unexpected contrast
- Direct address of pain
- Bold claim with proof
- Question that implies the answer
Examples of headlines I like:
[paste 3-5 real headlines you admire]
The key detail: including examples of headlines you like. This gives the AI a concrete target instead of the statistical average.
Landing Page Hero Sections
Write a landing page hero section for [product].
The visitor just clicked an ad that said: "[exact ad copy]"
They expect to see: [what the page should immediately confirm]
Target customer: [detailed persona — job title, company size, specific frustration]
Structure:
- Headline (under 12 words, makes a specific claim)
- Subheadline (one sentence, supports the claim with a proof point)
- 3 bullet points (each starts with a verb, each is under 8 words)
- CTA button text (under 5 words, action-oriented)
Voice: [paste a paragraph of copy you've written that nails your voice]
Do NOT use: "seamless," "cutting-edge," "solution," "leverage," "empower"
Notice the constraint about matching the ad copy. Message match is one of the highest-leverage conversion tactics, and AI helps you maintain it across campaigns at scale.
Ad Copy (Facebook and Google)
Write 10 Facebook ad variations for [product/offer].
Primary text: 2-3 sentences max.
Headline: Under 40 characters.
Description: Under 30 characters.
Target audience: [specific — not just demographics, but psychographics]
The one thing this ad must communicate: [single core message]
Social proof to reference: [specific number, testimonial, or credential]
Winning ad examples from our account (match this energy):
[paste 2-3 of your best-performing ads]
Each variation should test a different:
1. Hook (first 5 words)
2. Emotional angle
3. Proof point emphasis
Email Subject Lines
Write 30 email subject lines for [campaign type].
Context: [what the email contains, what action you want]
Sender name: [your from name — affects what feels natural]
Audience relationship: [cold prospect / warm lead / existing customer]
Constraints:
- Under 50 characters (mobile preview)
- No ALL CAPS words
- No spam trigger words (free, guarantee, act now)
- No emoji unless I specify
Test these angles across the 30:
- Curiosity gap
- Specificity (numbers, names)
- Direct benefit statement
- Pattern interrupt (unexpected)
- Social proof (borrowed authority)
Our top-performing subject lines historically:
[paste 3-5 with open rates if you have them]
Training AI on Your Brand Voice
This is where most people's AI copywriting goes from "okay" to genuinely useful. A brand voice prompt is a reusable block of context you prepend to every copywriting request.
How to Build Your Brand Voice Prompt
Start by collecting 10 to 15 pieces of copy you have written that nail your voice. Could be emails, social posts, landing pages, even Slack messages that got a strong reaction.
Then extract the patterns:
Sentence structure. Do you use short, punchy sentences? Long flowing ones? A mix? What is the ratio?
Word choice. What words do you use a lot? What words do you never use? Do you lean formal or casual? Do you use slang? Industry jargon?
Rhetorical devices. Do you ask questions? Use analogies? Reference pop culture? Tell stories?
Emotional register. Are you warm? Direct? Sarcastic? Urgent? Calm?
Write all of this into a prompt block:
BRAND VOICE GUIDE:
- Sentence length: Mix of very short (3-5 words) and medium (10-15 words). Never over 20 words.
- Tone: Direct, confident, slightly irreverent. Like a smart friend who happens to be an expert.
- Reading level: Grade 7-8. No jargon unless defining it.
- Always: Use "you" more than "we." Lead with specifics. Use real numbers.
- Never: Hedge with "might," "could," "potentially." Never use "leverage," "synergy," "robust."
- Personality: Opinionated. Take stances. It is okay to say "don't do this."
- Structure: Short paragraphs (1-3 sentences). Frequent subheads. Bullet points for lists of 3+.
- Examples of our voice:
[paste 3-5 short samples]
Save this somewhere accessible. Paste it at the top of every copywriting prompt. Update it quarterly as your voice evolves.
The Voice Calibration Test
After generating copy with your brand voice prompt, run this test: show the copy to someone who reads your content regularly. Can they tell it was AI-assisted? If yes, the voice prompt needs work. If no, you are calibrated.
Specific Copy Types: What Works and What Does Not
Landing Pages
What AI does well: Structure. AI is surprisingly good at organizing landing page sections in a logical flow — hero, problem, solution, social proof, features, CTA. It understands conversion page architecture because it has seen thousands of them.
What AI does poorly: Specificity. AI landing pages read like templates because they are. Every claim is generic. Every testimonial sounds made up. Every feature description could apply to any product.
The fix: Use AI for structure and first drafts of each section. Then replace every generic element with something specific to your product. Real customer quotes. Real performance numbers. Real screenshots. The specific, concrete details that your competitor cannot copy because they are yours.
Email Campaigns
What AI does well: Sequence logic. AI understands email nurture flows — welcome sequences, onboarding drips, re-engagement campaigns. It can map out a logical progression of messages.
What AI does poorly: Personality. AI emails sound like they were written by a committee. They are polite, professional, and completely forgettable.
The fix: Write the first email yourself. Nail your voice. Then use AI to generate the remaining emails in the sequence, using your first email as a voice reference. Edit each one to add personal anecdotes, specific references, and genuine personality.
