Paste your product, your buyer, and one goal — get copy that sounds human and actually sells, not word soup.
Stop describing the product and start describing what changes for the buyer once they own it. The AI does this fast if you feed it the friction the product removes.
It can't guess your voice, but it copies well. Give it three examples of writing you'd sign your name to and it locks onto the pattern.
The trick is forcing different angles, not different words. Ask for ten angles first, then one line each — otherwise you get ten synonyms of the same sentence.
The subject line does 80% of the work and the first line does the rest. Write those with the AI before you touch the body.
Don't rewrite blind. Find out what they actually hated, then hand the AI the copy plus that reason so it fixes the real problem, not a random one.
Good copy isn't clever — it's clear. It says one thing the reader cares about and gets out of the way. The AI can write clever all day; your job is to force it toward clear.
Because it's trained to be safe and average, and 'average marketing copy' is a graveyard of 'unlock,' 'elevate,' and 'game-changing.' It reaches for the most common phrase, which is exactly the phrase everyone else already used.
You beat it by being specific. Feed it a real detail — a number, a name, a weird true fact about your product — and the generic mush has nothing to hide behind.
AI writes the copy; it doesn't know what people search for. That's where a tool like Semrush comes in — it tells you the real words your buyers type, so your copy targets demand that actually exists.
Pick one channel and one message. With three days you don't have time to be everywhere. Feed the AI your product, your one buyer, and your one goal, then have it draft a landing page, a launch email, and five social posts off that single message.
Then you edit ruthlessly for truth — the AI will overclaim, and you're the one who has to deliver. Cut anything you can't back up before it ships.
Add a hard rule to your prompt: 'Every claim must be one I can prove — flag anything you're inventing.' It'll actually mark the lines where it guessed at a benefit or a number.
Then either kill those lines or replace them with something true. Overpromising isn't just cringe — it's the fastest way to a refund and a bad review.
It can help you stop posting the same shape of post. Give the AI five of your dead posts and ask it what pattern is killing them — usually it's all announcement, no hook, no reason for anyone to stop scrolling.
Then have it rewrite three of them to open with a question, a number, or a strong opinion. It won't fix a bad product, but it'll fix boring.
They're not rivals — they do different halves of the job. Semrush tells you what to write about; Claude writes it. The best workflow uses both.
| Job | Claude Opus 4.8 | Semrush |
|---|---|---|
| Write the actual copy | Yes, its whole point | No |
| Find what people search | No, it guesses | Yes, real data |
| Check the competition | Vague | Yes, hard numbers |
| Match your brand voice | Yes, with samples | No |
If you know your buyer, write the rough draft yourself and let AI tighten it — your version carries truth the AI can't invent. If you're staring at a blank page, let AI draft and you edit for truth.
| Situation | You draft, AI improves | AI drafts, you edit |
|---|---|---|
| You know the buyer well | Best — keeps your insight | Loses your edge |
| Blank page paralysis | Slow | Best — beats the blank page |
| High-stakes copy | Best — more control | Risky, verify hard |
| Speed on volume | Slower | Best — fast batches |
It replaces the part of your job that was typing the obvious sentence — and honestly, good riddance. What it can't do is know your buyer's real fear, catch the claim that'll get you sued, or decide the strategy behind the words.
The people losing work are the ones who only did the typing. The ones who own the thinking now do it three times faster. Be the second kind.
Trust the writing, verify every fact. AI will confidently invent a statistic, a testimonial, or a guarantee you never made — and in marketing, a made-up claim is your problem, not the model's.
Read every number, every 'clinically proven,' every 'best in class' and ask: can I prove this? If not, cut it. The AI drafts fast; you're still the one who signs off.