AI wrote your funnel copy and now it sounds like everyone else's

Why AI-written funnels are converting at 2019 rates

I reviewed 12 AI-generated landing pages last month. I could tell within three seconds that each one was written by AI. Technically clean, logically organized, perfectly forgettable. The conversion rates were consistently bad — not terrible, just bad in the specific steady way that means nobody hated the page enough to leave immediately, they just weren't moved enough to buy. The pages weren't wrong. They were empty. Someone had typed "write me a landing page for a marketing course" into a chat window, accepted whatever came back, and published it without changing a word.

AI is not the problem. I use AI for copy every single week and the output is faster than any human copywriter I've worked with. The problem is that most people hand AI the wrong inputs — and AI, being an extremely agreeable tool, produces output at exactly the level of specificity it receives from you.

• Solo founders who use AI to write their entire funnel without giving it a single line of real customer language first.
• Course creators who tried AI copy, watched it underperform their old pages, and went back to expensive copywriters without figuring out what went wrong.
• Agency owners who scaled AI content volume without noticing that their clients' conversion rates flatlined at the same time.

Funnel Baby's four-step AI copy brief

Step 1: Feed it your customer's actual words, not your description of them

The brief is not "my customer is a 35-year-old entrepreneur who wants to scale." The brief is what your customer said at 11pm when they couldn't sleep.

AI generates at the level of specificity it receives. Give it a demographic description and it writes for a demographic description — smooth, general, and identical to every other page targeting that demographic. Give it verbatim testimonials, real objections from sales calls, and word-for-word language from your community — and it writes copy that sounds like it was pulled directly from those conversations. The brief is the strategy. Everything after it is execution.

• Pull 5-10 verbatim testimonials — paste them directly into the brief, not a summary or paraphrase of them.
• Include the exact objections — "I've tried programs before and nothing ever stuck" is better input than "they're skeptical of online education."
• Add before-and-after language — what did life look like before they solved this problem? Quote it directly; do not paraphrase.
  • No real customer language yet? Find it in Facebook groups, Reddit threads, or product reviews in your niche.

Step 2: Write the hook yourself, then hand AI the body

The hook is the one part AI consistently gets wrong. Write it before you open the chat.

AI can structure a mechanism section, draft bullet points, and write transitions faster than anyone I know. What it cannot do is write a hook that feels genuinely unexpected — because AI hooks are constructed from patterns in training data and they read like hooks. Human hooks come from a specific, uncomfortable, honest observation that no model has been trained to reach first. Write the opening paragraph yourself. It can be rough. Then paste it into the chat and tell the model to match that tone and angle for everything that follows. The output will be meaningfully better.

• One specific hook — one scene, one number, one uncomfortable truth. No compound ideas in the opener.
• First person, real incident — "I audited 12 pages last month" lands harder than "many businesses struggle with copy."
• Lock the hook in the brief — tell the model the first paragraph is fixed and non-negotiable; ask it to match the voice, not improve the words.

Step 3: Edit for voice, not grammar

Your job after the AI draft is to remove the smooth, not to add polish.

Most people edit AI copy by cleaning up awkward phrasing and improving flow. That's backwards. The problem with AI output isn't roughness — it's that it's too smooth. AI removes all the friction that makes a real voice sound like a real person. Your edit pass should be looking for: sentences that sound like everyone else, transitions that are too clean, arguments that are too easy to agree with. Make it slightly harder to read. Slightly more uncomfortable. Add the thing you'd actually say in a conversation, even if it stings a little.

• Read it out loud — if you stumble nowhere, that's a bad sign. Real voice has texture and rhythm breaks.
• Punch up the word choice — replace "utilize" with "use," "facilitate" with "help," and "leverage" with literally anything else.
• Add one uncomfortable truth — something that might make a bad-fit buyer self-select out. AI never puts these in because AI writes for approval, not for fit.
  • Writing for fit and writing for approval are different goals. Only one of them converts.

Step 4: Test one section at a time, not the whole page

If the AI page underperforms, you won't know what broke unless you tested sections individually.

The most common post-writing mistake with AI copy is running the entire AI page against the entire current page. That's not a test — it's a coin flip with extra steps. If the AI version loses, you know nothing about why. If it wins, you know nothing about what specifically improved. Test the headline first. Then the hook. Then the mechanism section. Section by section, you build actual knowledge about what your audience responds to — and that knowledge applies to every page you build after this one.

• Test the headline first — it has the highest leverage. One clean variable, readable result within 500 visitors.
• Use heatmaps on the body — scroll depth and click-through reveal exactly where the AI version loses people versus where your current version does.
• Don't call it early — give each variation at least 500 unique visitors before you declare a winner.
  • Calling a winner at 80 visitors is how you make permanent decisions based on noise.

The honest part

"AI doesn't have bad taste. It has no taste. Taste is what you supply. If you don't supply it, the output is the average of everything the model has ever read."

The dropout rate here is high. Doing this correctly — real customer language in the brief, a human-written hook, a voice-edited draft, and section-by-section testing — takes longer than typing a prompt and accepting whatever comes back. Most founders spend 90 seconds on the brief and then spend four hours wondering why the page doesn't move people. I almost threw my laptop the last time I saw someone complain about AI copy while their brief was four words long. The brief is the work. The model is just faster than a blank page.

What this is really about

AI is a leverage multiplier for people who already know their customer. It is not a substitute for knowing them. The founders winning with AI-generated copy are not winning because the model is better — they're winning because they gave it a better brief than everyone else did. Good copy has always been about knowing your reader more specifically than anyone else in the room. AI accelerates whoever has done that research. It does nothing useful for whoever skipped it.

The model is not the shortcut. The brief is the shortcut. And writing the brief is the actual work most people are trying to avoid by using AI in the first place.

What to do this week

1. Pull five verbatim quotes from customers, prospects, or community members describing their problem in their own words. Save them to a document. That document is your brief.
2. Write the hook for your top-traffic page yourself — one sentence, one specific incident, first person. Do not use AI for this step.
3. Feed your brief and your hook to an AI model and ask it to write the rest of the page. Then edit the output for voice, not grammar.
4. Set up an A/B test: your current headline vs. the AI-generated headline on your best-traffic page. Run it for at least 30 days before you touch anything else.

The Bottom Line

AI copy that converts is built by founders who did the research — not founders who outsourced it. The model gives you speed. The brief gives you accuracy. Without the brief, you're just making generic mistakes faster than you used to.

Funnel Baby's pick: DotCom Secrets — the book that built ClickFunnels — the value-ladder playbook.