Ad creative dies from the 3rd variant onward. The team starts paraphrasing instead of finding a genuinely new angle, so CTR plateaus and the algorithm punishes the whole ad set. These 12 prompts force the model to vary the angle (problem framing, persona, status, fear, time, money) before it varies the words. Pair them with the campaign slogan prompts for top-of-funnel headline work.
TL;DR
- Paste any prompt below into ChatGPT (GPT-5.5), Claude (Sonnet 4.6), or Gemini 3.1 Pro and replace the
[bracketed]placeholders. - Every prompt forces a stated angle or persona so the model can’t fall back to bland benefit-listing.
- Front-load the hook: Meta truncates Feed primary text after ~125 characters, and only ~1% of users tap “See more” (as of June 2026).
- If AI generated or substantially edited the image, voice, or video, Meta requires an “AI info” disclosure — undisclosed AI is now the third-largest rejection category. Plain AI-written copy does not need a label.
Which model to use
These are copywriting prompts, so any current flagship works. As of June 2026: GPT-5.5 (ChatGPT default) tends to write punchier hooks; Claude Sonnet 4.6 holds a consistent brand voice across a batch; Gemini 3.1 Pro is strong when you paste a long review or transcript to mine. For data-driven prompts (#7, #8, #12) the 1M-token context on Sonnet 4.6, Opus 4.7, and Gemini 3.1 Pro lets you drop in raw export files without trimming.
Best for
- Meta / Instagram ads (Feed, Reels, Stories)
- TikTok ads
- Google Search / Performance Max
- Programmatic display
- Retargeting and lookalike refresh
1. 10 hook lines for a video ad
Product: [description]. Audience: [persona]. Write 10 first-3-second hook lines for a video ad. Mix: pattern-interrupt, contrarian statement, ultra-specific number, direct question, gentle confession. For each, mark the hook type and which scroll-stopper instinct it triggers (curiosity, identity, surprise, fear, status).The first 3 seconds decide roughly 71% of whether a TikTok viewer keeps watching, so generating 10 distinct openers to test is the highest-leverage thing this whole list does.
2. 5 distinct-angle headlines
For [product], generate 5 ad headlines using DIFFERENT angles: time saved, money saved, social proof, fear avoided, status gained. Mark which angle for each. Then rank by likely CTR for a cold audience vs warm audience, with a 1-line reason per ranking.Keep the winning headline at or under ~27 characters for Meta Feed and 40 for the headline field; longer ones get clipped on mobile.
3. Primary text variants by length
For [product], write 3 primary text versions: punchy (≤90 chars, fully visible before Meta's "See more"), medium (≤150 chars), long (≤300 chars). Same offer, different fit by placement. Each version must start with a hook that stands alone if Meta truncates after ~125 characters. No emoji walls.The ~125-character truncation point is why the punchy version matters most: only about 1% of people expand “See more,” so anything past line two is effectively invisible in Feed (as of June 2026).
4. Carousel ad copy
For [product], write a 5-card carousel: each card has a headline + 30-word body. Cards form a narrative: problem → why now → solution intro → proof → CTA. The last card must give one concrete next step, not "learn more". Note the swipe motivation card-by-card.5. Retargeting copy by drop-off stage
Write 5 retarget ad copy variants for [product]. Segment by drop-off stage: visited home page, visited PDP, added to cart, started checkout, bounced from pricing. Each variant addresses the objection most likely at that exact stage — not a generic "come back".6. UGC script direction
Brief a creator to film a UGC ad for [product]. Write the brief: hook angle, 3 key beats to hit, 1 thing to avoid, ideal CTA, format (15s vertical). Include the b-roll list, the on-screen text overlay at each beat, and the caption hook the creator should post with the video.7. Ad copy from a review
Below is a glowing customer review. Extract 4 ad-ready phrases and write 2 ad copy variants using them with proper attribution. Mark which phrase carries the strongest emotional pull and which works best as the hook vs the proof line.
[paste review]8. Ad-fatigue refresh
My current ad has been running 4 weeks and CTR has dropped ~30%. Below is the copy + image description. Suggest 5 specific refresh variants without changing the core offer: change hook only, change image only, change CTA only, change format, change persona-pitch. Predict which is most likely to revive performance.
