---
title: "Run a Meta Ads Audit in Claude—and Approve the Fixes"
description: "Run six Meta Ads checks in Claude through groas, verify findings against your records, and approve account fixes with guardrails and an undo plan."
url: "https://groas.com/post/a-meta-ads-audit-from-claude-what-it-checks-and-what-it-can"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/ac9fa0f1-25d1-445e-992b-f12fad0a1d6e.png"
published: "2026-11-19T15:50:48.408Z"
modified: "2026-11-19T15:50:48.504Z"
---

November 19, 2026 · 11 min read

# Run a Meta Ads Audit in Claude—and Approve the Fixes

[](https://groas.com/author/alexander-perelman)

[Alexander Perelman](https://groas.com/author/alexander-perelman)Personally Tested

Head Of Product @ groas

[LinkedIn](https://www.linkedin.com/in/alexander-433793253/)

![Tiny workers tag coin-leaking pipes in a miniature plumbing model and wait as a giant hand lowers an approval stamp before any fix is made.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/ac9fa0f1-25d1-445e-992b-f12fad0a1d6e.png)

In this article

1. [Setup: connect the account and define the control](#setup-connect-the-account-and-define-the-control)
2. [Test 1: check purchase signals before judging performance](#test-1-check-purchase-signals-before-judging-performance)
3. [Test 2: give every ad set a job and enough budget](#test-2-give-every-ad-set-a-job-and-enough-budget)
4. [Test 3: compare each creative with its own baseline](#test-3-compare-each-creative-with-its-own-baseline)
5. [Test 4: find the delivery segment hidden by the average](#test-4-find-the-delivery-segment-hidden-by-the-average)
6. [Test 5: check budget stability before touching bid caps](#test-5-check-budget-stability-before-touching-bid-caps)
7. [Test 6: use rival creative as a hypothesis, not proof](#test-6-use-rival-creative-as-a-hypothesis-not-proof)
8. [Turn the findings into an approval-ready fix list](#turn-the-findings-into-an-approval-ready-fix-list)
9. [Approve one change, read it back, keep the log](#approve-one-change-read-it-back-keep-the-log)
10. [Judge the result without confusing attribution with sales](#judge-the-result-without-confusing-attribution-with-sales)
11. [Start with the connection, finish with a decision](#start-with-the-connection-finish-with-a-decision)

A Meta audit that ends in a PDF changes nothing in the account. I want a fix list Claude can act on, using real account data, with nothing applied until I approve it.

That is the question behind this protocol: **can Claude identify wasted spend in a Meta account and apply only the fixes you approve?** Connect Claude to Meta through groas, run six tests, then turn verified findings into controlled changes. Tracking comes first. Rival creative comes last. No screenshot parade in between.

## Setup: connect the account and define the control

I would start with one business question: which part of this account appears to waste money, and what happens to cost per result after an approved fix?

[Plug groas into Claude](https://groas.com/claude) over MCP, then connect your Meta ad account and website. The groas MCP endpoint is `https://mcp.groas.ai/mcp`. This puts account data, competitor ads, funnels and Reddit context in the same chat rather than scattered across tabs.

Before running the tests, give Claude four instructions:

1. **State the goal.** For example: “We spend $20k a month to generate purchases at a $45 CPA.” Those are illustrative numbers; use your own.
2. **Set the window.** Use the last 14 to 28 days, not yesterday.
3. **Stay read-only.** Ask for findings and proposed changes, not execution.
4. **Expose gaps.** Require Claude to say when a requested field or diagnostic is unavailable rather than infer an answer.

**Verify the figures behind every proposed fix against Ads Manager before approving it.** That is the control. A live connection avoids selective pasting, but it does not remove the need to check the answer.

An [export can support a one-off review; a live connection supports action](https://madgicx.com/answers/use-claude-for-meta-ads). I use groas as the setup for this protocol because the account and market context sit together. Keep the approval rule just as close.

![Lab notebook with a Meta ad account under a magnifying glass](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/bc5d80ce-059f-4a5c-b8e3-49c622b27897.png)

## Test 1: check purchase signals before judging performance

**Question:** Is Meta optimizing toward a usable purchase signal, or counting noise?

Ask Claude:

> Check our pixels and whether purchase events dropped or doubled. List each available dataset, active event, browser versus server source, and possible duplicate. Show daily event counts for the last 14 days. Mark anything you cannot inspect.

