Ad Copy Generator

Overview

Write Meta ad copy that’s matched to the actual image creative — not generic copy pasted across every ad. Each image gets copy that reinforces its specific message, written in the brand’s voice, informed by what’s already working in the account. The critical rule: Copy reinforces the image, never repeats it. If the image shows “$17K/month”, the body tells the story behind that number. If the image shows a stressed host, the copy describes the chaos. Image and copy are two halves of one message.

Brand Memory Integration

This skill reads brand context to make every piece of copy sound like the brand, not a template.

Reads

optional

Match tone, vocabulary, forbidden words. Show: “Voice loaded — [tone summary].”

optional

Use angle as copy’s foundation. Show: “Positioning: [angle].”

optional

Know who we’re talking to, their awareness level, their language. Show: “Writing for [audience summary].”

If workspace/brand/ doesn’t exist, the skill works standalone. It will ask for ICP, voice notes, and forbidden words inline.

The Process

Step 1: Pull What’s Already Working

Before writing a single word, look at the account. What copy is converting?

# Top performers by CTR — last 30 days
curl -s "https://graph.facebook.com/v22.0/ACT_ID/insights?\
level=ad&fields=ad_name,impressions,clicks,ctr,cpc,cost_per_action_type\
&date_preset=last_30d&sort=ctr_descending&limit=20\
&access_token=$FACEBOOK_ACCESS_TOKEN"

Then pull copy from winners:

# Get creative ID from the ad
curl -s "https://graph.facebook.com/v22.0/AD_ID?fields=creative{id}&access_token=$FACEBOOK_ACCESS_TOKEN"

# Get the actual copy
curl -s "https://graph.facebook.com/v22.0/CREATIVE_ID?fields=asset_feed_spec&access_token=$FACEBOOK_ACCESS_TOKEN"

Extract patterns:

Headline length and structure (questions? numbers? commands?)
Body copy hooks (opens with pain? data? question? story?)
Social proof claims (X customers, Y% increase, Z revenue)
Reading level (short punchy vs. narrative)
CTA that converts (Learn More vs Get Started vs Book Demo)

Show the user what you found: “Your top 3 ads all open with a specific number and close with social proof. Average headline: 32 chars. I’ll match that pattern.”

Step 2: Load Brand Context

Need	Where	Fallback
ICP	`audience.md` or CLAUDE.md	Ask: “Who’s this for?”
Voice	`voice-profile.md`	Ask: “Any words to avoid? Tone preference?”
Pain points	`audience.md`	Extract from top-performing copy
Key stats	Brand brief / CLAUDE.md	Ask: “What proof points can I use?”
Forbidden words	`voice-profile.md`	Default ban list (see below)

Default forbidden words (always banned unless brand explicitly uses them): revolutionary, game-changer, cutting-edge, innovative solution, transform your business, unlock your potential, unleash, harness the power, leverage (as verb), next-level, best-in-class, world-class, seamless, robust, synergy

Step 3: Analyze Each Image Creative

For every image that needs copy, use vision to identify:

Visual format — Notes app, receipt, tweet, chart, kitchen photo, poster, review, slack, etc.
On-image text — What copy is already baked into the image?
Angle/hook — What psychology is the image pulling?
Mood — Organic/found-footage, professional, urgent, casual, dark, bright
Funnel stage — Awareness (problem), consideration (solution), conversion (offer)

Step 4: Write Copy Using Psychology

Every ad pulls at least one psychological lever. Rotate across variants so Meta can test.

The Four Horsemen (Kallaway Framework)

Horseman	Trigger	Opening Pattern
Money	Revenue loss, ROI, cost savings	”Every [thing] you miss is worth $X.”
Time	Wasted hours, efficiency, speed	”Your [person] isn’t a [wrong role].”
Status	Peer comparison, industry moves	”X+ [peers] already [did thing].”
Fear	Loss aversion, competitive threat	”The [competitor] down the street already has this.”

Matching Psychology to Image Format

Image Format	Best Psychology	Copy Tone
Notes app / screenshot	Fear + Money	Casual first-person discovery
Fake tweet / social	Status + Fear	Conversational, peer voice
Data visualization	Money + Status	Authority, numbers-forward
Receipt / POS	Money + Time	Specific, transactional
Kitchen / still life	Fear + Time	Emotional, atmospheric
Review / testimonial	Status + Money	Third-person proof
Slack / workplace	Time + Status	Insider, operational
News article	Fear + Status	Urgent, newsjacking
Before/after	Money + Time	Contrast, transformation
Bold text / poster	Any	Direct, declarative
Google review	Status + Money	Social proof, customer voice
Whiteboard / letter board	Time + Fear	Physical, tangible
Lock screen / notifications	Fear + Time	Urgency, real-time

Step 5: Apply Copy Specs

These specs come from analyzing hundreds of winning Meta ads:

