The Claude skill that spies on your competitors' ads, Quentin IA Marketing

Every ad your competitors run on Facebook and Instagram is visible, for free, in Meta's Ad Library. The creative, the copy, the button, the landing page, and above all the date each ad started running.

That date is the most underused piece of information in marketing. A brand does not leave an ad running for three months by accident: it keeps paying for it because it brings something back.

An ad that has been running for months is a competitor showing you what they keep paying for.

The problem is time. Going through five competitors, logging every ad, working out how long each has run, breaking down the hooks: that is a whole afternoon. This Claude skill does it for you, then turns the analysis into six ad concepts, with the briefs and the test plan.

Claude AI Lab

The Claude AI Lab is my Skool community where I share my Claude systems and the more advanced modules. Access is $67/month.

Join the Lab →

✅What you need

A paid Claude plan (Pro, Max, Team or Enterprise). That is what Claude in Chrome requires.

The Claude in Chrome extension, installed and signed in to your account.

A list of 5 to 8 competitors. The skill helps you build it.

The full skill is here: download meta-ad-spy.md. Rather copy it? Open it in your browser, select all, copy.

Install Claude in Chrome

🧩 Chrome Web Store

The Ad Library only loads in a real browser. Paste an Ad Library link into a regular Claude chat and it cannot open the page, so it sees no ads at all. Claude in Chrome gives it access to your own Chrome: it opens the pages, scrolls the ads and reads the details itself.

Add the extension from the Chrome Web Store.

Pin it and sign in. Click the puzzle icon in the Chrome toolbar, pin Claude, then sign in with your Claude account.

On Claude Code? Run /chrome, check the integration is enabled, then start a new chat so the browser tools load.

No paid plan

The skill still works. It tells you exactly which screenshots to send for each ad: the start date, the ad's Library ID, the copy and the button.

Run the skill

▶️ Claude in Chrome

Two ways to do it, one minute each.

The fast way. Open Claude in Chrome, paste the whole skill, then send: "Run the Meta Ad Spy skill. My brand is [your brand] and here are my competitors: [names or links]."

To keep it. In Claude Code, create the folder ~/.claude/skills/meta-ad-spy/ and save the file inside it as SKILL.md. On claude.ai, paste it into the instructions of a Project dedicated to your ad research.

The skill opens with a single batch of ten questions: your product, your customer, your offer, the proof you actually have, what you are not allowed to claim, what you can produce. Answer in one go. If you write "just go", it moves on with what it has and labels every assumption.

Give it the right competitors

🔎 facebook.com/ads/library

The quality of the analysis depends on the mix you give it. The skill asks for three groups:

🎯

3 to 5 direct competitors

Same product, same customer. The ones your prospects compare you to.

🏔️

1 or 2 role-model brands

Bigger than you in your category. They test more, so they have more ads that last.

↔️

1 adjacent brand

Different product, same customer. That is where you find the angles your category has never tried.

One trap in the Library: a keyword search also returns advertisers who simply mention the brand's name. Type the brand into the search box and pick its Page from the dropdown. That Page's link contains view_all_page_id= followed by a number: bookmark it, you will need it again next month.

If your competitor sells in Europe

Ads delivered in the EU show an estimated reach by age, gender and country, and stay visible for a year after they stop. So you also see what they killed fast, not only what survives.

What you get back

📋 one report

The report always follows the same order, and every detail says where it came from: browsing, screenshot or pasted text. Anything the skill did not see is marked "not seen", never guessed.

📒

A ranked ledger

Every ad with its days live, format, verbatim hook, offer, button and landing page, scored from 1 to 5.

🔬

The top ads broken down

The hook in three layers (what you see, read and hear), the angle, the awareness stage, the proof, and two or three hypotheses on what makes the ad work.

🧭

Patterns and open angles

What everyone runs, what the long runners share, and the angles, formats and objections nobody covers.

💡

Six concepts for your brand

Two on the proven pattern, two on open angles, one only you can make, one wild card. Three hooks and a script each.

🧾

Briefs and a test plan

Briefs a creator or editor can shoot without calling you, with kill and scale rules.

Reading a long-running ad

⏱️ the start date

Runtime is a strong signal, and the skill treats it as a signal, never as proof.

The tiers. Under 30 days: still in testing, do not learn from it yet. 30 to 44 days: the ad survived the first cuts. 45 to 59: likely working. 60 days and more: strong signal, they have paid for it for two months.

The best signal is the relaunch. The same creative relaunched as several separate ads, with fresh start dates, means the brand is feeding a winner.

"Multiple versions" means little. That label usually comes from dynamic creative, where Meta mixes images and copy. It is a setting, not a budget increase.

Watch out for retargeting. An ad that talks about "your cart" can run for months on a small warm audience. It lasts, but it teaches you nothing about acquisition.

Keep in mind

The Library shows no spend, clicks or sales for regular commercial ads. A start date tells you how long an ad has been running, not how much it makes.

Testing what it makes

🧪 Ads Manager

According to Motion's Creative Benchmarks 2026 (578,750 creatives analyzed between September 2025 and January 2026), about one creative in twenty becomes a winner. Which is why you test several truly different concepts at once rather than one polished one.

One concept, one ad. Launch each concept separately so it gets its own read. Ads Manager's Creative testing tool tests up to 5 creatives inside one ad set, with the test budget split evenly.

Judge on spend, not days. Roughly $500 to $1,000 per concept, or 2 to 3 times your target cost per purchase, before you call it.

Read in order. Hook rate first (3-second plays divided by impressions), then click-through rate, then cost per purchase.

Hooks come later. You only test new hooks once a concept has won.

What the skill refuses to do

Copy an ad. No competitor logo, name, tagline or footage in your concepts, no line that assumes a personal trait of the reader ("Are you struggling with debt?" gets rejected by Meta), and no fake testimonial, AI-generated or not.

Run it again every month

The real lever is repetition. Run the skill every 30 days on the same competitors: it notes which ads died and which appeared. An ad that disappears teaches you as much as one that stays, because it shows you what your competitor paid to learn did not work.

Once your concepts are live, the account needs a structure to run without you: that is what the team of 7 Claude Code agents for a Meta Ads account sets up.

And if you have not done it yet, download the skill and run it tonight on your first competitor.