How to run Meta ads with Claude

Pull my last 30 days at account level and give me the baseline I should be judging everything against from now on.

I want total spend, results, cost per result, ROAS, CTR, CPM and frequency, then the same breakdown per campaign, sorted by spend. Add a column for what share of total spend each campaign took, because that is usually where the surprise is.

Then tell me in plain language which three campaigns are carrying the account and which three are being carried, and flag anything where the sample is too small to draw a conclusion from yet rather than pretending it means something.

Find everything quietly wasting money in this ad account over the last 14 days.

Specifically: ad sets that have spent more than the cost of one conversion with zero conversions, ads with a frequency above 3.5 where CTR has fallen week on week, campaigns where cost per result has risen more than 25% versus the previous 14 days, and any audience overlap where I am bidding against myself.

Rank what you find by dollars at risk per week, not by how bad the percentage looks, and for each one tell me the specific action you would take and what you would expect to happen to the account total if I took it.

Audit my tracking setup and tell me whether I can trust the numbers I am looking at.

Check pixel and Conversions API event coverage, event match quality, whether any events are firing twice or not at all, and whether any of my key conversion events have dropped in volume in a way that is not matched by a drop in traffic.

For anything you flag, explain what it is doing to my reported results and my optimisation, tell me how serious it is on a scale I can act on, and give me the fix in the order that matters most. If everything is healthy, say so plainly instead of finding something to worry about.

Take my best performing ad from the last 30 days and my worst performing ad at similar spend, and tell me what is structurally different between them rather than just restating the numbers.

Compare them on the things you can actually see: the ad copy angle, the objective, the placement mix, the audience definition, the frequency curve, the click-to-conversion drop-off, and the day the gap between them opened up. Tell me which single difference you think explains most of the gap, and how confident you are in that on a scale of low, medium or high.

Then write the hypothesis as a testable sentence I could put into my next round of ads, and name the one metric I should watch to know within a week whether the hypothesis was right. You cannot see the images or video, so say clearly which part of your explanation is inference rather than something the data shows.

Based on the last 14 days, write me a staged budget plan that moves money out of the losers and into the winners without breaking the learning phase.

Rules: no single ad set changes by more than 20% in one move, nothing scales on fewer than 50 conversions in the window, and total account spend stays flat unless you make an explicit case for raising it.

Give it to me as a table of the exact changes, in the order I should make them, with the date to make each one. Under the table, tell me what you expect the account to look like in seven days if it works, and the specific signal that would tell me it is not working so I know when to stop.

Build me a pause list for this ad account based on the last 14 days. My target ROAS is [your number] and my margin means I lose money below [your break-even number].

Include every ad set that is below target on a fair sample, and for each one give me the name, the spend in the window, the conversions, the actual ROAS, and how many days it has been under target. Exclude anything still in the learning phase, anything with fewer than 50 conversions in the window, and anything where the drop is a single bad day dragging the average down, and tell me what you excluded and why so I can overrule you.

Do not pause anything yet. Show me the list, tell me how much weekly spend it frees up, and wait. When I reply with the word CONFIRM plus the specific names I approve, pause exactly those and nothing else, then read the account back to me so I can see the result.

Build me a new campaign in this ad account, and leave everything paused so I can review it before anything spends.

The offer is [describe your offer], the customer is [describe who you are targeting and what they already believe], the objective is [conversions / leads / traffic], the daily budget is [amount], and the landing page is [URL].

Structure it the way you would if you were spending your own money: tell me how many ad sets and why, what the targeting is on each, and how the budget splits. Write three ad copy variations per ad set that test genuinely different angles rather than reworded versions of the same one, and tell me what each angle is testing. Then list exactly what I need to check in Ads Manager before I turn any of it on.

Write my weekly ads report, the version I would send to a client or put in front of my own team.

Open with the three numbers that matter most this week and what each one did versus last week. Then: the top three and bottom three ads by cost per result, what changed and what caused it, what I did last week and whether it worked, and the three things I am doing this week with a one-line reason for each.

Write it in plain language with no jargon and no filler, put the bad news near the top rather than buried, and keep the whole thing under 400 words. If a number moved for a reason you cannot identify from the data, say that instead of inventing a story for it.

Build me a reusable Claude skill called weekly-ads-review that runs my whole Meta ads review through the connected ads MCP, and use skill-creator to package it.

The description should say: use this when I ask for my weekly ads review, ads report or account check.

The steps are: pull the last 7 and previous 7 days at account and campaign level, compare them, then run the waste check (spend with no conversions, frequency above 3.5 with falling CTR, cost per result up more than 25%), then check pixel and Conversions API signal health, then produce a plain-language report under 400 words with the three headline numbers, the top and bottom three ads, and three recommended actions ranked by dollars at stake.

Two hard rules in the skill: never make a change to the account, only recommend, and always state the date range and the level for every number so I can reconcile it against Ads Manager. If a metric is missing or the sample is too small, say so instead of estimating.