I expect tracking to be an early source of trouble. The mechanism is boring and expensive: Meta bids toward the event it counts. If that event fires twice or stops firing, the bidder receives the wrong lesson. Creative analysis cannot repair that.

Check the correct Pixel or dataset, the relevant `ViewContent`, `AddToCart`, `InitiateCheckout` and `Purchase` or `Lead` events, and browser versus Conversions API activity. For events sent through both sources, inspect the IDs used to [deduplicate browser and server events](https://ali-raza-ppc.github.io/blog/meta-ads-audit-checklist/).

Then compare a full week with your shop or CRM. **Use that comparison to investigate discrepancies, not demand identical totals.** [Shopify records orders while Meta attributes conversions to ads](https://www.kunads.ai/answers/why-meta-and-shopify-numbers-dont-match). Attribution alone can create a gap; a gap is not proof of broken tracking.

I would look for unexplained drops, doubled events and mismatched event IDs. If the evidence points to duplicate purchases, stop here and investigate deduplication before interpreting downstream CPA. If Claude cannot inspect the necessary data, record the test as unresolved, not passed.

![Tangled tracking wires and duplicate purchase receipts spilling from a laptop](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/76d1c2bb-a385-4106-ad91-1bf57a56b7de.png)

## Test 2: give every ad set a job and enough budget

**Question:** Does each ad set serve a distinct purpose, or does the structure divide the same job into underfunded pieces?

Ask Claude:

> List every active campaign and ad set, budget, optimization event, audience and Learning status. Flag Learning Limited, duplicate audiences and mismatched optimization events.

My expectation is that fragmentation will explain more than an elaborate naming convention does. Meta needs about 50 optimization events per ad set per week to exit learning. Dividing $100 a day across eight thin ad sets makes that difficult. The spreadsheet looks organized. Delivery gets crumbs.

Review three things:

- **Role:** Is this prospecting, retargeting or testing?
- **Signal:** Does its optimization event match the business outcome?
- **Resources:** Does its budget support the event volume it needs?

The [recurring structural red flags](https://ali-raza-ppc.github.io/blog/meta-ads-audit-checklist/) are small ad sets stuck in Learning Limited, unclear roles, and optimization for link clicks when the business wants purchases.

**Treat duplicate jobs as consolidation candidates, not automatic merge orders.** If Claude finds three broad-audience ad sets at $20 a day each, ask what distinguishes them before approving a consolidated structure. Record the budgets, events and audiences that would change.

I expect consolidation to help when fragmentation is the constraint. The test is whether the proposed structure fixes that constraint, not whether it contains fewer rows.

## Test 3: compare each creative with its own baseline

**Question:** Which established ads are deteriorating, and which still deserve budget?

Ask Claude:

> For each active ad with 14 or more days live, show 7-day frequency, CTR, cost per result and Delivery flag versus its own first 7 to 14 days. Flag decay rather than rank ads by current CPA alone.

My expectation is that a few established winners will deserve closer inspection. Repeated exposure can coincide with falling engagement and rising acquisition costs. Comparing an ad with itself helps separate deterioration from the fact that two ads started from different places.

Use the [creative-fatigue warning ranges](https://aiseoshift.com/blog/creative-fatigue/) as investigation flags:

- Seven-day frequency above 3 to 4 on cold prospecting; retargeting can tolerate more.
- CTR down 20 to 30% against the ad’s own baseline.
- CPA up 15% or more over several days.

**No single threshold earns a pause.** Read frequency, CTR, CPA, reach and Delivery status together. A fatigue label is not an early-warning system, and a rising CPA alone does not identify the cause.

Tell Claude to classify each ad as _scale_, _watch_ or _replace_, with the numbers behind the label. Pause nothing yet. The audience and placement check comes next, and it may change the diagnosis.

## Test 4: find the delivery segment hidden by the average

**Question:** Which slice of delivery pulls CPA upward while the account average conceals it?

Ask Claude:

> Break down the last 28 days by age, gender, placement and platform where available. Show spend share, purchase share and cost per result for each row. Identify small samples explicitly.

I expect a segment-level change to explain some apparent creative problems. [Breakdowns separate results by dimensions such as age, placement and device without changing delivery](https://www.circlify.ai/blog/meta-ads-breakdowns-report). That lets the audit inspect the mix without first building a new ad set for every slice.

