Prompt Engineering for Marketing Copy: AIDA, PAS, and Beyond

Marketing prompts that say write compelling copy produce compelling nothing. Frameworks give the model a skeleton — Attention before Action, Problem before Solution — so output follows persuasion logic instead of adjective soup.

This guide maps classic copy frameworks to prompt templates for 2026 models. Use GPT-5.5 when you need strict section headers; Claude Opus 4.8 when voice and nuance matter; Gemini 3.5 Flash for bulk ad variants.

Always provide: audience, product, one proof point, tone, word limit, CTA. Framework name in the prompt is non-negotiable.

Before vs after: vibes vs framework

Weak prompt:

Write compelling landing page copy for our project management SaaS. Make it exciting.

Typical output: generic superlatives (revolutionary, game-changing), invented testimonial, feature list in random order, weak CTA (Learn more), 400 words with no structure.

Strong prompt:

ROLE: Direct-response copywriter.

FRAMEWORK: AIDA — label each section.

PRODUCT: FlowStack — cuts sprint planning from 2 hours to 35 minutes

AUDIENCE: Engineering managers at 50–200 person SaaS companies

PROOF: Meridian Pay case study — 35-min planning (verified)

OUTPUT: Attention headline + subhead | Interest 2 sentences | Desire 3 bullets | Action CTA + support line

TONE: Confident, peer-level

CONSTRAINTS: No invented testimonials. No fake urgency. Max 180 words.

Expected output: labeled AIDA sections, one proof source, single primary CTA, defensible claims only.

Why frameworks beat vibes

Frameworks constrain structure. The model fills slots with your product facts instead of inventing a random flow.

Label output sections in the prompt AND require labels in output — GPT-5.5 follows labeled schemas reliably.

Without framework: model optimizes for fluent adjectives. With framework: model optimizes for slot-filling with your facts.

AIDA (Attention, Interest, Desire, Action)

Best for: landing page hero + above fold, short sales pages.

Prompt:

ROLE: Direct-response copywriter.

FRAMEWORK: AIDA — label each section.

PRODUCT: [name + one-line value prop]

AUDIENCE: [who + pain]

PROOF: [stat, testimonial quote, or award — real only]

OUTPUT:

• Attention: headline + subhead (max 20 words headline)
• Interest: 2 sentences expanding the problem
• Desire: 3 bullets of outcomes
• Action: CTA button text + supporting line

TONE: [confident / friendly / technical]

CONSTRAINTS: No invented testimonials. No fake urgency.

Worked AIDA example (filled):

Attention: Stop losing Tuesday to sprint planning | Subhead: FlowStack gets your team from backlog to commit in 35 minutes

Interest: Two-hour planning meetings mean less shipping and burned-out leads. Your engineers feel it every sprint.

Desire bullets: Cut planning time 70% | Async prep replaces live debate | Integrates with Jira and Linear

Action: Start free trial — no credit card | Join 400+ eng teams this quarter

All proof marked [VERIFY] if you have not confirmed the 400+ stat before publish.

PAS (Problem, Agitate, Solution)

Best for: pain-aware audiences, email campaigns, ad copy.

Prompt:

FRAMEWORK: PAS

PROBLEM: [specific pain in customer words]

AGITATE: What happens if unresolved — 2 sentences, no fear-mongering

SOLUTION: [product] as mechanism — how it fixes, not feature list

FORMAT: Under 200 words. End with CTA: [action]

AUDIENCE: [segment]

Agitate means consequences, not insulting the reader.

Worked PAS example (filled):

Problem: Your attribution reports still live in three spreadsheets merged every Monday morning.

Agitate: Every hour on manual merges is an hour not fixing campaigns — and one wrong VLOOKUP sends budget to the wrong channel.

Solution: Acme Analytics pulls ad platform + CRM data into one dashboard. Monday reports generate in 45 minutes, not 6 hours.

CTA: See your data unified — book a 15-min demo

Agitate names cost of inaction without threatening the reader.

BAB (Before, After, Bridge)

Best for: transformation stories, case study intros, SaaS onboarding emails.

Prompt:

FRAMEWORK: BAB

BEFORE: Current state — [specific frustration]

AFTER: Desired state — [measurable outcome]

BRIDGE: [product] — 2 sentences how it connects before to after

OUTPUT: 3 labeled paragraphs. Under 150 words total.