Element	Spec	Why
Headline length	25-40 chars, never >50	Truncation on mobile kills CTR
Body word count	50-120 words	Short enough to read, long enough to convince
Paragraphs	2-3, each 1-2 sentences	Mobile readability — walls of text = scroll past
Numbers required	≥1 per variant ($, %, or count)	Specificity = credibility
Social proof	Include where natural	”X+ customers” is the highest-converting pattern
Opening line	Pain, stat, or bold claim	Never brand name, never “Introducing”
Closing line	Social proof + soft CTA	”X+ already made the switch.”
Line breaks	After every 1-2 sentences	Scannability on mobile
No em dashes	Use hyphens or restructure	Rendering inconsistency across placements
No emojis in body	Unless brand voice uses them	Most B2B/professional ads don’t

Step 6: Generate Degrees of Freedom Variants

Meta’s algorithm tests combinations. Give it options: Per creative:

3-5 headline variants (different angles)
3-5 body copy variants (different psychology)
1-2 description variants (supporting stat or benefit)

Rules for variant quality:

Each variant MUST hit a different Horseman. V1=Fear, V2=Money, V3=Status, etc.
No two headlines should use the same structure. Mix questions, statements, numbers, commands.
No two body variants should open the same way.
At least one variant should feel like organic content (not an ad).

Step 7: Cross-Reference with Account Winners

Before finalizing, check your copy against Step 1 findings:

Does headline length match what’s working?
Are you using similar hook structures?
Does reading level match?
Are your social proof claims consistent with what the account already says?

Flag any mismatches: “Your top performers use question headlines. 2 of my 3 headlines are questions — aligned.”

Funnel Stage Adjustments

Stage	Objective	Copy Focus	CTA	Proof Level
ToFu	Awareness / Traffic	Problem awareness, curiosity	Learn More	Light — “X+ use this”
MoFu	Consideration / Leads	Solution proof, specific features	Get Free Audit / See Demo	Medium — case study snippets
BoFu	Conversion / Sales	Urgency, risk reversal, pricing	Start Free Trial / Book Demo	Heavy — ROI numbers, testimonials

Adjust body copy density:

ToFu: 50-80 words (fast read, curiosity-driven)
MoFu: 80-120 words (more detail, proof points)
BoFu: 60-100 words (tight, urgent, specific offer)

Output Format

For each creative:

## [Creative Name]
*Image: [brief description of what the image shows]*
*Format: [notes app / tweet / receipt / etc.]*
*Psychology: [primary Horseman] + [secondary]*
*Funnel: [ToFu / MoFu / BoFu]*
*Matched to image: [one line on how copy reinforces the visual]*

### Headlines
1. [headline] ([XX] chars) — [Horseman]
2. [headline] ([XX] chars) — [Horseman]
3. [headline] ([XX] chars) — [Horseman]

### Body Copy

**V1 — [Horseman]: [Angle name]**
[body text]
*[word count] words*

**V2 — [Horseman]: [Angle name]**
[body text]
*[word count] words*

**V3 — [Horseman]: [Angle name]**
[body text]
*[word count] words*

### Descriptions
1. [description]
2. [description]

asset_feed_spec Output

The skill outputs copy in Meta’s Graph API format, ready for upload:

{
  "bodies": [
    {"text": "Your V1 body copy here. Pain point opener, specific stat, soft CTA."},
    {"text": "Your V2 body copy here. Different angle, different psychology."},
    {"text": "Your V3 body copy here. Social proof heavy. Numbers up front."}
  ],
  "titles": [
    {"text": "Headline One - 35 Chars"},
    {"text": "Question Headline?"},
    {"text": "Stat-Driven: 3X Results"}
  ],
  "descriptions": [
    {"text": "Supporting benefit statement."},
    {"text": "Secondary proof point."}
  ]
}

This JSON format is required by Meta’s Graph API asset_feed_spec field when creating creatives. It enables Degrees of Freedom optimization where Meta tests all combinations.

Campaign File Output

After generation, save to campaign directory:

workspace/campaigns/{campaign-name}/ads/
  {creative-name}.md          <- Full copy document
  {creative-name}.json        <- asset_feed_spec JSON

Append to workspace/brand/assets.md:

| {creative-name} copy | ad-copy | YYYY-MM-DD | {campaign-name} | draft | 3 headlines, 3 bodies |

Graph API Integration

Fetch Top Performers

TOKEN="$FACEBOOK_ACCESS_TOKEN"
ACCOUNT="act_123456789"

# Get top ads by CTR
curl -s "https://graph.facebook.com/v22.0/$ACCOUNT/insights?\
level=ad\
&fields=ad_id,ad_name,impressions,clicks,ctr,cpc\
&date_preset=last_30d\
&sort=ctr_descending\
&limit=20\
&access_token=$TOKEN" | jq .