Compare share of spend with share of purchases. Do not let an impressive ROAS on four orders run the meeting. If Audience Network spend doubled while purchases stayed flat, investigate the delivery shift before declaring the ad exhausted.

**Keep small rows out of the fix list until they earn a decision.** Leave a thin segment alone for another week and record the trend. Ask Claude whether the finding changes the creative diagnosis from Test 3.

The control is structural: do not split a consolidated ad set into six micro ad sets to chase a 7-day blip. That reverses Test 2 and recreates the problem the audit is supposed to remove.

## Test 5: check budget stability before touching bid caps

**Question:** Do budget and bidding decisions support the goal, or keep interrupting delivery?

Ask Claude:

> For each active ad set, show daily budget, bid strategy, days since the last edit, optimization events in the last 7 days, Learning status, and cost per result versus my break-even CPA. Flag spending below 80% of budget and edits within the last 3 days.

My expectation is that the change history will explain more than the cleverness of the bid settings. Material budget or bid changes can send an ad set back into learning. An account full of edits can look carefully managed while making its own performance harder to interpret.

Compare stable ad sets with frequently edited ones. Check whether under-spending, thin event volume and Learning Limited appear together. Do not assume that every struggling ad set needs more budget simply because it spends what it receives.

A passing candidate looks boring: a clear job, a steady budget, an appropriate bid strategy and enough unchanged delivery to assess it. Leave that candidate alone.

**No numbers behind a budget recommendation, no approval.** If Claude proposes a 20% increase, require the 7-day spend, event count, CPA and Learning status behind it. Resolve structural fragmentation before experimenting with bid caps.

## Test 6: use rival creative as a hypothesis, not proof

**Question:** What are rivals running, and which part belongs in the next controlled creative test?

Ask Claude:

> List rival ads by start date and hook, show the landing page behind each, and compare their offer and promise with ours sentence by sentence. Flag claims our page cannot support.

My expectation is that one long-running angle will deserve a test while others deserve ignoring. [Meta’s Ad Library shows active ad creative, copy, start dates and platforms](https://www.usecarly.com/blog/meta-ads-mcp/). For ordinary commercial ads, it does not show spend, impressions or results.

**Longevity is a clue, not a performance report.** Through groas, Claude can bring rival ads, funnels and Reddit language beside your account numbers. The useful comparison is your CPA next to their hook and landing-page promise. It is not an invented estimate of their ROAS.

![Rival ad cards beside a faded creative, illustrating hypotheses rather than proof of performance](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/ec85ea2a-9adf-48f0-9561-e7b270eff6d2.png)

Rate each candidate on three points:

- How long has it run?
- How specific is the hook?
- Does the landing page deliver on the promise?

A specific hook, a proof point and a matching page make a useful test candidate. Add one angle to the queue. Do not copy a rival’s claim unless your own page can support it.

## Turn the findings into an approval-ready fix list

After the six tests, ask Claude:

> Give me fixes sorted by expected cost impact. For each, state the account object, current value, proposed value, supporting numbers, guardrail and undo step. Separate unresolved diagnostics from actionable findings. Do not apply anything yet.

That last sentence matters. **A recommendation is not permission.** Claude through groas can apply approved account changes, including budgets, status and structure. Keep the proposed action narrow enough to verify afterward.

For this protocol, I would divide the work like this:

| Apply through Claude after approval         | Keep outside this execution pass       |
| ------------------------------------------- | -------------------------------------- |
| Consolidate or pause overlapping ad sets    | Pixel and Conversions API code changes |
| Adjust ad set budgets                       | Landing-page copy and checkout changes |
| Pause fatigued ads or launch approved ads   | New creative concepts and brand claims |
| Change audience exclusions and bid settings | Budget guardrails and final sign-off   |

This is a boundary for the test, not a claim that every other task is impossible. Tracking code, checkout changes and new promises deserve their own review. Mixing them into a budget cleanup makes the result harder to interpret and the mistakes harder to contain.

groas is [flat at $19 for a whole team](https://groas.com/). Every change is logged with its reason and can be undone in one step. Autopilot stays off until switched on; leave it off for this approval-by-approval protocol.

![Claude review checklist in groas with approved account actions and manual review items](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/0d3e2d8b-3a7b-4a7a-88ad-057626af2947.png)

## Approve one change, read it back, keep the log

An audit that changes an account without a record replaces one problem with another. Before execution, set three guardrails:

1. A daily spend ceiling.
2. A pause threshold that triggers a request for your approval, such as spending twice the target CPA with zero purchases.
3. A review or stop date if performance does not improve.

