Claude Prompts for Facebook Ads: 25 Prompts That Work

How to use this library

Each prompt below has blanks in [brackets]. Fill them, paste, run. A few habits make all of them work better:

• Paste a real ad. Prompts that say "reference ad" want a full script or primary text, with its estimated spend and days live.
• Keep a facts block. Write your product facts and forbidden claims once, save them in a Claude Project, and every prompt can refer to them.
• Ask for tables. Tables are easier to check and to paste into a brief.

Some numbers we use as constraints come from the 2,443 winning Meta ads on our board (October 9, 2026, all with $50k+ in estimated spend):

Prompts that mention search_ads, get_ad, winning_patterns, advertiser_hooks, get_swipe_prompt or list_taxonomy need the Ad Radar MCP connector. Without it, paste the ads yourself.

Research (prompts 1 to 5)

1. Niche sweep

Use search_ads for niche "[niche]", active, sorted by score, limit 30.
Table: brand, opener (verbatim), format + length, est. spend, days
live, variants, hook pattern. Then 3 bullets on what repeats.
Label spend as estimated. Never invent an ad.

2. Separate tests from winners

From this list of ads, keep only those live 90+ days or with 2+
variants. For the rest, say "possible test". Explain in one line why
days live and variant count are stronger signals than likes.

3. One brand's playbook

Call advertiser_hooks for [brand]'s advertiserId. Then search_ads for
that brand and quote the 3 openers with the highest estimated spend.
What is the repeatable pattern, in one sentence?

4. What beats the baseline

Call winning_patterns for [video/image]. List the 5 tag values with
the highest lift and the 5 lowest. Flag any value with fewer than 20
ads as a small sample. Which 2 should my next test use?

5. Transcript search for an idea

search_ads with query "[phrase, e.g. 'I was skeptical']", limit 20.
For each hit quote the sentence containing the phrase and the next
one. Which variations of the phrase appear in ads with 90+ days live?

More on the research process in competitor ad research inside Claude.

Hooks (prompts 6 to 10)

6. Openers in different patterns

Product facts: [facts]. Audience: [who], [awareness level].
Write 10 openers, 2 each in: problem/benefit, specific number,
first-person story, question, warning. 8 to 15 words each.
Spoken English. No brand name in the first 5 words.

7. New hooks on a proven body

This script is our current winner: [script]. Keep everything after
the first sentence. Write 8 new first sentences, each in a hook
pattern the original does not use, each leading naturally into the
original second sentence.

8. First frames, not just first words

For each opener below, describe the first 2 seconds on screen in one
line: what the camera sees, any text overlay (max 7 words), and the
sound. It must be filmable with a phone. [openers]

9. Hook from a review

Here are 30 customer reviews: [reviews]. Find the 5 most specific
sentences (numbers, situations, objections overcome). Turn each into
an opener in the customer's own words, max 15 words.

10. Hook critique

Score each opener 1-5 on: specificity, who it calls out, curiosity,
believability. One-line fix for anything under 4. [openers]

A reference ad worth pasting into prompt 7, because it shows a first line that only works with the second:

For the patterns behind these prompts, see Facebook ad hooks.

Copy (prompts 11 to 15)

11. Primary text in five patterns

Facts: [facts]. Forbidden claims: [list]. Reference ad: [text].
Write 5 primary texts, 50 to 90 words, one per hook pattern.
A concrete detail or number in the first 2 sentences.
If you need a fact I didn't give, write [NEEDS FACT].

12. Headlines

Write 15 headlines for this primary text, 4 to 8 words. Five state
the outcome, five state the offer, five name the audience.
No emoji in ten of them. [primary text]

13. Long-form story ad

Write a 250-word first-person story ad. Narrator: [persona]. One
specific scene in the first sentence (place, time, a number).
The product appears after word 120. End on the offer.
Match the sentence rhythm of this reference: [story ad].

14. Match copy to awareness level

Rewrite this ad for each awareness level: unaware, problem-aware,
solution-aware, product-aware. Change only what the first 2
sentences assume the reader knows. [ad]

15. Compliance pass

Flag every claim in this copy that needs proof or could breach Meta's
ad policies (health outcomes, before/after promises, personal
attributes like "Are you overweight?"). Suggest a safer line for each.
[copy]

Prompt 15 is a first pass, not legal advice. Claims remain the advertiser's responsibility. Awareness levels are explained in awareness levels in ads.

Video scripts (prompts 16 to 19)

16. Beat map of a winner

Call get_ad on [ad]. Split the transcript into beats with the job of
each (hook, problem, mechanism, proof, offer, CTA) and the share of
total words each beat uses.

17. Swipe the structure

Call get_swipe_prompt on [ad], kind "swipe", lang "en", product:
"[product facts]". Follow the prompt it returns. Afterwards, list any
line in your draft that is within 5 words of the original.

18. Length cut

This script runs [X] seconds. Cut versions at 30s and 15s. Keep the
hook and the offer word for word. Tell me what you removed and why.

19. Demo script

Write a 45-second demo script for [product]. Every line must describe
something happening on screen. Max 2 lines of talk without a visible
action. End with the product in use, then the offer.

Here is a reference for prompt 19: a demo that has run a full year with almost no pitch.

Script workflow in more depth: writing video ad scripts with AI.

Briefs and review (prompts 20 to 22)

20. Brief from a winning ad

Extract the decisions behind this ad: awareness, angle, hook pattern,
first frame, body beats, proof (shown or told), offer, format, length.
Write "not visible" for anything you can't see. Then fill my brief
template for [product] with them. [ad] [template]

21. Creator brief

Turn this brief into a one-page note for a UGC creator: what to say
in the first 3 seconds word for word, what to show, what they may
improvise, what they must never say. Plain language. [brief]

22. Weekly review note

Here is last week's ad export: [CSV]. Rules: winner = CPA under
[target] with 2x target spent; loser = CPA over 1.3x target after
2x target spent. List winners, losers, still testing. 5 bullets on
what to brief next, each citing an ad name.

Analysis (prompts 23 to 25)

23. Why this ad works

Call get_swipe_prompt on [ad] with kind "analysis". Follow it. Then
add: which part of the ad I could test for my product within a week,
and which part depends on assets I don't have.

24. Tag my own ads

Call list_taxonomy. Using those exact values, tag each of my ads on
hook pattern, angle, awareness and proof level. Table. Then compare
my mix with winning_patterns and name the gap.

25. Fatigue check

Here is daily data for one ad over 6 weeks: [CSV with spend, CPM,
CTR, CPA, frequency]. Has it fatigued? Show the numbers that say so.
If yes, propose 3 new openers on the same body (see prompt 7).

Tagging your own ads properly is covered in tagging ads with AI.

Three rules that make any prompt better

1. Give it the evidence, ask for the evidence. Paste real ads in, and ask Claude to cite the ad behind every pattern it claims.
2. Name the patterns. "Ten hooks" gets you one idea ten ways. "Two each in five patterns" gets you five ideas.
3. Make missing facts visible. [NEEDS FACT] and "not visible" are the two most useful strings in any ad prompt.

Figures marked as estimated spend come from Ad Radar's model of engagement on public Meta Ad Library ads. They are estimates, labeled as such, and are best used to rank ads against each other.