AI Copywriting for Ads: 5 Frameworks with Prompts

Most marketers use AI to generate ad copy and end up with generic output that loses to a skilled human copywriter. The problem isn’t the model — it’s the prompt. No structure, no results.

Five copywriting frameworks, applied through AI prompts, that produce ad copy worth testing.

Why AI without a framework produces mediocre copy

A prompt like “write an ad for an online Excel course” gives you exactly what you’d expect: generic phrases, a feature list, a “sign up now” call to action. The text doesn’t hook because the model has no constraints on how to build its argument.

A framework fixes this. It tells the model which emotion to trigger first, when to introduce the product, and how to lead to action. AI follows structural instructions well. The more precisely you define the framework in the prompt, the closer you get to copy that a skilled copywriter would write.

The five frameworks below cover 90% of ad copywriting tasks — from short ads to landing pages.

PAS: Problem — Agitate — Solution

The most direct framework. Three steps: name the problem, amplify the pain, offer the solution. Works well in short formats: search ads, social posts, email subject lines.

When to use: the product solves a specific, conscious pain point the audience already recognizes.

Prompt

You are a direct response copywriter. Write ad copy using the PAS framework.

Product: [name and brief description]
Target audience: [who buys]
Main audience problem: [one specific pain point]
Format: [social post / email / ad]
Length: [word count or character count]

Structure:
1. Problem — open with a problem the audience will immediately recognize as their own. A specific situation, not an abstraction.
2. Agitate — show the consequences of inaction. What happens if the problem goes unsolved. Use numbers or time frames.
3. Solution — present the product as the logical way out. One key result, not a feature list.

Constraints: no exclamation marks, no word "unique," no generalizations like "everyone knows" or "we've all been there."

Example: online Excel course

Before (no framework):

Learn Excel at a professional level! Our course includes 50+ lessons, practical assignments, and a certificate. Suitable for beginners and advanced users. Sign up now with 40% off.

After (PAS):

The report your colleague assembles in 20 minutes takes you half a day. Every week that’s 10+ hours of manual spreadsheet work: copying data between sheets, formulas that break, recalculating totals by hand. Over a quarter, that adds up to 130 hours spent on tasks that three functions could handle. The “Excel for Analysts” course closes that gap in 6 weeks. Week one result: pivot tables and VLOOKUP that cut your daily report down to 15 minutes.

The difference: the first version describes the course. The second describes the reader’s situation and shows a concrete result with a time frame.

AIDA: Attention — Interest — Desire — Action

A classic framework — it’s been around since 1898. Four phases guide the reader from first contact to action. Best for longer formats: landing pages, email newsletters, product descriptions.

When to use: the audience doesn’t yet recognize the problem or isn’t familiar with the product.

Prompt

You are a conversion copywriter with direct response experience. Write copy using the AIDA framework.

Product: [name and description]
Target audience: [demographics + usage context]
Key benefit: [one, the strongest]
Social proof: [numbers, testimonials, case studies]
CTA: [specific desired action]
Format: [landing page / email / product card]

Structure:
1. Attention — a fact, statistic, or provocative statement that stops the scroll. Not a question — a statement.
2. Interest — provide context. Why this matters right now. Tie it to a trend, deadline, or market shift.
3. Desire — show the outcome through a concrete usage scenario. Numbers, timelines, before/after comparisons.
4. Action — one action, one button, one reason to do it now (no artificial pressure).

Tone: professional, no hype. Every sentence should deliver new information.

Example: project management SaaS

Before (no framework):

Manage projects efficiently with ProjectFlow. Kanban boards, Gantt charts, time tracking, and integrations with 50+ services. Try free for 14 days.