Example: Before — support queue triaged in Slack threads. After — median first response under 2 hours. Bridge — TicketFlow routes by priority and SLA automatically.

4Ps (Promise, Picture, Proof, Push)

Best for: long-form sales sections, webinar scripts.

Prompt:

FRAMEWORK: 4Ps — separate labeled sections.

PROMISE: One bold outcome claim (must be defensible)

PICTURE: Sensory description of life after — 3 sentences

PROOF: [data, logos, quote — provided facts only]

PUSH: CTA + risk reversal (guarantee, trial, demo)

Max 350 words.

Use when you have strong proof assets — weak proof makes Promise section dangerous.

FAB (Features, Advantages, Benefits)

Best for: product pages, comparison tables, sales enablement.

Prompt:

Input features: [list]

For each feature output table row: Feature | Advantage | Benefit to [audience]

Benefit must answer so what for the buyer. No jargon in benefit column.

Example row: Real-time sync | No stale dashboard data | Catch budget overspend same day, not next week

Framework picker

| Goal | Framework |

| Hero + CTA fast | AIDA |

| Pain-led email | PAS |

| Transformation narrative | BAB |

| Long sales section | 4Ps |

| Feature list → buyer value | FAB |

| Retargeting ad (pain reminder) | PAS compressed |

| Onboarding email 1 | BAB |

Channel-specific: landing page hero

FRAMEWORK: AIDA

CHANNEL: Landing page above fold only

PRODUCT: [name + value prop]

AUDIENCE: [segment]

PROOF: [one stat — verified]

OUTPUT:

• Headline: max 10 words
• Subhead: max 25 words
• 3 benefit bullets: max 8 words each
• Primary CTA: 2–4 words
• Secondary CTA: optional, softer

CONSTRAINTS: No paragraph over 2 sentences above fold. Mobile-first line breaks.

GPT-5.5: strict word counts per element. Claude: smoother subhead rhythm.

Channel-specific: paid social (Meta/LinkedIn)

FRAMEWORK: PAS compressed

CHANNEL: Paid social — primary text under 125 characters, headline under 40

PROBLEM: [one line pain]

SOLUTION: [product mechanism in one line]

PROOF: [one stat or social proof fragment]

OUTPUT:

• Primary text (PAS in 125 chars)
• Headline (40 chars)
• Description (optional, 30 chars)
• CTA button label from platform list

Generate 3 variants: pain-led, outcome-led, proof-led. Same facts.

Gemini 3.5 Flash: batch 10 variants from one FAB table for ad set testing.

Channel-specific: Google Search RSA

FRAMEWORK: AIDA micro — headline = Attention, description = Interest + Action

CHANNEL: Responsive search ads

OUTPUT:

• 15 headlines: max 30 characters each, include 3 with keyword [keyword]
• 4 descriptions: max 90 characters each
• Pinning suggestion: which headline for H1 vs H2

CONSTRAINTS: No duplicate headlines. No unverified superlatives. Include keyword in exactly 3 headlines.

GPT-5.5 excels at character-limit counting for RSA.

Channel-specific: email campaign body

FRAMEWORK: PAS or BAB depending on list warmth

CHANNEL: Marketing email

AUDIENCE: [cold / warm / customer]

If cold: PAS. If warm/customer: BAB.

FORMAT: Subject line + preview text (under 90 chars) + body under 250 words

OUTPUT: Label framework sections in body as comments for reviewer

CTA: One primary link action

Add: Preheader complements subject, does not repeat it.

Channel-specific: product detail page (mid-fold)

FRAMEWORK: FAB table + 4Ps Push for bottom CTA block

OUTPUT:

1. FAB markdown table (5–8 rows)
2. 4Ps Push: guarantee + CTA paragraph under 80 words

Audience: consideration stage — benefits over hype.

Weak vs strong: PAS agitate

Weak agitate prompt:

Make them feel terrible about their current situation.

Output: insulting copy, fear-mongering, unsubstantiated disaster scenarios.

Strong agitate instruction:

AGITATE: Name 2 concrete business consequences of the problem — time cost, error cost, or opportunity cost. No insults. No health or safety scare unless product is in that domain.

