Find Your Winning Ad Angles With Claude and Meta Data

Start with one slice of your account that you can fairly compare. Export at the Ads level, not Campaign or Ad Set level, so each row points to one creative.

For a purchase-focused ecommerce review, include the fields that let Claude identify the ad, understand where it ran, and judge the business outcome:

• Identity: campaign name, ad set name, ad name, and Ad ID.
• Delivery: amount spent, impressions, reach, frequency, CPM, link clicks, link CTR, and CPC.
• Funnel outcomes: landing-page views, adds to cart, initiate checkout, purchases, cost per purchase, purchase conversion value, and website purchase ROAS.

If you want a Facebook-versus-Instagram read, add the Publisher platform breakdown before you export. If you want a more granular placement read, include Platform position too.

Do not export every available column. The useful export is focused enough to audit. More importantly, do not combine campaigns with different objectives, offers, attribution settings, currencies, or date windows in the same review.

Pro tip: Meta’s purchase labels can vary by account setup. Export the purchase event you optimize for, plus its cost, value, and ROAS fields. The exact label matters less than keeping it consistent across the rows you compare.

Create a new Cowork folder called meta-angle-review. Add the CSV, then paste in your brand’s public Meta Ad Library link.

The export tells Claude what happened. The Ad Library gives it a creative reference for active ads: copy, offers, format, and the visual execution it can inspect. Ask Claude to be strict about this match. An old exported row may no longer be active, and a similar-looking ad is not necessarily the same one.

Use this project instruction before you begin the review. It keeps the analysis grounded instead of letting Claude fill gaps with plausible-sounding creative interpretations.

You are analyzing a Meta Ads creative review for [BRAND]. Treat the performance CSV as the source of truth for delivery and purchase outcomes. Treat the public Meta Ad Library as the source of truth for active-ad copy and visual context.

Match an exported ad only when the ad name, offer, or copy makes the match clear. Label every export row as matched, ambiguous, or unmatched. Do not describe creative for ambiguous or unmatched rows.

Use purchases, cost per purchase, purchase conversion value, and website purchase ROAS as the primary evidence. Use CTR only as a diagnostic. Keep source files unchanged. Do not claim that an angle caused performance or will win in the future.

Now give Claude one clear analysis job. This prompt asks for the report you will use with your creative team, rather than a loose summary of top ads.

Analyze the Meta performance CSV in this project and open the provided Meta Ad Library link. Keep the source CSV unchanged.

First, flag non-comparable rows or thin evidence. For every export row, label the creative match as matched, ambiguous, or unmatched. Do not describe a creative for ambiguous or unmatched rows.

Find repeated ad angles only from matched rows when the hook, promise, or visual treatment is genuinely similar. Use purchases, cost per purchase, purchase conversion value, and ROAS as the primary evidence; use CTR only as a diagnostic.

Create a self-contained visual report called meta-ad-angle-report.html. Include:
- an executive summary;
- angle clusters with supporting ads and placements;
- performance ranges, confidence, and likely confounders;
- misleading high-CTR/low-purchase examples, if present; and
- three distinct next concepts, each with a hook, visual treatment, hypothesis, primary metric, and pause-or-remix rule.

Clearly separate observed evidence from hypotheses. Do not claim an angle caused performance or will win.

When the report opens, review the evidence behind one conclusion before you act on it. A useful angle is more specific than “video” or “UGC.” It should name the repeatable combination that the creative team can actually remake: for example, a problem-first hook plus a Day 1-to-Day 7 proof structure, or a customer result tied to one recognizable routine moment.

Pro tip: A high CTR is not proof that an angle is working. If the ad attracts clicks but has weak purchases, CPA, or ROAS, treat it as a diagnostic: the opener may be strong while the promise, product fit, or landing-page expectation is off.

Keep the final output narrow. Ask Claude to package only three next tests, each based on a different observed pattern. Each test should tell a copywriter or designer what to make and how you will decide whether it earned another iteration.

Use this follow-up:

Using only the matched evidence in the report, create next-creative-tests.md with three test cards.

For each card, include:
- the angle to test;
- a distinct hook;
- the visual treatment and proof moment;
- the hypothesis;
- the primary purchase metric; and
- a pause-or-remix rule.

Do not repeat the same concept three times. Keep the recommendations proportional to the evidence and flag assumptions that still need validation.

Then run the same review on a CSV filtered to only Instagram or only Facebook. That placement-specific pass helps you isolate where an angle is worth adapting, instead of assuming the same execution belongs everywhere.