Extract Copy from Winning Ads

# Get creative from ad
AD_ID="23847111111111"
curl -s "https://graph.facebook.com/v22.0/$AD_ID?\
fields=creative{id,name,asset_feed_spec}\
&access_token=$TOKEN" | jq '.creative.asset_feed_spec'

Response:

{
  "bodies": [
    {"text": "Every reservation you miss is worth $47. That's the average..."},
    {"text": "Your host isn't a secretary. But they're spending 3 hours..."}
  ],
  "titles": [
    {"text": "$47 per missed call"},
    {"text": "Your host deserves better"}
  ]
}

Anti-Patterns

Avoid these common mistakes:

Reusing copy across creatives — Every image gets unique copy that matches IT
Same psychology across variants — V1/V2/V3 must hit different Horsemen
Starting with brand name — Nobody cares about your brand in ToFu
Generic SaaS language — See forbidden words list
Emoji bullet lists — Looks like every other ad on Meta
Ignoring account data — Always check what’s working first
Copy that repeats the image — Image says the number, copy tells the story
Long paragraphs — 3+ sentences = scroll past on mobile
Same headline structure — Mix questions, stats, commands, statements
Hard sell CTAs in ToFu — “Sign up now” in awareness = low CTR

Quick Mode

If the user provides images and says “write copy for these” without specifying a full workflow:

Skip account data pull (or use cached if available)
Load brand context if it exists
Analyze images → write matched copy → output in standard format
Still suggest checking against account data as a next step

Integration with Other Skills

Workflow

ad-creative-monitor → Detects "Notes App Hero" fatigued (CTR dropped 40%)
ad-copy-generator   → Write 3 new body variants matched to the same image
ad-upload           → Push fresh creative, attach to existing ad
meta-ads            → Monitor new creative performance next 7 days

Data Flow

# What's already working?
curl "https://graph.facebook.com/v22.0/$ACCOUNT/insights?..."
# → Extract: headline length, hook types, social proof format

{
  "bodies": [{"text": "..."}, {"text": "..."}, {"text": "..."}],
  "titles": [{"text": "..."}, {"text": "..."}, {"text": "..."}],
  "descriptions": [{"text": "..."}, {"text": "..."}]
}

# ad-upload skill uses this JSON
curl -X POST "https://graph.facebook.com/v22.0/$ACCOUNT/adcreatives" \
  -d @{creative-name}.json

Environment Variables

required

Facebook access token with ads_read scope for pulling account performance data

# Get token from social-cli config
export FACEBOOK_ACCESS_TOKEN=$(jq -r '.profiles.default.tokens.facebook' ~/.social-cli/config.json)

Next Steps

Ad Upload — Push generated copy to Meta via Graph API
Meta Ads — Monitor performance of new copy
Ad Creative Monitor — Detect when copy needs refreshing again

Get Started

Core Concepts

Skills

Configuration

Reports

Actions

Advanced

Overview

Brand Memory Integration

Reads

The Process

Step 1: Pull What’s Already Working

Step 2: Load Brand Context

Step 3: Analyze Each Image Creative

Step 4: Write Copy Using Psychology

The Four Horsemen (Kallaway Framework)

Matching Psychology to Image Format

Step 5: Apply Copy Specs

Step 6: Generate Degrees of Freedom Variants

Step 7: Cross-Reference with Account Winners

Funnel Stage Adjustments

Output Format

asset_feed_spec Output

Campaign File Output

Graph API Integration

Fetch Top Performers

Extract Copy from Winning Ads

Anti-Patterns

Quick Mode

Integration with Other Skills

Workflow

Data Flow

Environment Variables

Next Steps

​Overview

​Brand Memory Integration

​Reads

​The Process

​Step 1: Pull What’s Already Working

​Step 2: Load Brand Context

​Step 3: Analyze Each Image Creative

​Step 4: Write Copy Using Psychology

​The Four Horsemen (Kallaway Framework)

​Matching Psychology to Image Format

​Step 5: Apply Copy Specs

​Step 6: Generate Degrees of Freedom Variants

​Step 7: Cross-Reference with Account Winners

​Funnel Stage Adjustments

​Output Format

​asset_feed_spec Output

​Campaign File Output

​Graph API Integration

​Fetch Top Performers

​Extract Copy from Winning Ads

​Anti-Patterns

​Quick Mode

​Integration with Other Skills

​Workflow

​Data Flow

​Environment Variables

​Next Steps

Overview

Brand Memory Integration

Reads

The Process

Step 1: Pull What’s Already Working

Step 2: Load Brand Context

Step 3: Analyze Each Image Creative

Step 4: Write Copy Using Psychology

The Four Horsemen (Kallaway Framework)

Matching Psychology to Image Format

Step 5: Apply Copy Specs

Step 6: Generate Degrees of Freedom Variants

Step 7: Cross-Reference with Account Winners

Funnel Stage Adjustments

Output Format

asset_feed_spec Output

Campaign File Output

Graph API Integration

Fetch Top Performers

Extract Copy from Winning Ads

Anti-Patterns

Quick Mode

Integration with Other Skills

Workflow

Data Flow

Environment Variables

Next Steps