After (AIDA):

Project managers spend more than half their working hours on coordination, not execution (PMI, 2024). [Attention]

With distributed teams on the rise, the problem has gotten worse. When your team spans four time zones, status meetings don’t scale. You need a system where progress is visible without calls. [Interest]

Bookmate moved 12 projects to ProjectFlow and cut weekly status meetings from 6 hours to 40 minutes. All task communication moved into card context. After one month, coordination time dropped to 25%. [Desire]

Start with a free 14-day trial. Importing from Trello, Asana, or Jira takes 10 minutes. [Action]

Each block does one thing. Attention doesn’t sell. Interest doesn’t push to action. Desire works through someone else’s results, not a promise.

BAB: Before — After — Bridge

A contrast framework. Show the current situation, the desired situation, and the bridge between them (your product). It hits hardest when the audience knows their problem but can’t see a way out.

When to use: the product changes a process or habit. You need to show a transformation.

Prompt

You are a copywriter specializing in transformation marketing. Write copy using the BAB framework.

Product: [name and description]
Target audience: [who and in what situation]
Current audience situation (Before): [specific description of pain points and processes]
Desired situation (After): [what will change — measurably]
Format: [email / post / landing page]

Structure:
1. Before — describe the audience's current reality. Everyday details that trigger recognition. Don't exaggerate, don't dramatize.
2. After — show the alternative reality. Concrete changes in numbers and actions, not emotional descriptions.
3. Bridge — explain how the product moves them from Before to After. The mechanism, not the promise.

Rules: no "imagine," "dream of," or "forget your problems." Show, don't tell.

Example: email marketing automation service

Before (no framework):

Automate your email marketing with MailPulse. Build funnels, segment your list, track conversions. All the tools in one place.

After (BAB):

Right now: you’re sending a single blast to your entire list of 8,000 subscribers. Open rate has been dropping steadily: 22% six months ago, 17% now, 14% last month. Unsubscribes are climbing. You know you need segmentation, but setting it up manually takes 4–5 hours per campaign.

In 30 days: your list is split into 6 behavioral segments automatically. Each segment gets relevant content. Open rate is back to 24%. Campaign prep time is down to 40 minutes because content is matched to each segment’s template.

How it works: MailPulse analyzes each subscriber’s interaction history and assigns a behavioral tag. The first automatic segmentation happens within 48 hours of connecting your list. From there, the system suggests a content template for each segment based on your past highest-engagement campaigns.

BAB works because the reader sees themselves in “Before” and wants to be in “After.” The bridge makes the transition concrete and believable.

FAB: Features — Advantages — Benefits

A technical framework for products where specs matter. It translates features into advantages, and advantages into user benefits. A must-have for B2B, technical products, and comparative marketing.

When to use: the audience is knowledgeable and actively comparing options.

Prompt

You are a B2B copywriter with a technical background. Write copy using the FAB framework.

Product: [name and description]
Target audience: [role, level of technical expertise]
3–5 key product features: [list with technical details]
Competitors: [main alternatives]
Format: [website description / commercial proposal / product card]

Structure (repeat for each feature):
1. Feature — technical specification. Exact numbers, specs.
2. Advantage — why this is better than alternatives or the current approach. A comparison.
3. Benefit — what this concretely gives the user in their work. Time, money, outcome.

Output format: table or structured list. Not running prose.

Example: CDN provider

Before (no framework):

Fast CDN for your website. Servers worldwide, DDoS protection, simple setup. Speed up your page load times.

After (FAB):

Feature	Advantage	Benefit
380 PoPs across 102 countries	Twice as many PoPs as most Fastly- and Akamai-tier competitors	Load times for Southeast Asian and African users drop from 3.2s to 0.8s
HTTP/3 + Brotli compression enabled by default	Most CDNs require manual HTTP/3 activation	15–20% less traffic without any code changes. Bandwidth savings at 1M+ pageviews: $200–400/month
DDoS mitigation up to 15 Tbps	Covers the largest known attacks without needing an Enterprise plan	No extra charge for advanced protection
Wildcard SSL with auto-renewal	Many CDNs cap SSL certificates per plan	One config covers all subdomains. Zero manual certificate renewal

FAB works best when the buyer is already comparing options. They’re not looking for inspiration — they need arguments.

