"Can I use ChatGPT for ad copy?"
The common question since 2023. Can general AI like ChatGPT, Claude, and Gemini replace human copywriters? Results from a 3-month informal experiment and the realistic answer.
Test structure
A: Human copywriter (5+ years experience) B: ChatGPT 4 (simple prompt) C: Claude 3 (detailed prompt + brand guide)
Three ecommerce accounts, 5 ads each, measured 2 weeks per variant.
Results (typical pattern)
| Metric | Human | ChatGPT (simple) | Claude (detailed) |
|---|---|---|---|
| CTR | 1.8% | 1.2% | 1.6% |
| CPA | $22 | $31 | $24 |
| Ad Quality | 6.5/10 | 4.8/10 | 6.1/10 |
Takeaways:
- Simple-prompt AI drops performance significantly ("write me some ad copy" level)
- Detailed prompt + brand guide AI gets close to human
- Humans still lead in the fine details
AI copy's strengths
1. Speed
- 20 variations in 5 minutes
- Humans produce 3–4 in the same time
2. Variety
- Easy tone, length, and structure variations
- Large volume of A/B test material
3. Draft quality
- More efficient to refine an AI draft than start from scratch
- Human time focuses on fine edits
AI copy's weaknesses
1. Brand voice
- Can't lock in "our brand tone"
- Even with a brand guide injected, nuance gets diluted
2. Policy violation risk
- Banned expressions like "100% effective" and "guaranteed" auto-generated
- Meta ad review rejection rate is 20–30% higher on AI copy
3. Cultural context
- Weak on local humor and consumer psychology
- Translation tone → awkward phrasing common
4. Number and fact hallucinations
- Numbers like "40% off" or "1,000 users" get auto-generated
- Can turn into false claims if the advertiser doesn't verify
How to use it in practice
Use 1: Draft generation (strongly recommended)
- AI generates 10 hooks
- Human picks and refines 2–3
- 70% time savings
Use 2: Variation mass production (recommended)
- Feed one winning copy to AI for 10 variations
- Build A/B test material fast
Use 3: Fully automated (not recommended)
- Ship AI copy as-is
- Policy violation + brand damage risk
Prompt quality is everything
Bad prompt example:
"Write 5 ad copy for our company"
Result: generic, clichéd, possibly including policy violations.
Good prompt example:
"Our brand is premium skincare for female professionals in their 30s. Tone: professional yet approachable. Key values: science-based, non-irritating. Write 5 ad copy options for the following product. Constraints: no '100%', 'perfect', or 'guaranteed' expressions. Length: 40 characters max. Structure: problem → solution → CTA."
Result: significantly more usable drafts.
Meta built-in vs. external AI
| Use case | Meta Advantage+ Creative | External AI (ChatGPT, Claude) |
|---|---|---|
| Copy variations | Good (ads-domain trained) | Good (general LLM) |
| Image generation | Stable Diffusion based | DALL·E, Midjourney |
| Brand tone preservation | Weak (auto-variation) | Depends on prompt quality |
| Policy compliance | Strong (Meta review aware) | Weak (external doesn't know the criteria) |
| Speed | Instant (inside Ads Manager) | Requires separate tools |
| Cost | Free (part of ad plan) | API/paid subscription |
Conclusion: Parallel use is optimal. Advantage+ Creative for auto-variation + external AI for brand-specific copy.
So what do we do?
Beginner advertisers:
- Advantage+ Creative first (Meta built-in, free)
- Add external AI only when brand identity really matters
Intermediate:
- External AI for 10 drafts → human picks 2–3
- Advantage+ Creative for variation generation
- Document your brand guide prompt
Advanced:
- Automate the AI copy pipeline (Make, Zapier, etc. for prompt management)
- Verification loop: AI generate → policy check → A/B test
Longer outlook
AI copy quality is improving fast. The human copywriter's role shifts to:
- Strategy (what to say)
- Brand definition (how to say it)
- Final verification (check before ship)
Actual typing and variation work moves to AI-driven. Should standardize within 2–3 years.
Creative planning and iterative AI creative production are covered in Meta Ads Book 3.