Turn 50 winning ads into 10 fresh angles with Claude

You are a senior performance creative strategist. Below is a set of real ads I collected from public ad libraries in my category. Each row has the hook word for word, the format, the offer, the dominant emotion, the CTA and landing page, and how long the ad has been running.

Read the whole set before you write anything, and analyse it as a body of evidence rather than ad by ad. Tell me: the promise each hook opens with and how many ads share that promise, the specific belief or objection each cluster is attacking, which offer shapes repeat, which emotions do the heavy lifting, and which formats pair with which promises. Give me counts, not adjectives, so I can see how strong each pattern is.

Then flag the outliers: the two or three ads that do something nobody else is doing, and say whether that looks like a smart untapped angle or like a test that quietly died. Separately, tell me what is conspicuously ABSENT from this whole set, meaning the promises, audiences or objections that nobody in my category is addressing. Finish with a short honest note on where this sample is weak or biased.

My product and audience: [describe what you sell, who buys it, and what they are usually using instead].

Now group every hook in that set into named hook types, and build me a reusable taxonomy rather than a summary.

For each type: give it a short memorable name, write the underlying formula as a fill-in-the-blank template with the variable parts in square brackets, quote two real examples from my set, name the audience state it works on (unaware, problem-aware, solution-aware, product-aware), and say what makes it fail or feel cheap when it is done badly. Order the types by how much evidence my set gives each one, and say how many ads sit in each.

Then write one line per type explaining what it would sound like for MY product specifically, so I can tell at a glance which types transfer to me and which do not. Keep every template short enough to say out loud in under four seconds, because that is the real constraint on a video hook.

Using the hook taxonomy and the absences you found, write me 10 fresh ad angles for my product. An angle here is the core argument the ad makes, not a slogan and not a hook line.

Hard rules: no angle may be traceable to any single ad in my set, and every angle must combine at least two of the patterns you identified, or take one pattern into an audience or objection that my set showed nobody is covering. Do not write anything my product cannot actually back up.

For each angle give me: a one-line statement of the argument, the exact belief it changes in the buyer's head, who it is for and where they are in awareness, the proof or demonstration the ad would need on screen for it to land, three hook lines in different hook types from the taxonomy, the format I should shoot it in, and one honest sentence on why this angle might flop. Rank all ten by how cheap they are to test against how much they would change my results if they worked, and tell me which three to shoot first and why.

My product, my proof, and what I can realistically film or design: [fill this in properly, the angles are only as good as this paragraph].

Take angle number [X] and turn it into a production brief I can hand to a person.

Write the 30 second script with the hook in the first two seconds, the on-screen text for each beat, and the shot or asset needed for every line. Then give me three alternate hooks for the same angle so the test is a hook test rather than a whole new concept, and a first-frame and thumbnail idea that makes sense with the sound off.

Add the test plan underneath: what I am measuring in the first 48 hours, roughly what a fair test costs at my budget of [your daily budget], what result would mean I keep going, and what result would mean the angle is dead rather than the hook. Keep the whole brief on one page and write it so somebody who was not in this conversation can shoot it without asking me a single question.

Build me a reusable Claude skill called "ad-angles" that runs my whole ad research process end to end.

The description should say to use it whenever I paste a set of competitor ads and ask for angles, patterns or hooks. The instructions should run four stages in order: pattern breakdown across the whole set with counts, a named hook taxonomy with fill-in-the-blank templates, ten fresh angles that each combine at least two patterns and are never traceable to one source ad, and a one-page production brief for the top-ranked angle.

Build in these rules: never invent performance numbers or spend for an ad, treat longevity and repetition as the only honest proxies for what is working, always name what is absent from the sample as well as what is present, and always end with an honest note on sample bias. Before you write the SKILL.md, ask me what category I work in, what I sell, and what my typical test budget is, so the output is shaped to my actual situation.