Claude Skills for Meta Ads: 6 Meta-Only Skills and 15 Prompts

Why Meta needs its own skills

Meta punishes slow reaction. An ad that held 1.3% CTR for two weeks slides to 0.7% in eight days, frequency passes 4, and the CPA damage lands a week after the CTR did. The six skills here exist because those checks are numeric, repeatable, and skipped whenever someone is busy. Each one is a file in the Claude skills for Meta Ads folder of the Ryze marketing-skills repo; this page is the only place each is described in full.

Everything here is Meta-only. Nothing about Google Ads, and nothing that belongs to the cross-platform Claude ads skills, which have their own page. If you have never used a skill, start with the prompts: they are the same instructions without the install step.

Which Meta Ads skill for which problem

Match the symptom to the skill, pull the export in the third column, run it on the cadence in the fourth. The first row is where nearly every account should start.

What are Claude skills for Meta Ads?

A Claude skill is a folder with a SKILL.md inside: YAML frontmatter with a name and a description that says when it applies, then the steps, the input and the output. Claude loads the full file only when your request matches the description. The six Meta files add one line to the frontmatter, metadata: platform: Meta, and every description mentions Meta ad sets, frequency, creatives or the Ad Library, so a request about Meta triggers them and a request about search terms does not.

What makes a Meta skill different from a generic one is the input contract. Meta’s problems live in breakdowns: fatigue is invisible without a daily breakdown, overlap is invisible without ad-set targeting, placement waste is invisible without the delivery breakdown. Each skill below names the Ads Manager export it needs, down to the tab, columns and breakdown, because a skill run on the wrong export returns confident nonsense.

How to install Claude skills for Meta Ads

The six files live in one folder of the Ryze marketing-skills repo. Put each SKILL.md in its own folder named after the skill before installing. The repo root also ships all-marketing-skills.zip if you want the whole library at once.

claude.ai (web and mobile)

Go to Settings › Capabilities and turn on code execution and file creation. Open Customize › Skills, click +, Create skill, Upload a skill, upload the folder as a zip, toggle it on. Then attach an export and ask in plain words: “which of these ads are fatigued?” Claude picks the skill and says so.

Claude Code

Copy each folder into ~/.claude/skills/ for yourself or.claude/skills/ inside a repo for the team. Invoke with /creative-fatigue-detection or let Claude trigger it. This is where the Thursday check becomes a script: pull the export, run the skill, write the table to a file.

Claude Desktop and Cowork

The desktop app shares your claude.ai skills and adds a schedule. Give Creative Fatigue Detection a Thursday 08:00 slot and the flagged list is waiting before you open Ads Manager. Claude Cowork for Meta Ads walks through the scheduled setup.

Connector note. A skill does not grant data access. Without a connector it works on the file you attach. With the Claude MCP connector for ad accounts attached, delete the “attached export” wording from any prompt below and Claude queries the account itself. Keep it read-only for analysis; which Claude connector for Meta Ads compares the options.

The six Claude skills for Meta Ads

Each card shows the repo file, the export to pull, the frontmatter that triggers it, the prompt it wraps so you can run it before installing anything, and the output shape. Cadence is ours; the thresholds are the files’.

Source: the Ryze marketing-skills repo, file 04-meta-creative-fatigue-detection.md.

01. Catch decaying ads before the CPA shows it

Skill: Creative Fatigue Detection

Tracks CTR trajectory, frequency buildup, CPM movement and engagement decay per ad and sorts every active ad into urgent, warning or healthy. The file’s rule of thumb: three or more days of consistent decline is fatigue, not noise. Higher budgets burn creative faster, so rerun it after scaling.

Export it needs: Ads Manager › Ads tab, last 30 days, Breakdown › By time › Day. Columns: ad name, ad set, amount spent, impressions, reach, frequency, CTR (all), CPM, results, cost per result. Export as CSV.

Output: one table, ad | status | days active | spend | peak CTR | current CTR | drop % | frequency | verdict, sorted by spend; then days of life left for warning ads and a rotation schedule.

Cadence: Weekly, Thursday, and after every budget increase. Mode: read-only.

Source: the Ryze marketing-skills repo, file 08-meta-audience-overlap-analysis.md.

02. Find the ad sets bidding against each other

Skill: Audience Overlap Analysis

Maps overlap between ad sets from their targeting (custom audiences, lookalikes, interest stacks, age and gender, geo) and cross-checks it against delivery so you learn which overlap is inflating CPMs and which is theoretical. The file’s example: three of six prospecting ad sets sharing 60%+ of their audience ran 28% higher CPMs than the rest.