[paste copy and image description]9. Platform-native pacing rewrite
I have a Meta ad that performs but is being repurposed for TikTok. Rewrite it for TikTok-native pacing: first 1-second hook, no logo in the first 3 seconds, on-screen text that completes the spoken line, a swipe-friendly close. Keep the offer identical but rewrite the rhythm.
[paste current ad]Leading with a logo or product shot is the most common way to waste the first 3 seconds; lead with the problem, the benefit, or a surprising claim instead.
10. Persona-split variant batch
For [product], write 4 ad variants — each pitched to a different persona of the same buyer: the budget-conscious version, the time-poor version, the social-proof-driven version, the early-adopter version. Same offer, same product, completely different angle. Mark which persona each targets.11. Single-feature ad
My product has [N] features but ads that name all of them perform worst. Pick the 1 feature that creates the biggest "oh, I actually need that" reaction and write 3 ad copy variants built entirely around that single feature. Hide the other features for now.12. Creative from search intent
My top 5 Google Search queries for [product] are: [list]. Write 1 ad creative per query that matches the exact intent behind that search. Don't pitch the product; answer the question implied by the query and let the product appear as the obvious next step.Platform copy specs at a glance (June 2026)
| Field | Meta Feed | TikTok | Google Search |
|---|---|---|---|
| Hook / first line | front-load before ~125 chars | first 1-3 seconds | first headline |
| Headline | ~27 visible (40 max) | overlay text, keep short | 30 chars × up to 15 |
| Body / primary text | truncates ~125 chars | spoken + on-screen | 90 chars × 4 descriptions |
| ”See more” / expand rate | ~1% tap to expand | n/a | n/a |
Treat these as the visible-without-clicking thresholds, not hard caps. Meta lets you enter up to 2,200 characters of primary text; almost no one reads past the fold.
Common mistakes
- 10 variants all paraphrasing the same hook — vary the angle, not the words.
- No angle or persona stated in the prompt, so the model defaults to bland benefit-listing.
- Burying the hook past the ~125-character Meta truncation line where ~1% of users ever see it.
- Ignoring fatigue signals and rewriting copy when the image or pacing is the real problem.
- Repurposing a Meta ad to TikTok without adjusting first-second pacing.
- Trying to pitch every feature in one ad instead of leading with the single strongest.
- Forgetting Meta’s AI-content disclosure when the visual or audio was AI-generated or heavily edited — undisclosed AI is now a top rejection reason.
FAQ
Do I have to disclose that AI wrote my ad copy? No. As of June 2026, Meta’s “AI info” / “Made with AI” label applies to AI-generated or substantially modified images, audio, and video — not to text written with a chatbot. Standard edits like cropping and color correction are exempt. So a human-shot video with AI-written copy needs no label, but an AI-generated product image does.
How many variants should I generate per prompt? Generate 5-10, then ship 3-4 genuinely different angles into one ad set. More than that splits the budget too thin for Meta or TikTok to find a winner before the learning phase resets.
Which AI model writes the best ad copy? There’s no single winner. For short punchy hooks, GPT-5.5 tends to land harder; for keeping one brand voice across a batch, Claude Sonnet 4.6 is more consistent; for mining a long review or call transcript, paste it into Gemini 3.1 Pro or Sonnet 4.6 (both 1M-token context). Test the same prompt across two and keep the better output.
Why does my CTR drop after a few weeks even with new copy? Fatigue is often visual or pacing-driven, not a copy problem. If you’ve only rewritten text and CTR keeps sliding ~30% by week 4, change the image, the format, or the first-second pacing (prompts #8 and #9) before writing more headlines.
What’s the most important place to put the hook? The first line. Meta Feed hides primary text after ~125 characters behind “See more,” which only ~1% of users tap, and TikTok decides ~71% of watch-through in the first 3 seconds. If the hook isn’t in the opening, it effectively doesn’t exist.
Related
- Campaign slogan prompts
- Email marketing prompts
- Competitor Ad Analysis Prompts for Meta Ad Library
- Conversion-Focused CTA Prompts for Button Copy and A/B Tests
Tags: #Prompt #E-commerce #Ad creative