4U: Urgent — Unique — Ultra-specific — Useful

A framework for headlines and short formats. Four criteria every piece of copy must pass. Use it for email subject lines, ad headlines, and landing page headers.

When to use: you need to pack maximum conversion power into 5–15 words.

Prompt

You are a headline and microcopy specialist. Generate 10 headline variants using the 4U framework.

Product: [name]
Offer: [what's being offered]
Target audience: [who]
Placement context: [email subject / ad headline / landing page header]

Each headline must pass 4 criteria:
1. Urgent — a reason to act now (deadline, limit, changing conditions). No fake urgency.
2. Unique — a differentiator from competing offers. What exists only here.
3. Ultra-specific — concrete numbers: timeframes, percentages, dollar amounts. No "fast," "a lot," "significantly."
4. Useful — a clear benefit in a single action.

Format: table with four columns (U-U-U-U) + final headline.
Constraint: no more than 70 characters per headline.

Example: targeted advertising webinar

Before (no framework):

Free webinar: how to set up targeted advertising

After (4U), 5 variants:

1. Cut your cost per lead by 40% in 3 days — a breakdown of Meta Ads settings (recording until Friday)
2. 3 Meta Ads campaigns delivering leads at $18 — live breakdown on March 27 at 7 PM ET
3. $18 CPL instead of $60: a Meta Ads funnel for info products (80 spots left)
4. Setting up lookalike audiences in Meta Ads in 15 minutes — step-by-step walkthrough with data
5. From $60 to $18 per lead: 3 Meta Ads changes tested on a list of 12,000

Every variant has a specific number, a time frame, and a clear action. None contains an empty promise.

How to choose the right framework

The choice depends on three things: audience awareness, format length, and product type.

Situation	Framework	Reason
Audience knows the problem, looking for a solution	PAS	Direct path: pain → solution
Audience doesn’t recognize the problem	AIDA	Need an “attention hook” before the sale
Need to show a transformation	BAB	Before/after contrast motivates more than arguments
Technical product, comparison stage	FAB	Buyers need specs, not emotions
Headline or email subject	4U	Maximum conversion in minimum words

In practice, you combine them. A 4U headline, a PAS body, an AIDA landing page. AI handles composite prompts well when the structure is spelled out.

Prompt engineering for ad copywriting

The framework gives you structure. But the quality of what you get back depends on the context you feed the model.

Set constraints, not freedom. “Write copy” gets you mediocre output. “Write copy in 150 words, no superlatives, with one number in each paragraph” gets you something you can actually work with. More on how prompt structure affects output is covered in the context engineering guide.

Provide tone examples. Instead of “write in a professional tone,” show the model 2–3 paragraphs of text whose tone you like. Few-shot examples work better than abstract descriptions. For more on structuring prompts systematically, see the prompt engineering system.

Specify anti-patterns. Explicitly banning specific phrases (“no words like ‘unique,’ ‘innovative,’ ‘best in class’”) removes marketing noise better than positive instructions.

Iterate through critique. Generate the copy, then in the next prompt: “Evaluate this text on three criteria: specificity (are there numbers?), recognizability (will the reader see themselves?), actionability (is it clear what to do?). Rewrite the weak parts.” This mirrors the LLM-as-judge pattern, where the model generates first, then evaluates its own output.

Adapting to formats and platforms

One framework, multiple formats. You need to adapt the prompt to each platform’s constraints.

Search advertising (Google/Microsoft Ads). Headline up to 30 characters, description up to 90. Use 4U for the headline, the first sentence of PAS for the description. Specify exact character limits in the prompt.

Email newsletters. Subject line by 4U (under 50 characters for mobile), first paragraph by PAS or BAB. Generate the preheader (preview text) in a separate prompt with a 40–70 character limit.

Social media. The first 2 lines decide whether the reader expands the post. A PAS opening works best: problem in line one, agitation in line two. AIDA for the rest.

