Facebook Ad Copy

---
name: facebook-ad-copy
version: 1.1.0
description: "Use this skill to generate Facebook and Instagram ad copy grounded in real audience insight. Triggers include: requests to create Facebook or Meta ad campaigns, write primary text and headlines for Facebook ads, refresh stale social ad creative, build ad sets for different audience segments, or generate audience-tested ad variations. Uses OriginalVoices Digital Twins (ask_twins) to deeply research the target audience's language, preferences, concerns, and experiences, then generates 10-15 audience-informed ad variations rooted in what real people actually said — not AI guesswork."
---

# Facebook Ad Copy Skill

## Overview

Generate Facebook and Instagram ad copy informed by real audience insight. Instead of relying on AI guesswork to write ads, this workflow first talks to your target audience to understand their language, preferences, concerns, and experiences — then uses those insights to create ads that speak to what real people actually care about.

## Workflow Steps

### Step 1: Gather Inputs

Only two things are required:

- **Target audience**: Who are these ads for? (e.g. "Women aged 25-40 in the US interested in skincare and wellness")
- **Product/concept**: Either a description in their own words, or a link to a product page / landing page / pitch deck

Optional:
- **Number of ad variations**: How many do they want? Default to 10-15 if not specified
- **Tone or brand voice preferences**: Any specific tone to match?

### Step 2: Audience Research

This is the most important step. Use `ask_twins` to understand the audience's world — their language, what they care about, what gets in their way, what grabs their attention, and how they relate to the problem space. Ask broad questions about the general topic area, not just about the specific product.

**Question quality — apply every rule to every question before sending.** The goal is the widest possible range of honest answers. A weak question is narrow, leading, or closed, so respondents converge on the same obvious reply and you learn nothing. Run each question through these rules:

1. **Anchor in a specific moment or behaviour, not an abstract definition.** Ask "Walk me through the last time…" or "Tell me about a time…", never "How do you define / think about [concept]?".
2. **Keep the valence open — never lead.** Don't presuppose the experience was good or bad. Banned phrasings: "what frustrates you most about…", "the time [X] annoyed you", "what do you love about this". Ask "how did it go?" or make it two-sided: "what worked well, and what didn't?".
3. **One idea per question.** Split anything double-barrelled into separate questions.
4. **Open-ended only — no yes/no, no multiple choice.** Every question invites How / What / Why / Walk me through / Tell me about. Never hand respondents a menu of options to pick from — let them supply their own.
5. **No socially-correct answers.** If almost everyone would answer the same way ("Is quality important?" → yes), reframe with a trade-off ("when is quality *not* worth it?").
6. **Ground intent in real behaviour.** Don't ask hypotheticals like "would you buy / click / switch?" — they inflate. Anchor in what they last actually did ("the last time you bought something like this, what tipped the decision?").
7. **Build in room to diverge.** Prefer a fork, trade-off, or spectrum ("what would make you — and what would hold you back?") over a flat single-answer prompt.

Before sending, check each question passes both gates — **Range:** could most people answer it the same way, or is there an obvious "right" answer? **Neutrality:** does it assume how they feel, or hand them the answer? If it fails either, rewrite it.

**Capturing the audience's own words.** This skill generates copy from what real people say, so the research must surface their *own language*, not abstract opinions — generic questions produce generic copy. Include at least two moment-anchored questions built to elicit verbatim phrasing (e.g. "The last time you looked for [X], what did you actually type or say?", "How would you describe [problem] in your own words, to a friend?"). When you later test copy, replace "would you click / stop scrolling?" (demand effect + inflated intent) with "what would you do next, if anything?" — and ask "what do you take this to be offering?" (comprehension) before whether they like it.

```
ask_twins(
  audience: "[detailed target audience description]",
  questions: [
    "Walk me through the last time [general topic area, e.g. skincare, fitness, cooking, managing finances] came up for you — what did you actually do, and how did it go?",
    "How would you describe [the problem/topic] in your own words — the way you'd explain it to a friend who was dealing with it too?",
    "Tell me about your most recent experience with a [product/solution in this space] — what worked well, and what didn't?",
    "When you think about trying a new [product in this category], what pulls you toward it — and what holds you back?",
    "Has a brand in this space ever earned or lost your trust? What did they do, and what would it take to earn it?",
    "Think about the last time you discovered a [product in this category] you decided to try — how did you come across it, and what made you give it a go?"
  ]
)
```

**Why this matters:** These questions surface the real language people use, the emotions they feel, the objections they hold, and the experiences that shape their decisions. This is what makes the ads resonate — not generic marketing copy.