**Keep every action behind explicit approval.** Do not quietly turn the pause threshold into permission for unattended changes.

Approve a specific item: `Apply fix 2 now, read back the new budget and status, and log the reason.` If Claude cannot read back the new value from Ads Manager, treat the change as unverified. Do not build the next decision on it.

Keep the log with the previous value, new value, timestamp, CPA and event count before the change. The [reason and one-step undo](https://groas.com/) belong beside the action, not in somebody’s memory.

Re-run Tests 1 and 5 after 7 days. Check whether the signal remains usable and whether delivery addresses the original finding. If the problem persists, review the change and its undo step before adding another intervention. The point is an interpretable sequence, not a busy change log.

## Judge the result without confusing attribution with sales

**Keep the reporting basis consistent before calling a win.** [Meta attributes conversions under its attribution setting](https://www.kunads.ai/answers/why-meta-and-shopify-numbers-dont-match), including view-through and modeled conversions. Your shop records orders. Refunds, tax, shipping and attribution-setting changes can also move the comparison.

Use Meta-versus-Meta comparisons under the same attribution window for campaign decisions, and shop blended ROAS for the business decision. If Claude reports a win that your CRM does not confirm, do not scale on the Meta figure alone. Revisit Test 1 and the attribution difference before deciding what improved.

That discrepancy is a finding, not an invitation to choose the prettier dashboard. The PDF version would leave a recommendation to scale. This protocol should leave a specific question to resolve first.

## Start with the connection, finish with a decision

[Connect Meta Ads to Claude through groas](https://groas.com/post/connect-meta-ads-to-claude-the-setup-the-first-ten-prompts), set the spend ceiling and run the six prompts in order. **Approve only fixes with supporting numbers, a guardrail and an undo step.**

If the tests identify a supported fix, make it and measure what changes. If they find nothing actionable, do not manufacture account edits to justify the audit. Move the next controlled creative test forward instead. If key data is unavailable, the next job is closing that gap, not declaring the account clean.

Either way, the result changes the next decision: fix a demonstrated problem, test a new hypothesis or collect the missing evidence. Bring that fix list to your next review, not the PDF.

## Frequently Asked Questions

### How can I tell whether Meta is counting purchase events correctly?

Compare Meta’s daily purchase events with your shop or CRM records, and check for unexplained drops, doubled events, and mismatched browser and server event IDs. A difference between Meta and your shop does not by itself prove tracking is broken, because attribution can create a gap. If the required data is unavailable, record the check as unresolved.

### When should I consider consolidating Meta ad sets?

Consider consolidation when multiple ad sets have overlapping audiences or duplicate jobs and limited event volume. Meta needs about 50 optimization events per ad set per week to exit learning, but overlapping ad sets should not be merged automatically: first check their roles, optimization events, audiences, and budgets.

### How do I check whether a Meta ad is showing signs of creative fatigue?

Compare an established ad’s frequency, CTR, cost per result, reach, and Delivery status with its own earlier performance. Frequency above 3 to 4 on cold prospecting, CTR down 20 to 30% from baseline, or CPA up 15% or more over several days are investigation flags, not automatic reasons to pause.

### How should I use Meta age, placement, and platform breakdowns in an audit?

Compare each segment’s share of spend with its share of purchases and cost per result to see whether a delivery shift may explain the account average. Treat small samples cautiously: leave a thin segment alone for another week and record the trend rather than making it a fix-list item.

### What should I check before changing a Meta ad set’s budget or bid strategy?

Review the ad set’s recent edits, daily budget, bid strategy, optimization events, Learning status, and cost per result against your break-even CPA. A budget recommendation should include supporting figures such as 7-day spend, event count, CPA, and Learning status; resolve structural fragmentation before experimenting with bid caps.

### Can Meta Ad Library show which rival ads are performing well?

No. Meta’s Ad Library shows active ad creative, copy, start dates, and platforms for ordinary commercial ads, but it does not show spend, impressions, or results. Treat an ad’s longevity as a clue for a possible test, then check whether its hook and landing-page promise are relevant and supportable for your business.

### How should I approve and verify a Meta Ads change suggested by Claude?

Approve one specific change at a time, with supporting numbers, a daily spend ceiling, a pause threshold that requests approval, and a review or stop date. After the change, read back the new value from Ads Manager; if Claude cannot verify it, treat the change as unverified. Keep the previous and new values, timestamp, reason, and undo step in the log.