Landing pages. Headline by 4U, first section by PAS or BAB, benefits block by FAB, overall structure by AIDA. Generate each block in a separate prompt with context from previous blocks.

Common mistakes in AI-generated ad copy

Generating without audience data. A prompt without a target audience description produces copy “for everyone” — which means for no one. Minimum: role, problem, level of product awareness. If you haven’t defined your ICP yet, the ICP definition with AI guide walks through that process.

Too many promises. AI defaults to being persuasive and overloads copy with benefits. Constrain it: “one main promise, one piece of proof, one action.”

No specificity. “Increase your sales” vs. “Increase average order value by 23% in the first month.” If you don’t have real numbers, say so in the prompt: “Use plausible but conservative figures and label them as examples.”

Uniform output. Generating 10 variants in one prompt produces 10 similar texts. Generate 2–3 at a time, switching frameworks between runs. A PAS variant, an AIDA variant, a BAB variant of the same offer — that gives you real diversity for testing.

Ignoring funnel stage. Copy for a cold audience and for someone who’s already visited your site needs different frameworks. Top of funnel: AIDA (capture attention). Middle: BAB (show the transformation). Bottom: PAS (remove the final objection) or FAB (give them closing arguments).

Where these frameworks fall short

Frameworks give structure, but they don’t replace market knowledge. A perfectly structured PAS ad still fails if the pain point is wrong or the audience segment is too broad.

AI-generated copy also tends to sound polished but safe. It rarely takes a bold stance or uses the kind of odd, specific language that makes real ads memorable. You’ll still need a human pass to add edge — the weird metaphor, the sentence fragment, the line that makes someone stop scrolling.

Finally, these frameworks assume a single product with a clear value proposition. Complex B2B sales with multiple stakeholders, long cycles, and regulatory constraints don’t map neatly to PAS or AIDA. In those cases, frameworks are a starting point for individual touchpoints, not a full messaging strategy.

Where to start

You don’t need all five frameworks at once. Here’s a practical plan for your first week.

Days 1–2: PAS. Pick one product or service. Write a prompt using the template from this article. Generate 3 variants. Choose the best one, refine it by hand. The key test: does the reader recognize their own situation in the first sentence?

Days 3–4: 4U for headlines. Take 5 existing ad headlines from your campaigns. Rewrite each using the 4U prompt. Compare the CTR of old vs. new variants.

Days 5–7: AIDA for longer formats. Write an email or product description using AIDA. Read only the Attention block. If it doesn’t make you want to read further, rework it.

Each framework gets stronger when the model gets enough context. The process of repurposing content across different formats and platforms is covered in the content repurposing formula article.

After the first week, add BAB and FAB as you need them. Save working prompts to a library. Within a month, you’ll have a set of tested templates for every type of ad task.

FAQ

Which AI copywriting framework converts best?

PAS (Problem — Agitate — Solution) consistently performs well for short-format ads because it mirrors how people make decisions: recognize the pain, feel its weight, grab the fix. But “best” depends on your audience’s awareness level and the ad format. Test PAS against AIDA or BAB on the same offer to find what works for your specific case.

Can AI fully replace a human copywriter?

Not yet. AI with a framework produces solid first drafts — often 70–80% of the way there. The remaining 20–30% is where a human adds brand voice, unexpected phrasing, and the kind of specificity that comes from actually knowing the product and customers. The real productivity gain is in speed: what took 2 hours now takes 20 minutes of generation plus editing.

How many ad variants should I generate per framework?

Two to three per prompt run. Generating more than that in a single prompt makes the outputs converge — the model starts repeating itself with minor variations. Switch frameworks between runs (a PAS version, then an AIDA version of the same offer) to get genuinely different angles worth A/B testing.

Do these frameworks work with any AI model?

Yes. PAS, AIDA, BAB, FAB, and 4U are structural instructions, not model-specific features. They work with GPT-4, Claude, Gemini, and open-source models like Llama. The quality difference between models shows up in tone and creativity, not in following the framework structure.