Social Media Copy
What AI does well: Volume and variation. Need 30 tweet variations testing different hooks? AI handles this in minutes.
What AI does poorly: Cultural awareness. AI does not know what is trending right now, what memes are relevant, or what tone the discourse has taken this week. It also tends to over-explain. Social copy should be punchy and incomplete — it should make you want to click, not give you the whole story.
The fix: Use AI for the raw variations, then edit each one down to half the length. Cut the explanations. Add the timely references. Make it sound like something a real person would actually post.
Ad Copy
What AI does well: Hook generation. The first line of an ad determines everything, and AI can generate dozens of hooks quickly for testing.
What AI does poorly: Authenticity. AI ad copy often sounds like ad copy — polished, professional, and clearly trying to sell you something. The best-performing ads in most markets sound conversational, slightly rough, real.
The fix: Generate hooks with AI, then rewrite each one as if you were texting a friend about the product. Remove the polish. Add the imperfections that make it feel human.
Tools Compared: What to Use and When
I have tested most of the major AI copywriting tools. Here is what I actually think.
Claude is my primary tool for long-form copy and anything requiring nuance. It handles complex brand voice instructions better than anything else I have used. It is also better at following structural constraints — when I say "under 10 words," it actually stays under 10 words.
ChatGPT is strong for brainstorming and ideation. It is good at generating large volumes of variations quickly. The custom GPTs feature is useful for creating reusable copywriting assistants with baked-in brand context.
Jasper adds a layer of templates and campaign workflows on top of the AI. If you have a team producing copy at scale, the workflow features justify the cost. For solo operators, the underlying models do the same thing with good prompts.
Copy.ai is best for short-form content — social posts, ad copy, product descriptions. It has purpose-built templates that speed up these specific use cases.
Writer focuses on brand consistency. If you are a larger organization with multiple people producing copy, Writer's style guide enforcement and brand voice features are genuinely useful.
The honest truth: the tool matters less than your prompting skill and editing process. A skilled copywriter with any of these tools will produce better output than a novice with the "best" tool.
The Editing Checklist
After generating AI copy, run through this checklist before publishing:
1. The Specificity Check. Find every generic claim. Replace it with a specific one. "Helps you grow" becomes "helped 340 DTC brands increase repeat purchase rate by 18%."
2. The Hedge Hunt. Search for "may," "might," "could," "potentially," "help," "can." Either make the claim stronger or cut the sentence.
3. The Voice Test. Read it aloud. Does it sound like you? Or does it sound like LinkedIn on a Monday morning? Edit until it sounds like something you would actually say.
4. The "So What" Test. After every sentence, ask "so what?" If the answer is not obvious, the sentence needs to either make its point clearer or get cut.
5. The Proof Test. Every claim should have a proof point within two sentences. No proof? Add it or remove the claim.
6. The Jargon Sweep. Flag every industry term. Is your audience guaranteed to know it? If not, explain it or replace it.
7. The Length Cut. Take your final copy and cut 20%. Almost all AI copy is too long. Cutting forces you to keep only what earns its place.
Common Mistakes and How to Avoid Them
Mistake: Using AI output as final copy. AI gives you a first draft. Always. Treating it as finished work produces mediocre results and trains your audience to tune you out.
Mistake: Not providing examples. The single highest-leverage thing you can do in an AI copywriting prompt is include examples. Examples of what you want. Examples of what you do not want. The AI learns more from three good examples than from three paragraphs of instructions.
Mistake: Asking for one version. Always generate multiple variations. The best output is rarely the first one. Ask for five to ten versions and pick the best elements from each.
Mistake: Forgetting the audience context. AI does not know your audience unless you tell it. The more specific your audience description, the more targeted the output. "SaaS founders with $1-5M ARR who are struggling with churn" beats "business owners."
Mistake: Ignoring the format constraints. A headline has different rules than body copy. An email subject line has different rules than a blog title. Always specify the format, the character limits, and the platform.
Building Your AI Copywriting Workflow
Here is the workflow I recommend for teams adopting AI copywriting:
Week 1-2: Build your brand voice prompt. Collect examples, extract patterns, write the prompt, test it against real copy needs.
Week 3-4: Create prompt templates for your three most common copy types. Headlines, emails, and ads are where most teams start. Build the templates, test them, refine based on output quality.
Month 2: Establish the editorial workflow. Who generates the AI drafts? Who edits? What is the quality bar? Create a lightweight checklist (use mine above as a starting point) and make it part of the process.
Month 3 and beyond: Measure and optimize. Track which AI-assisted copy performs better or worse than fully human copy. Use the data to refine your prompts and editing process. Build a library of your best-performing prompt-output pairs.
The Bottom Line
AI copywriting is not about replacing human copywriters. It is about making them faster and more prolific. The best results come from combining AI speed with human judgment — AI for volume and structure, humans for voice and specificity.
The companies that win with AI copywriting will not be the ones with the best tools. They will be the ones with the best editing processes and the clearest understanding of their own brand voice.
Start with the workflow. Build your brand voice prompt. Create your prompt templates. Edit ruthlessly. The AI is the easy part. The discipline is what makes it work.