Steal Your Competitor's Ad Playbook with Claude + Meta Ad Library

1. What You End Up With

Before the steps, here is the finish line. At the end of this workflow you have one page that answers five questions about your market:

• What each competitor is saying. Every active ad, grouped by hook and angle, with the exact first line.
• What is probably working. The ads that have run for 60 days or more, and the ones copied across many ads.
• What everyone is saying. The crowded angles where one more ad like theirs will just blend in.
• What nobody is saying. The open angles, offers and formats you can own.
• What to make next. Five hooks, three angles and the formats to test, written for your product, not copied from theirs.

2. What Claude Can and Can't See

Claude reads the Ad Library page the same way you would, just faster and without getting bored. That means it sees exactly what the page shows, and nothing more:

Why browsing beats the Meta connector for this

If you already run Meta ads with Claude through the MCP connector, you might expect it to do this job. In my tests it returned only the newest ads, without their text or their start dates, which removes the two things this method depends on. Letting Claude browse the Ad Library website gives it the full page instead.

Prefer to do it by hand, or want to understand every signal first? My step-by-step guide to the Meta Ad Library is the manual version of this workflow.

3. Step 1: Point Claude at the Ad Library

You need a version of Claude that can open web pages and scroll them, such as the Claude in Chrome extension or the browser built into the Claude desktop app. Then pick three competitors: the brands your customers compare you with, not the biggest names in your industry.

Paste the prompt below and swap in your three Page names. Claude takes a few minutes per competitor; a big advertiser with 50+ ads takes longest, because the page loads ads as you scroll. If one competitor has hundreds of ads, add "stop after the first 60 ads, sorted by most recent" to keep it manageable.

Open the Meta Ad Library (facebook.com/ads/library).
Set country to India, ad category to All ads, and active status to Active ads.

For each of these Facebook Pages, search the Page name and open its ads:
1. [Competitor 1]
2. [Competitor 2]
3. [Competitor 3]

Scroll until every active ad has loaded. For every ad, record:
- Library ID
- "Started running on" date, and days running as of today
- Format: video (with length), image or carousel
- Versions: "multiple versions" or "X ads use this creative"
- The first line of the ad text, word for word
- The offer (price, discount, free element, guarantee)
- Button text and landing page domain

Give me one table per competitor. Do not summarise yet.

4. Step 2: Group Every Ad by the 5 Lenses

A table of 80 ads is still just a pile. The second prompt turns it into patterns by sorting every ad through the same five lenses. Tap each lens to see what Claude looks for:

The grouping prompt

Paste this straight after Claude finishes the tables, in the same chat:

Now analyse all the ads above, across all three competitors.

1. Hooks: group the first lines by type (question, audience call-out, bold claim, price or free, story, urgency). Count how many ads use each.
2. Angles: what is the core promise (career, money, risk removed, AI future, who it is for)? Count by competitor.
3. Offers: list every offer and which competitor uses it.
4. Formats: video vs image vs carousel per competitor, and the most common video length.
5. Longevity: sort ads into Testing (under 14 days), Working (14 to 60) and Proven (60+). Flag any message copied across 3 or more ads.

End with two short lists: "Everyone is saying" and "Nobody is saying".

5. Step 3: Find the Ads That Are Actually Working

This is the step most people skip, and it is the one that matters. Claude will flag the long-runners, but read its list with these rules in mind:

• 60+ days is your strongest signal. An ad that has run for two months or more is very likely paying for itself.
• Copies mean scaling. When the same text and creative appear in many ads, the advertiser is pushing more money into something that works.
• Lots of new ads means testing. A burst of ads that all started in the last week tells you what they are trying, not what works yet.
• One message for months is a winner. If a competitor runs a single hook for half a year while testing nothing else, that hook is carrying their business.
• Ask Claude why. For each proven ad, ask: "In one line, why do you think this ad keeps running?" The answers become your hypotheses.

6. Step 4: Turn It into a Creative Brief

The last prompt is where you stop studying them and start planning your own ads. Describe your product honestly, because the brief is only as good as what you tell Claude.

Then edit the brief yourself. Claude is very good at spotting patterns and terrible at knowing your customers the way you do. Keep the angles that sound true, cut the ones that sound like marketing.

Using everything above, write a creative brief for my product:
[What you sell, the price, who it is for, your real proof]

Include:
- 3 angles I can own that none of these competitors use
- 5 hooks for those angles, in my own voice
- Which offers to test, and which to avoid because everyone uses them
- 2 video ideas and 2 image ideas, with the format and length
- What to borrow from the proven ads (structure, not wording)

Do not copy any competitor's wording. Flag anything you're unsure about.

7. A Real Run: "Performance Marketing Course" Ads in India

To show what this looks like, I ran the workflow on 21 October 2026 on my own market: an exact-phrase search for "performance marketing course" in India, active ads only. It returned about 130 ads, and I analysed the 84 that loaded. Three advertisers ran most of them. I've described them instead of naming them, because the pattern is the point.

What the brief said

Here is the condensed brief Claude produced for a course like mine, a ₹2,499 practical performance marketing course taught by someone who runs real accounts:

• Everyone is saying "AI". Every advertiser sells "AI-powered" or "AI + performance marketing". It no longer sets anyone apart, so lead with something else.
• Nobody shows proof. Not one ad showed a real campaign the instructor ran, a real ad account or real numbers. That is the open lane for a practitioner.
• Nobody speaks Hindi or Hinglish. The biggest advertiser wins in Malayalam; the rest run in English. A Hinglish angle is unclaimed.
• Risk reversal works here. The 8-month winner removes the fear of wasting money. Borrow the structure, a clear guarantee or low-risk entry, not the words.
• Call out your buyer. Competitor B shows that naming who the course is not for builds trust. Test "Built for freshers who want proof, not certificates".
• Hooks to test: "Here is the ad account I learned on", "₹500 of ads taught me more than 50 hours of videos", "What a ₹2,499 course should show you before you pay".

8. The Full Workflow on One Page

Run this once a month, or before any big campaign, and keep each month's output. Over time you will see which competitors are stuck on the same message and which ones are learning.

9. Mistakes That Waste the Whole Exercise

The workflow is quick, which makes it easy to misuse. Avoid these:

• Copying ads word for word. It is lazy, it can be a legal problem, and Meta's system treats near-identical creative as the same idea, so you compete head-on with an advertiser who has a bigger budget. Since the Andromeda update, genuinely different creative matters more than ever.
• Trusting a single ad. One new ad tells you nothing. Patterns across many ads and many weeks are what count.
• Ignoring the landing page. A great hook pointed at a weak page still fails. Open the top three landing pages yourself.
• Picking the wrong competitors. A national brand with ten times your budget is playing a different game. Study the ones your buyers actually compare you with.
• Never writing your own hooks. Research is the start, not the job. My three hook formulas turn the brief into ads you can launch this week.