Output: quantified time waste and one operational risk — reader stays respected.

Variant generation prompt

Using [framework] output above:

Generate 3 headline variants: (1) outcome-led, (2) curiosity gap, (3) social proof led.

Keep body structure. Change headline/subhead only.

For ads add: Generate 3 primary texts under [N] characters — same framework, same proof.

Voice and compliance guardrails

Add to any framework prompt:

BANNED: superlatives without proof, fake countdown, medical/financial guarantees

REQUIRED: CTA matches funnel stage [awareness/consideration/decision]

YMYL: if health/finance — no outcome promises; use qualifies language

Mark any unverified stat with [VERIFY] in output — human must confirm before publish.

Funnel stage × framework matrix

| Stage | Best framework | CTA type |

| Awareness | BAB, curiosity AIDA headline | Learn more, read guide |

| Consideration | PAS, FAB | Compare, watch demo |

| Decision | 4Ps Push, AIDA Action | Start trial, buy, book |

| Retention | BAB (new feature) | Upgrade, explore feature |

Mismatch — decision CTA on awareness copy — kills conversion and triggers ad rejection.

Common mistakes

Framework name omitted — model drifts to generic essay

No word limit — hero copy becomes 500-word manifesto

Proof left blank — model invents testimonials and stats

Agitate instruction vague — output shames the reader

Same prompt for LinkedIn and landing page — channel limits ignored

Feature list in Solution slot — mechanism missing

Multiple CTAs — reader paralysis

Model notes

GPT-5.5: strict section labels, ad character limits, RSA headline counts

Claude Opus 4.8: better first-draft human voice for PAS agitate and BAB Picture sections

Gemini 3.5 Flash: bulk ad variants from one FAB table

Gemini 3.1 Pro: long 4Ps sales pages with multiple proof blocks — keep context of all provided facts

PromptMake workflow

/text → pick framework + channel + paste product facts → Improve tightens CTA → human verifies claims.

Run variant prompt as second pass — never merge variant generation into first prompt or structure loosens.

FAQ

AIDA or PAS for landing page hero?

AIDA for above-fold hero. PAS for email driving to that page or retargeting ads.

How do I prevent fake testimonials?

CONSTRAINTS: Use ONLY proof provided below. If no testimonial provided, omit testimonial section — do not invent.

Can one prompt do all channels?

No — character limits and structure differ. Use channel-specific templates above; share PRODUCT and PROOF blocks only.

How many variants for ad testing?

Three minimum per angle (pain, outcome, proof). Gemini 3.5 Flash batch-generates; human removes any [VERIFY] lines before launch.

B2B vs B2C tone in frameworks?

Same frameworks — change AUDIENCE and TONE. B2B: peer-level, metric-led Desire. B2C: sensory Picture in 4Ps, emotional PAS Problem.

What about SEO blog copy?

Frameworks apply to intro and CTA blocks — not full article. See Perfect SEO content formula for long-form.

Regulated industries?

Add YMYL block. Require legal review flag on output. Ban guarantee language unless PROOF includes approved disclaimer text.

Worked before/after: RSA headlines

Weak:

Write Google ad headlines for our analytics tool.

Output: 15 headlines over 30 characters, duplicate angles, unverified Best-in-class claim.

Strong (channel-specific RSA template filled):

FRAMEWORK: AIDA micro. CHANNEL: RSA. KEYWORD: marketing attribution

OUTPUT: 15 headlines max 30 chars, 4 descriptions max 90 chars

CONSTRAINTS: keyword in exactly 3 headlines. No unverified superlatives.

Output: 15 compliant headlines, 3 contain attribution keyword, descriptions fit limits, [VERIFY] on any stat.

Related articles

Email prompt templates — 10 RTF examples

Perfect SEO content formula — long-form content prompts

Positive framing — banned words as positive constraints

Claude vs ChatGPT vs Gemini — voice vs structure routing

Bottom line

Name the framework in the prompt. Label output sections. Supply real proof or mark [VERIFY]. Match framework to channel and funnel stage. AIDA, PAS, BAB, 4Ps, FAB turn write copy into fill-in-the-blanks persuasion — when facts come from you, not the model.