Export it needs: Ads Manager › Ad sets tab, last 30 days, no breakdown. Columns: ad set name, campaign, targeting (audience), reach, impressions, frequency, CPM, results, cost per result. Add the percentages from Audiences › select 2 to 5 audiences › Show Audience Overlap if you have them.

Output: an overlap map of ad-set pairs with severity, the estimated CPM inflation, the specific interests, lookalikes or custom audiences causing it, and consolidation or exclusion recommendations with projected savings.

Cadence: Monthly, and whenever you add ad sets. Mode: read-only.

Source: the Ryze marketing-skills repo, file 13-meta-competitor-creative-analysis.md.

03. Read the Ad Library for angles nobody is running

Skill: Competitor Creative Analysis

Breaks a competitor’s active ads into formats, messaging themes (pain point, benefit, social proof, urgency), CTAs, offers and refresh rate, then maps that against your own creative to find white space. The file’s example finds three SaaS competitors all leading with feature comparisons and demo CTAs, and none using testimonial creative with a trial offer.

Export it needs: No Ads Manager export. Open the Meta Ad Library, filter by advertiser, and collect for 3 to 5 competitors: number of active ads, format per ad, primary text and headline, CTA, and start date. Paste as text or a table. Ryze’s competitor Ad Library view collects the same fields.

Output: a per-competitor inventory by format, theme and CTA, refresh rate, messaging gaps you can own, and a launch timeline that shows strategy shifts. Pairs with the automated creative workflow in automated ad creation.

Cadence: Quarterly, and before creative planning. Mode: read-only.

Source: the Ryze marketing-skills repo, file 18-meta-frequency-cap-recommendations.md.

04. Find the frequency where impressions stop converting

Skill: Frequency Cap Recommendations

Correlates frequency with conversion rate, CTR and CPA per campaign type to find the inflection point, then recommends caps by objective and estimates the spend you would redirect to fresh reach. The file’s example: prospecting conversion rate peaked at frequency 2.3 and fell 40% by 4, while the account averaged 5.1; recommended cap 3 per 7 days for prospecting, 5 for retargeting.

Export it needs: Ads Manager › Ad sets tab, last 30 days, Breakdown › By time › Week. Columns: ad set name, campaign objective, reach, impressions, frequency, CTR, results, cost per result, amount spent.

Output: the frequency distribution, optimal frequency per campaign type from your own data, recommended caps with projected savings, and the reach that spend would buy instead.

Cadence: Monthly, and when scaling. Mode: read-only.

Source: the Ryze marketing-skills repo, file 22-meta-retargeting-window-analysis.md.

05. Size the retargeting window from conversion lag, not habit

Skill: Retargeting Window Analysis

Uses time-to-conversion data to decide whether the 30-day window should be 7, 14 or 60 days per segment, and where retargeting spend stops returning. The file’s example: 78% of conversions from Meta-sourced visitors landed within 7 days and 4% after day 14, so the 14 to 30-day window for that traffic was cut.

Export it needs: Two inputs. Ads Manager › Ad sets tab for ad sets built on website custom audiences of 7, 14, 30 and 60 days: spend, results, cost per result, frequency, last 90 days. Plus the GA4 Advertising › Attribution › Time lag report, or your own days-to-conversion export, split by source.

Output: a lag distribution per segment, the recommended window per segment with confidence, spend and conversion share per window, and bid modifiers by recency. The marginal-CPA logic behind it is in marginal CPA vs average CPA on Meta.

Cadence: Quarterly, and after funnel changes. Mode: read-only.

Source: the Ryze marketing-skills repo, file 41-meta-ads-audit.md.

06. The onboarding audit, scored out of 100

Skill: Meta Ads Audit

Runs the others in sequence: fatigue, overlap, tracking health, structure, prioritized actions. Its thresholds are explicit: frequency above 3 to 4 means refresh or expand, a 15 to 20% CTR drop over 7 days means new creative, CPM up 30 to 40% in two weeks means a creative reset, overlap over 30% means consolidate or exclude, frequency above 7 is severe. It asks for real iOS tracking-loss numbers rather than assuming them.

Export it needs: Three exports, minimum 14 days, Breakdown › By time › Day: campaign level, ad set level (with targeting and budget type) and ad level, each with spend, impressions, reach, frequency, CTR, CPM, results, cost per result. Plus a note or screenshot from Events Manager: Pixel status, Conversions API status, Event Match Quality score.

Output: a health score out of 100, a fatigue alert table, overlap findings per ad-set pair, a tracking checklist, scaling opportunities, and a confidence level. The manual version of this audit is in how to use Claude for Meta Ads.

Cadence: Onboarding, then monthly. Mode: read-only.

