Audit a Meta Ads account with AI
A prompt-by-prompt walkthrough that finds tired creative, budgets split too thin, and spend going to the wrong placements.
This is a complete Meta Ads audit you can run in Claude or ChatGPT, one prompt at a time. It assumes your Meta ad accounts are connected through Adplane, so every answer comes from your live Facebook and Instagram data, not general advice. Each step says what to look for and what to do about it, and every fix is reversible.
Meta reports results as actions, one count per type: purchases, leads, registrations, add to cart. Name the one that matters to your business and every answer gets that count and the cost per action next to spend. If you sell online with the pixel, purchases and purchase ROAS are the place to start; if you collect leads, ask for leads instead.
Step 1: the shape of the account
Fernwood Outdoor on Meta, the last 30 days, September 11 to October 10:
| Campaign | Spend (USD) | Purchases | Cost per purchase | ROAS |
|---|---|---|---|---|
| Prospecting | 3,640 | 70 | 52.00 | 1.88 |
| Retargeting | 905 | 58 | 15.60 | 6.28 |
| Fall jackets | 600 | 4 | 150.00 | 0.93 |
The account spent 5,145 for 132 purchases, 38.98 each, at a ROAS of 2.54. Prospecting takes 71% of the spend. Fall jackets costs nearly four times the account average per purchase and returns less than it spends.
You are looking for concentration and for anything that returns less than it costs. The purchase count is only as good as the pixel behind it: if a campaign shows no purchases at all, check the tracking before you judge the campaign. You should see every campaign with its spend, purchases, cost per purchase and ROAS, and the dates the numbers cover.
Step 2: budgets and delivery
Every ad set that spent in these 30 days is active and delivering. Prospecting and Retargeting budget each ad set; Fall jackets uses one campaign budget of 20 dollars a day, shared by four ad sets:
| Ad set | Campaign | Daily budget | Spend (USD) | Purchases |
|---|---|---|---|---|
| Broad interests | Prospecting | 65 | 1,940 | 25 |
| Lookalike buyers | Prospecting | 60 | 1,700 | 45 |
| Retargeting 30 days | Retargeting | 35 | 905 | 58 |
| Men's jackets | Fall jackets | shared | 210 | 2 |
| Women's jackets | Fall jackets | shared | 180 | 1 |
| Kids' jackets | Fall jackets | shared | 110 | 1 |
| Hikers interest | Fall jackets | shared | 100 | 0 |
I cannot read Meta's learning label; Ads Manager shows it in the Delivery column. By Meta's guideline of about 50 results a week to leave learning, Fall jackets has no chance: 20 dollars a day split four ways bought about one purchase per ad set in these 30 days. Two ad sets would give Meta twice the signal per ad set.
Retargeting 30 days spends about 30 a day against its 35 budget. The audience is small, so more budget would mostly show the same ads to the same people.
Too many ad sets on too little budget is a common problem in a small Meta account. The fix is fewer ad sets, not more money. You should see each ad set's daily budget, or "shared" where the campaign holds it, next to what it spent.
Step 3: creative fatigue
Weekly, from the week of September 12 to the week of October 3 to 9:
| Ad | Frequency | CTR |
|---|---|---|
| Tent bundle carousel | 1.6 to 3.0 | 2.4% to 1.1% |
| Summit tent image | 1.3 to 2.4 | 1.6% to 0.9% |
| Rain shell image | 1.4 to 2.2 | 1.8% to 1.7% |
Tent bundle carousel, in Retargeting 30 days, is tired: the same people see it almost twice as often and click half as much. Summit tent image, in Broad interests, is heading the same way. Rain shell image is holding up. I would pause the carousel and give Retargeting 30 days a new ad.
Rising frequency with falling CTR over two or three weeks is creative fatigue. More budget does not fix it; new creative does. Ask for weekly rows, not one 30-day total, or the trend disappears. You should see where each ad started and ended, so the direction is plain.
Step 4: where the money goes
| Placement | Spend (USD) | Purchases | Cost per purchase |
|---|---|---|---|
| Facebook feed | 1,420 | 34 | 41.76 |
| Instagram feed | 880 | 19 | 46.32 |
| Instagram Reels | 610 | 11 | 55.45 |
| Facebook Reels | 340 | 4 | 85.00 |
| Audience Network | 390 | 2 | 195.00 |
The feeds do most of the work. Audience Network took 390 dollars for 2 purchases. I can change an ad set's name, status, budget and bid, but not its placements, so removing Audience Network from these ad sets is a change to make in Ads Manager.
The same question works by age and gender: ask for Prospecting split by age, or by gender, and look for groups that spend without buying. Placements and demographics are where Meta's automatic choices can drift from what pays. You should see the placements add up to the campaign's spend and purchases.
Step 5: make the fixes
Done, one at a time. Read back from Meta: all three are paused. Fall jackets keeps its 20 dollar daily budget, now shared by Men's jackets and Women's jackets.
Lookalike buyers is your cheapest prospecting at 37.78 per purchase, and at a frequency of 1.8 it has room to reach more people. Raise it from 60 to 75 dollars a day? Say yes and I will make the change.
Done. Read back from Meta: Lookalike buyers now has a daily budget of 75 dollars, up from 60.
The readiness check passed: the account has a payment method and a Page. I uploaded the image, made a creative from the Fernwood Outdoor Page with a Shop now button, and added the ad Fall tents image to Retargeting 30 days. Read back from Meta: Fall tents image is stored with its creative, text and link.
Created pausedHere is how it looks in the Instagram feed: (preview link). Want to change the text before you turn it on?
Pauses are the safe first move: they stop spend and can be undone at any time. Move budget in steps of about a quarter, since a large jump can send an ad set back into learning. A raise the assistant suggests waits for your yes; a raise you ask for in your own words, such as "raise Lookalike buyers to 75 dollars a day", is applied as asked. You should see each change read back from Meta, and the new ad's preview before anything runs.
Step 6: the write-up
Found. Prospecting takes 71% of spend at 52.00 per purchase. Fall jackets returned 0.93 on 600 dollars, with its budget split four ways. Tent bundle carousel and Summit tent image are fatiguing. Audience Network took 390 dollars of Prospecting for 2 purchases.
Changed. Paused Tent bundle carousel, Kids' jackets and Hikers interest. Lookalike buyers raised from 60 to 75 dollars a day. A new ad, Fall tents image, waiting in Retargeting 30 days for your review.
Check on October 25. Lookalike buyers' cost per purchase at the new budget, Fall jackets with two ad sets, frequency on Retargeting 30 days once the new ad runs, and a replacement for Summit tent image. Remove Audience Network in Ads Manager if you have not yet.
Keep the summary in the conversation: next time, start with it and ask what changed since.
Why this is safe to run
The reporting steps cannot change anything. The fixes go through Adplane's write tools. A new campaign, ad set or ad cannot deliver until you turn it on, and one update tool (meta_update_object) is the single path to turning anything on or raising a budget. Every change is read back from Meta before it is reported as done. The full safety model is in the docs.
Run it on your account
Adplane connects to Claude in one click from the connector directory, or to ChatGPT from the plugin directory. Then connect Meta Ads on your accounts page and sign in with Facebook. The free trial is 14 days and does not require a card. Connect your account and run step 1.
Try it in your own chat
Sign in with Google, connect your ad accounts, and ask your first question.