Claude Prompts for Ad Copy: 7 Frameworks That Write Scroll-Stopping Ads in Minutes

It's 4pm and your campaign is supposed to go live tomorrow. The ad account is open, the audience is built, the budget is set. The only thing missing is the part that actually does the selling: the copy. You type a headline. Delete it. Type another. Twenty minutes later you have one mediocre variation and a creeping sense that you're about to spend real money sending strangers to a sentence you don't believe in.

Ad copy is the highest-stakes writing most business owners ever do, because every word is attached to a dollar. A good ad lowers your cost-per-click and makes the whole funnel cheaper. A weak one quietly taxes every sale you make for the life of the campaign. And unlike a blog post, you can't fix it slowly over time — the auction punishes you immediately.

This is exactly the kind of work AI is built for, if you prompt it correctly. The right Claude prompts for ad copy don't just spit out generic hype — they force the model to think like a direct-response copywriter: lead with the prospect's problem, sell the outcome, and give one clear reason to click now. Below are seven frameworks you can paste into Claude today, each engineered to produce ads you'd actually run.

Why most people get bad ad copy from AI

The reason "write me a Facebook ad for my product" returns garbage isn't the model — it's the prompt. A one-line request gives Claude nothing to work with, so it falls back on the blandest pattern in its training data: adjectives, exclamation marks, and "Unlock the power of…" openers that every disengaged reader has learned to scroll past.

Great ad copy comes from constraints and context, not from asking nicely. Before any of the prompts below, give Claude four things: who the customer is, the single most painful problem they have, the one outcome they want, and the platform the ad will run on. Those four inputs do more for output quality than any clever phrasing. The frameworks below are built to demand them.

1. The problem-agitate-solve ad

The oldest direct-response structure still works because it mirrors how buying decisions actually feel. Use this prompt:

"You are a direct-response copywriter. Write 3 Facebook ad variations for [product] using the Problem-Agitate-Solve framework. The customer is [describe them]. Their most painful problem is [problem]. The outcome they want is [outcome]. For each variation: open with the problem in their own words, agitate the cost of leaving it unsolved, then position [product] as the solution. Keep each under 100 words, conversational, no hype words like 'unlock' or 'revolutionary.' End with a clear CTA."

The "no hype words" constraint is doing heavy lifting — it strips out the exact phrases that signal "ad" to a skimming reader.

2. The big-promise headline battery

Headlines are where 80% of your ad's performance is decided, so generate them in bulk and test. Try:

"Generate 15 ad headlines for [product] aimed at [customer]. Mix these angles: the outcome they want, the problem they're escaping, a specific number or timeframe, a contrarian take, and curiosity. Each under 8 words. No clickbait that the product can't deliver on."

Fifteen headlines in one pass means you can pick the three strongest and split-test them instead of betting everything on the first thing you thought of at 4pm.

3. The platform-native rewrite

A Facebook ad and a Google search ad are different species. Once you have a winning message, have Claude reshape it for each placement:

"Take this core message: [paste]. Rewrite it as (a) a Facebook primary text under 125 characters, (b) three Google responsive search ad headlines at 30 characters max each, and (c) a TikTok-style hook that sounds like a person talking, not a brand. Preserve the core promise in all three."

This is where teams waste the most hours — manually re-cutting one idea for five channels. It's a 30-second job for a well-prompted model.

4. The objection-killer ad

Most clicks are lost not to disinterest but to doubt. Surface and answer the doubt inside the ad:

"List the top 5 reasons [customer] would hesitate to buy [product]. Then write one short ad that pre-empts the single biggest objection in the headline and dissolves it in the body. Make the reassurance feel honest, not defensive."

5. The social-proof angle

If you have any results, reviews, or numbers, turn them into the ad's spine:

"Write 3 ad variations for [product] built around this proof point: [testimonial / stat / result]. Lead with the proof, not the product. Make a skeptical reader think 'if it worked for them, it could work for me.'"

6. The ugly-but-effective native ad

Polished ads sometimes underperform a plain-text post that reads like a friend's recommendation. Ask for it directly:

"Write a Facebook ad for [product] that deliberately looks un-designed — like a real person posting a genuine recommendation in a group. First person, specific details, one moment of mild self-deprecation, then a soft mention of [product] as what fixed it. No salesy CTA — just a natural 'here's the link.'"

7. The angle-multiplier

When you've found a winner, don't stop — clone the structure across new angles:

"This ad is my best performer: [paste]. Without changing its structure or rhythm, write 5 new variations that each lead with a different customer pain point. Keep what's working; swap only the angle."

This single prompt is how small teams produce the 20-30 creative variations that modern ad platforms reward with cheaper delivery.

The mistakes that still cost you money

Even with strong prompts, three habits sink AI-written ads. First, shipping the first draft — Claude's job is to give you raw material; yours is to cut the 20% that sounds like a brand and keep the 80% that sounds like a person. Second, skipping the customer description — an ad written for "small business owners" converts at a fraction of one written for "solo bookkeepers who are drowning in client emails." Third, forgetting compliance — never let AI invent a claim, a statistic, or a guarantee your product can't back up; the auction and the platform's review team both punish it.

The pattern underneath all seven frameworks is the same: you supply the truth about your customer and your offer, and Claude supplies the volume and the structure. That division of labor is what turns a 4pm blank-page panic into ten testable ads before dinner.

Skip the setup entirely

Every prompt above is a starting point you'll refine over a few campaigns. If you'd rather skip the trial-and-error, the same thinking is already packaged into our 50 Claude prompts for business owners — a $17 pack with battle-tested prompts for ad copy, cold emails, proposals, and content, each pre-loaded with the constraints that separate ads people click from ads people scroll. Paste, fill in your customer and offer, and ship.

You don't need to become a copywriter to stop losing money on bad ads. You need a repeatable system that forces the model to do the disciplined part — and the discipline is in the prompt.