### How should I compare Meta results with shop or CRM sales after an account change?

Use Meta-versus-Meta comparisons with the same attribution window for campaign decisions, and shop blended ROAS for the business decision. Meta may include view-through and modeled conversions, while shop records orders; if the CRM does not confirm a reported Meta win, do not scale based on the Meta figure alone.

## Run Your Marketing On groas

Plug ChatGPT, Claude or any AI agent into groas and let it do the work of a whole marketing team: $19 a month flat with every feature included, and autopilot whenever you’re ready.

[Start For $19](https://groas.com/pricing)

[Or have groas run it for you](https://groas.com/done-for-you)

## Structured data

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://graph.groas.com/entity/groas","name":"groas","url":"https://groas.com/","description":"groas is the marketing infrastructure for your own AI: it connects your AI to all your marketing and gives it proprietary marketing data collected 24/7. Plug Claude, ChatGPT or any AI agent into groas and run all your marketing from it, for a tiny fraction of the usual cost, with the groas app there whenever you want to see more. It is the layer ChatGPT, Claude and any AI assistant run a business’s online marketing on. Through groas an assistant can do anything you could do natively in Google Ads, Bing Ads, Meta Ads and ChatGPT ads, on your website and in AI search, and, where the owner switches it on, can run connected campaigns by itself, so AI agents do the work a whole marketing team used to. groas also gives those agents large-scale proprietary marketing data it collects around the clock, on any schedule the owner chooses: every ad competitors run on Google and Meta, their landing pages and full funnels with the offers and prices in them, Reddit customer research, AI search answers from ChatGPT, Google AI Overviews, Gemini, Claude and Perplexity, and SEO data from Semrush and Ahrefs, backlinks included ($19 a month, flat, with everything included, for unlimited teammates on one shared workspace and unlimited ad accounts and websites, delivered as an MCP server). Separately, its done-for-you service has senior people and proprietary quantitative systems run paid ads, SEO and AI search for businesses, and white label for agencies, from $5,000 a month.","logo":{"@type":"ImageObject","url":"https://groas.com/icon-512.png","width":512,"height":512},"sameAs":["https://www.linkedin.com/company/groas/"]},{"@type":"WebSite","@id":"https://groas.com/#website","name":"groas","url":"https://groas.com/","inLanguage":"en-US","publisher":{"@id":"https://graph.groas.com/entity/groas"}},{"@type":"WebPage","@id":"https://groas.com/post/a-meta-ads-audit-from-claude-what-it-checks-and-what-it-can#webpage","url":"https://groas.com/post/a-meta-ads-audit-from-claude-what-it-checks-and-what-it-can","name":"Run a Meta Ads Audit in Claude—and Approve the Fixes","description":"Run six Meta Ads checks in Claude through groas, verify findings against your records, and approve account fixes with guardrails and an undo plan.","inLanguage":"en-US","isPartOf":{"@id":"https://groas.com/#website"},"about":{"@id":"https://graph.groas.com/entity/groas"},"primaryImageOfPage":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/ac9fa0f1-25d1-445e-992b-f12fad0a1d6e.png"},"breadcrumb":{"@id":"https://groas.com/post/a-meta-ads-audit-from-claude-what-it-checks-and-what-it-can#breadcrumb"},"dateModified":"2026-11-19T15:50:48.504Z","mainEntity":{"@id":"https://groas.com/post/a-meta-ads-audit-from-claude-what-it-checks-and-what-it-can#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/a-meta-ads-audit-from-claude-what-it-checks-and-what-it-can#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://groas.com/"},{"@type":"ListItem","position":2,"name":"Blog","item":"https://groas.com/blog"},{"@type":"ListItem","position":3,"name":"Run a Meta Ads Audit in Claude—and Approve the Fixes","item":"https://groas.com/post/a-meta-ads-audit-from-claude-what-it-checks-and-what-it-can"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/a-meta-ads-audit-from-claude-what-it-checks-and-what-it-can#article","url":"https://groas.com/post/a-meta-ads-audit-from-claude-what-it-checks-and-what-it-can","mainEntityOfPage":{"@id":"https://groas.com/post/a-meta-ads-audit-from-claude-what-it-checks-and-what-it-can#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"Run a Meta Ads Audit in Claude—and Approve the Fixes","description":"Run six Meta Ads checks in Claude through groas, verify findings against your records, and approve account fixes with guardrails and an undo plan.","datePublished":"2026-11-19T15:50:48.408Z","dateModified":"2026-11-19T15:50:48.504Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/ac9fa0f1-25d1-445e-992b-f12fad0a1d6e.