Ad operations · The copilot’s skills
12 Ad-Ops Skills for Claude
A skill is a focused check the AI runs on your own account data. These are the 12 we actually run across Meta, Google and GA4. Each one reads and flags one thing. The AI analyses and proposes. You decide. Nothing changes without your approval.
Skills, not a chatbot
One clever prompt helps once. A named skill is the same check, run the same way, every week, so nothing gets missed on the fourth account when you’re tired. Same Claude everyone else has. The edge is the routine built around it, because the campaign build itself stopped being the edge.
Every skill below is approval-gated by design: it reads your exports or connected data, flags what it finds, and hands you the decision. No bid, budget or campaign moves unless you say so.
A skill is also a check written down once, so you stop re-checking every answer by hand. Why that is where the hours go is in using AI on your ads, then checking every number.
The 12 skills
Or see what your team should look at first: the Paid Media Capacity Check, being built nowWasted spend
- wasted-spend. Flags every search term above a spend threshold with zero conversions in the window, ranked by wasted spend, with a negative-or-fix call for each.
- placement-drain. Finds placements taking more than a few percent of spend at a CPA above account average. Shows spend, conversions and CPA per placement.
Search terms & keywords
- search-terms. Groups search terms into themes, flags the ones irrelevant to the product, and drafts a negative-keyword list split exact vs phrase. You approve before anything is added.
Creative that converts
- creative-fatigue. Flags creatives whose CTR dropped more than a set amount versus their first week while spend stayed flat. Ranks by spend at risk.
- rsa-gaps. Reviews headlines and descriptions against the product. Flags duplicates, missing benefits and any asset rated Low, and suggests replacements.
PMax & Shopping
- pmax-inputs. Audits the four inputs you actually control (audience signals, asset quality, the feed, conversion setup) and tells you which is weakest and why.
Meta & paid social
- audience-overlap. Reads your ad sets and their targeting and flags which are likely bidding against each other, with a consolidation suggestion.
- advantage-audit. Lists every Advantage+ / automation setting enabled on the account, what each one changes, and what to check on it weekly.
Measurement
- conversion-sanity. Reviews the conversion setup and flags anything double-counted, page-view-based, or duplicated across Meta / Google / GA4 that would inflate results.
- discrepancy-check. Compares platform-reported conversions against GA4 for the same window and explains the gap by attribution model before you call it a tracking bug.
Decision & reporting
- pacing-vs-goal. Given spend-to-date, budget and ROAS target, tells you whether you’re pacing over or under and what to change first.
- weekly-read. Writes your Monday read from the exports: what changed vs last week, what needs a decision, and what’s fine, in that order. If a source has no data, it says “no data” instead of estimating.
How the copilot runs them
Think of it as one operator with a checklist, not a bot with the keys. A lead pass reads across your connected accounts. Each skill is a focused sub-check that looks at one thing and reports back. The findings roll up into a single weekly read you approve. Approval-gated end to end: nothing acts without you.
- What’s live today. The weekly read (the reporting piece) is the one skill we run in production right now. It reads Meta and Google and writes the summary. When a source hasn’t settled, it says so instead of inventing a number.
- What you run yourself. Every skill above is also a prompt you can paste into Claude or ChatGPT with your own exports, free, today. No tool required.
- What’s honest roadmap. The rest, running all twelve on a schedule across every account, is what we’re building in the open. The approval gate stays the rule: it proposes, you decide.
Questions people actually ask
How is a skill different from a prompt I saved?
A saved prompt is a good day. A skill is the same check run the same way on the fourth account when you are tired. Which, if you are honest, is the account where things actually get missed.
Do I need a paid plan or a special setup?
The same Claude everyone else has. The edge is not the model. It is having the twelve checks named, and running them in order instead of remembering which one you meant to do.
Will any of them move something in my account?
No. Every one reads and flags. No bid, budget or campaign moves unless you go and do it yourself, and that is a design decision rather than a limitation we plan to remove.
Twelve is a lot. Where do I start?
Wasted spend and search terms. Those two usually pay for the time on the first pass, and if they come back clean on your accounts, the other ten are probably not urgent this week.
Related reading. Skills para Claude, in Spanish, with how to package one as a Claude skill. The Weekly Ads Prompt Pack. The Weekly Ads Decision System. What not to automate in your Meta & Google ad accounts.
Want these running on your accounts?
The prompts are the manual version, yours to use now. We also run the weekly read across Meta, Google and GA4 as a service: same output, no exports, every decision still yours. No pitch. Tell us how many accounts you run.