15 Claude prompts for Meta Ads, grouped by job

Prompts the skills above do not already contain, each with the export to attach. They work in claude.ai, Claude Desktop and Claude Code with a CSV; with a connector, drop the “attached” wording. Industry CPM and CPA reference points for the measurement prompts are in Meta Ads cost benchmarks by industry. If you use ChatGPT for some of this, the ChatGPT prompts for ads page has the equivalents.

Creative (4)

Prompt 1: Creative briefs from what already wins

Attach: Ads tab, last 60 days: ad name, format, primary text, headline, CTR, conversion rate, spend.

Prompt 2: Eight copy variants, one variable each

Attach: Your best ad's primary text, headline, description and CTA, pasted.

Prompt 3: Creative scorecard, 1 to 100 per ad

Attach: Ads tab, last 30 days: ad name, format, CTR, conversion rate, ROAS or cost per result, spend, days active.

Prompt 4: Video hooks: who holds the first 3 seconds

Attach: Ads tab, video ads only, last 30 days: impressions, 3-second video plays, ThruPlays, average play time, CTR, cost per result.

Audiences (3)

Prompt 5: Which lookalikes are stale

Attach: Ad sets tab, lookalike ad sets, last 90 days, breakdown by month: spend, results, cost per result, CTR, audience size; plus the seed list dates.

Prompt 6: Twenty interests to test next

Attach: Ad sets tab, interest-targeted ad sets, last 60 days: targeting, spend, CTR, cost per result, ROAS.

Prompt 7: Age, gender, device and placement: who actually converts

Attach: Campaigns tab, last 30 days, Breakdown by Age and Gender; a second export by Placement; a third by Device. Columns: spend, results, cost per result, ROAS, frequency.

Budget and structure (3)

Prompt 8: CBO or ABO for this campaign

Attach: Ad sets tab for one campaign, last 30 days, breakdown by day: ad set budget, spend, results, cost per result, reach, frequency.

Prompt 9: Advantage+ or manual, per campaign

Attach: Campaigns tab, last 30 days: campaign type, objective, daily budget, spend, results per week, cost per result, ROAS, targeting notes.

Prompt 10: Budget split across funnel stages

Attach: Campaigns tab, last 30 days: campaign name, targeting type, spend, results, cost per result, ROAS, frequency, reach.

Measurement (3)

Prompt 11: Which attribution window is telling the truth

Attach: Campaigns tab, last 30 days, Compare attribution settings: results and cost per result under 1-day click, 7-day click, and 7-day click plus 1-day view.

Prompt 12: Tracking health: Pixel, CAPI, EMQ

Attach: Notes or screenshots from Events Manager: event list, Pixel status, Conversions API status, Event Match Quality per event, deduplication, AEM priority order.

Prompt 13: Why did CPM jump

Attach: Campaigns tab, last 30 days, breakdown by day: impressions, spend, CPM, frequency, CTR, reach.

Reporting (2)

Prompt 14: The Monday Meta digest

Attach: Campaigns tab for this week and last week: spend, impressions, CTR, results, cost per result, ROAS, frequency; plus the Ads tab for both weeks.

Prompt 15: Placement ROAS: where the money actually goes

Attach: Campaigns tab, last 30 days, Breakdown by Placement, split into prospecting and retargeting campaigns: spend, impressions, results, cost per result, ROAS.

Tools that automate the Monday and Thursday rows, including Ryze, are compared in the top AI tools for Meta Ads management.

A Meta-only weekly cadence

Two habits pay on Meta: rotating creative before the weekend, when spend and frequency both climb, and fixing structure monthly rather than every time a CPM looks odd.

The funnel these slots maintain is described in the advanced Meta Ads full-funnel strategy with Claude. If a slot keeps getting skipped, rewrite its output contract before you move the slot.

How to write a Meta Ads skill for Claude

The repo files are short by design, and the first thing you will change is the threshold: a B2B account with small audiences fatigues at frequency 5, not 3; a Reels-heavy account cares about hold rate more than CTR. Copy the file, edit the numbers, keep the four parts. Below is a complete SKILL.md for creative fatigue with the export spelled out, which the public file leaves implicit.

Two lines carry most of the value: the description, which makes the skill fire on “why is CTR falling?” without you naming it, and the echo-back step, which catches the wrong export. The general anatomy and the other 42 skills in the repo are in the complete guide to Claude marketing skills.

Troubleshooting: when a Meta Ads skill gives bad answers

The same fatigue, overlap and frequency checks run continuously inside Ryze AI, which then applies the change with an approval step. The Meta API side of a direct connection is in MCP for the Meta Marketing API.