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"claude, ads, meta, account, run, fixes, meta ads, test","wordCount":2261,"inLanguage":"en-US"},{"@type":"Person","@id":"https://groas.com/author/alexander-perelman#person","name":"Alexander Perelman","jobTitle":"Head Of Product @ groas","description":"My background includes Goldman Sachs and Stanford Computer Science.","image":"https://groas.com/media/blog/4f17cac3a81acc224cc1b2cfab8f94bbc295a3089d13e8421125771ef8d4064f.jpg","worksFor":{"@id":"https://graph.groas.com/entity/groas"},"sameAs":["https://www.linkedin.com/in/alexander-433793253/"]},{"@type":"FAQPage","@id":"https://groas.com/post/a-meta-ads-audit-from-claude-what-it-checks-and-what-it-can#faq","mainEntity":[{"@type":"Question","name":"How can I tell whether Meta is counting purchase events correctly?","acceptedAnswer":{"@type":"Answer","text":"Compare Meta’s daily purchase events with your shop or CRM records, and check for unexplained drops, doubled events, and mismatched browser and server event IDs. A difference between Meta and your shop does not by itself prove tracking is broken, because attribution can create a gap. If the required data is unavailable, record the check as unresolved."}},{"@type":"Question","name":"When should I consider consolidating Meta ad sets?","acceptedAnswer":{"@type":"Answer","text":"Consider consolidation when multiple ad sets have overlapping audiences or duplicate jobs and limited event volume. Meta needs about 50 optimization events per ad set per week to exit learning, but overlapping ad sets should not be merged automatically: first check their roles, optimization events, audiences, and budgets."}},{"@type":"Question","name":"How do I check whether a Meta ad is showing signs of creative fatigue?","acceptedAnswer":{"@type":"Answer","text":"Compare an established ad’s frequency, CTR, cost per result, reach, and Delivery status with its own earlier performance. Frequency above 3 to 4 on cold prospecting, CTR down 20 to 30% from baseline, or CPA up 15% or more over several days are investigation flags, not automatic reasons to pause."}},{"@type":"Question","name":"How should I use Meta age, placement, and platform breakdowns in an audit?","acceptedAnswer":{"@type":"Answer","text":"Compare each segment’s share of spend with its share of purchases and cost per result to see whether a delivery shift may explain the account average. Treat small samples cautiously: leave a thin segment alone for another week and record the trend rather than making it a fix-list item."}},{"@type":"Question","name":"What should I check before changing a Meta ad set’s budget or bid strategy?","acceptedAnswer":{"@type":"Answer","text":"Review the ad set’s recent edits, daily budget, bid strategy, optimization events, Learning status, and cost per result against your break-even CPA. A budget recommendation should include supporting figures such as 7-day spend, event count, CPA, and Learning status; resolve structural fragmentation before experimenting with bid caps."}},{"@type":"Question","name":"Can Meta Ad Library show which rival ads are performing well?","acceptedAnswer":{"@type":"Answer","text":"No. Meta’s Ad Library shows active ad creative, copy, start dates, and platforms for ordinary commercial ads, but it does not show spend, impressions, or results. Treat an ad’s longevity as a clue for a possible test, then check whether its hook and landing-page promise are relevant and supportable for your business."}},{"@type":"Question","name":"How should I approve and verify a Meta Ads change suggested by Claude?","acceptedAnswer":{"@type":"Answer","text":"Approve one specific change at a time, with supporting numbers, a daily spend ceiling, a pause threshold that requests approval, and a review or stop date. After the change, read back the new value from Ads Manager; if Claude cannot verify it, treat the change as unverified. Keep the previous and new values, timestamp, reason, and undo step in the log."}},{"@type":"Question","name":"How should I compare Meta results with shop or CRM sales after an account change?","acceptedAnswer":{"@type":"Answer","text":"Use Meta-versus-Meta comparisons with the same attribution window for campaign decisions, and shop blended ROAS for the business decision. Meta may include view-through and modeled conversions, while shop records orders; if the CRM does not confirm a reported Meta win, do not scale based on the Meta figure alone."}}]}]}
```