### Step 3: Analyse Audience Insights

Review all responses and extract:

- **Pain points in their own words**: The exact language they use to describe frustrations
- **Preferences and priorities**: What they look for and value in this category
- **Experiences and context**: What they've tried before and how it went
- **Concerns and objections**: What would hold them back from trying something new
- **Trust signals**: What builds or breaks confidence
- **Emotional drivers**: The feelings that motivate action (relief, excitement, validation, fear of missing out)

### Step 4: Generate Ad Variations

Using the audience insights, generate 10-15 ad variations (or the number requested). Every ad should be directly traceable to something the audience actually said — a pain point, a preference, a phrase, an emotion.

**Facebook Ad Copy Structure:**
- **Primary text**: Main body copy (125 chars visible before "See more"; full text up to 1,000+)
- **Headline**: Appears below the image/video (up to 40 chars recommended)
- **Description**: Below the headline (up to 30 chars recommended)
- **CTA button**: Shop Now, Learn More, Sign Up, Get Offer, etc.

**Spread variations across different angles, drawing from the research:**
- **Pain point ads**: Open with a frustration the audience described, in their words
- **Benefit-first ads**: Lead with the outcome or feeling they said they want most
- **Social proof / trust ads**: Built around the trust signals and credibility markers they mentioned
- **Experience-based ads**: Reference the common experiences they shared (what they've tried, what failed)
- **Objection-handling ads**: Directly address a concern or hesitation from the research

For each ad, note which audience insight it's built on.

### Step 5: Validate with Audience (Optional but Recommended)

Test 3-5 of the strongest ads back with the audience.

```
ask_twins(
  audience: "[same target audience]",
  questions: [
    "You're scrolling Facebook and this ad goes past: '[primary text + headline]'. In your own words, what do you take it to be offering — and what would you do next, if anything?",
    "Here are three ways an ad could open. A: '[opening A]' B: '[opening B]' C: '[opening C]'. Which one sounds most like it's meant for you, and what makes the other two miss?",
    "Is there anything in these ads that doesn't ring true or feels off to you? What is it, and why?"
  ]
)
```

### Step 6: Deliver Final Ad Set

Format the output using proper markdown headers: use `##` for the report title, `###` for major sections (e.g. by angle: Pain Point Ads, Benefit-First Ads), `####` for individual ad variations, `>` blockquotes for Digital Twin quotes, and `---` horizontal rules between major sections. Never use plain text or bold-only text as section headers — always use proper markdown headers.

Present the full set of ad variations with:
- Each ad in full (primary text, headline, description, CTA)
- The audience insight driving each ad clearly labelled
- Ads grouped by angle (pain point, benefit, trust, experience, objection)
- Validation results (if done)
- Top 3-5 recommended ads to test first
- A/B testing suggestions (which ads to test against each other)

## Facebook Ad Copy Guidelines

| Element | Recommended | Max |
|---------|-------------|-----|
| Primary text | 125 chars (before truncation) | ~1,000+ chars |
| Headline | 27-40 chars | 255 chars |
| Description | 27-30 chars | 255 chars |

## Tips for Best Results

- **The research makes the ads**: The audience research step is what separates these ads from generic AI output. Don't rush it
- **Write how people talk**: If the audience says "I just want something that actually works", put that in the ad — not "Experience seamless efficacy"
- **Front-load the hook**: First 125 characters are all that shows before "See more"
- **One insight per ad**: Each ad should be built on one clear audience insight, not five crammed together
- **Include the "why" for each ad**: Labelling which insight drives each ad helps the user understand why it should work and makes it easier to iterate
- **Variety matters**: 15 ads that all say the same thing in different words aren't useful. Spread across genuinely different angles and motivations from the research

## Related Skills

- **google-ad-copy**: For Google Search RSA assets
- **creative-testing**: To test ad concepts before writing full copy
- **icp-discovery**: To find the right audience before writing ads
- **landing-page-optimization**: To optimise the page ads drive traffic to