How to use the Ad Creative Generator agent skill

What the ad-creative skill makes the agent do

An agent skill is a folder of instructions and reference files that an AI agent loads when your request matches the skill's description. This one is for paid ad creative. Once it is active, the agent first asks about your platform and format, your product and offer, your audience, any existing ad results, and your constraints such as brand voice or compliance needs.

It then works in one of two modes. In generate-from-scratch mode it defines 3 to 5 distinct angles (such as pain point, outcome, social proof or curiosity), writes several variations per angle, and checks each piece against character limits. In iterate mode, where you supply performance data, it looks at which themes, structures and word patterns top performers share, and builds new variations on those while testing new angles.

• Platforms covered: Google Ads (Responsive Search Ads), Meta, LinkedIn, TikTok and Twitter/X
• Checks drafts against the character limits listed in the skill, and offers trimmed alternatives when a draft is over
• Organises output in a structure that maps to ad upload fields
• Points to a reference guide for generating ad images and videos with AI tools

Install the ad-creative skill

You can install the skill in one of two ways. Download the ZIP for ad-creative from the skill's page on this site, then choose the option that fits how you work.

For a personal setup available to all your projects, unzip the folder into ~/.claude/skills/. For a single project, unzip it into the .claude/skills/ folder inside that project. If you use the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into .claude/skills/ inside your project folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Check the skill is in the right folder

After unzipping, the skill folder should contain SKILL.md at its top level, next to a references folder and an evals folder. SKILL.md is the file the agent reads first. If the agent does not seem to use the skill, check that the folder structure is intact and that SKILL.md is present.

• Files included: SKILL.md, references/platform-specs.md, references/generative-tools.md, evals/evals.json

Trigger the skill with natural requests

You do not need a special command. The skill is described as being for creating and iterating paid ad creative, and for generating headlines, descriptions, primary text or large sets of ad variations. Ask for those things in plain language and name the platform and format if you know them.

If you already have a product marketing context file, the skill reads it first. The skill looks for .agents/product-marketing-context.md, or .claude/product-marketing-context.md in older setups, so it does not have to ask the same questions twice.

• "Write 10 Google Ads headlines for my project management app aimed at agency owners."
• "Create Meta primary text in three angles for a free trial of our bookkeeping tool."
• "Here is last month's ad export as CSV. Which themes won on CTR, and what should we test next?"
• "Draft LinkedIn intro text and headlines for a webinar, and check the character counts."

Work through a generation example

Suppose you ask for Google Ads headlines for a small accounting software product. The agent will ask who the audience is, what the free offer is, and what makes you different. Once you answer, it proposes angles such as a pain point (stop chasing invoices by hand), an outcome (get paid faster) and a comparison with spreadsheets.

Each draft is then checked against the 30-character headline limit and the 90-character description limit used in the skill's Google Ads table. Over-length drafts come back with a shorter version. Expect to review everything yourself before upload, because the skill itself says its specs are drafting targets.

• Good input: platform, format, offer, audience, one or two differentiators, banned words
• Good output to check: each headline reads on its own and in any combination, and includes a keyword, a benefit and a call to action

Iterate using performance data

For iteration, paste or attach your results. The skill asks which metric matters most (CTR, conversion rate or ROAS) and which creative is running. It works from data you supply; it does not pull live numbers from the ad platform itself.

The skill is explicit about the limits of this analysis. Compare the same attribution window, audience, placement and spend basis. Include impression, click, conversion and revenue counts, because rates alone do not establish significance. Google RSA asset labels are not independently randomised experiments, so treat rankings as descriptive rather than proof of cause.

• Include: impressions, clicks, conversions, revenue or spend per ad
• Keep the attribution window, audience and placement the same when comparing
• Treat the output as a list of hypotheses to test, not settled findings

Safety, requirements and what to review

The skill lists no required tools or packages. Its metadata says it does not run scripts and does not use the network. It works from text you give it, and its reference files are documents rather than executable code.

The skill is marked as high risk in its metadata, and it is not suitable for regulated ad claims. In finance, health and legal advertising, the skill only asks generic compliance questions and does not verify rules. Have a compliance reviewer check claims, and confirm every character limit and format against the platform's current documentation before you upload.

• Requires: nothing listed
• Runs scripts: no
• Uses network: no
• Confirm all specs with live platform requirements and previews
• Get compliance review for finance, health or legal ads