A client hands you a Meta ad account and asks, ‘What would you fix?’ ChatGPT can help assemble the evidence, compare performance and write the report. The useful output is a short list of decisions supported by account data. A confident paragraph without dates, ad IDs or a defined conversion is not an audit.
This playbook covers the first audit of a client's paid Facebook and Instagram campaigns. Use it for agency onboarding, a second opinion or a monthly account review. Facebook and Instagram advertising can run through the same Meta ad account; inspect the platform breakdown before treating them as separate budgets.
What is a Facebook and Instagram ads audit?
A Facebook and Instagram ads audit is a structured review of a Meta ad account's measurement, campaign setup, spending, audiences, placements and creatives against a business goal. It should explain what is happening, which conclusions the evidence supports, and what to check or change next. ChatGPT helps organize that work when you provide the data.
1. What do you need before auditing a client's Meta ads?
Agree on the business question before opening a report. A lead-generation client needs qualified enquiries and sales; an online store needs orders and revenue evaluated against its margins. A low cost per click does not answer either question.
- Account and authorization: the client's Meta ad account ID, your authorized access and permission to use its data with the selected tools. Keep each client in a separate conversation.
- Goal and target: the conversion event, acceptable cost per qualified lead or acquisition, and any revenue target. Ask for the client's target rather than inventing an industry benchmark.
- Comparable periods: start with the last 28 complete days versus the previous 28, excluding today. Extend the window for low-volume accounts and note promotions, budget changes or conversion delays.
- Measurement context: account currency, time zone, attribution settings and where conversions happen: a website, an Instant Form, messaging or another destination.
- Business evidence: aggregate CRM counts of qualified leads, booked appointments or sales; for ecommerce, orders, refunds and margins where available. Customer names and contact details are unnecessary for this comparison.
Keep missing inputs visible. If the client has no qualified-lead counts, the audit can report Meta's recorded leads but cannot establish cost per qualified lead.
2. How do you give ChatGPT access to Facebook and Instagram ads?
Connect the Meta ad account through LoomaScale
Follow the ChatGPT to Meta Ads connection guide, authorize Meta through its sign-in flow, and select the client's account. LoomaScale gives ChatGPT tools to retrieve campaign, ad set and ad-level performance and supported breakdowns. Ask it to confirm the account ID, currency and dates before analyzing anything.
The connection guide covers setup; this article covers the audit and client deliverable. If you are choosing a connector, the Meta Ads MCP server comparison explains the alternatives.
Keep this workflow diagnostic: ask for analysis and recommendations only. ‘Do not change anything’ is an instruction to the assistant, not a read-only permission setting. LoomaScale also has tools that can change campaigns, so review proposed changes separately.
Or upload Meta Ads Manager exports
Export campaign, ad set and ad-level tables for both periods, plus a separate platform and placement report where supported. Include IDs as well as names, date boundaries, spend, impressions, reach, the selected conversion event and its value where relevant. Label each file's reporting level and attribution settings. Upload the files to ChatGPT and ask it to list the columns and missing fields first. OpenAI’s data-analysis guide documents spreadsheet analysis and recommends reviewing the calculations and assumptions.
A CSV audit is a snapshot: ChatGPT cannot retrieve an omitted metric or a newer result from the file. Do not stack campaign, ad set and ad totals into one sum; those are different views of the same spending. For a file-based audit, replace the retrieval instructions in the following prompts with ‘Use only the attached exports.’
3. Which ChatGPT prompt starts the client audit?
Replace the bracketed inputs before running this prompt.
Audit my client's Facebook and Instagram ads in Meta ad account [ACCOUNT_ID]. Analysis only: do not create, edit, pause or activate anything. Business goal: [GOAL]. Primary conversion event: [EVENT]. Client target: [TARGET OR UNKNOWN]. Compare [START–END] with [PREVIOUS START–END], using complete days in the account time zone. First confirm account ID, currency, time zone, reporting dates and attribution settings where available. List missing information. Retrieve performance at account, campaign, ad set and ad level. Include entities that spent during either period, even if now paused. Show spend, impressions, reach, frequency, clicks with their exact definition, the selected conversion count, cost per conversion, and purchase value/ROAS where applicable and returned. Keep action types separate. Complete pagination where supported and flag any truncated or incomplete report before drawing conclusions. Retrieve campaign objectives and ad-set destination, optimization event, budget and targeting/exclusion settings where supported. If unavailable, mark those checks incomplete; do not infer settings from names or performance. Request Facebook versus Instagram and placement breakdowns separately where needed. If a metric or breakdown is unavailable, say so; do not treat missing data as zero. Check measurement gaps, objective alignment, budget allocation, audience settings, placements and possible creative fatigue. Compare ratios from totals, not averages of row-level ratios. Do not sum reach across overlapping audiences. Return up to five evidence-supported findings. For each: entity name and ID, dates, evidence, business implication, confidence, next check and proposed action. Separate observations from hypotheses. Do not estimate savings or claim causation without evidence. Finish with the unanswered questions for the client.
Read the account confirmation before the recommendations. An audit of the wrong client or a partial date range can still sound convincing. Compare the spend and selected conversion totals with Ads Manager using the same dates, event and attribution settings.
4. What should the Meta ads audit checklist cover?
| Check | Evidence to collect | Decision it supports |
|---|---|---|
| Measurement | Selected conversion event, attribution settings, Ads Manager totals and CRM or store totals | Whether the performance numbers are usable |
| Campaign objective | Objective, destination and optimization event for each spending campaign | Whether delivery matches the client's business goal |
| Budget allocation | Spend and relevant results by campaign and ad set, including paused entities with historical spend | Which areas deserve investigation before a budget move |
| Facebook vs Instagram | Platform and placement breakdowns with spend and the same conversion event | Where recorded results differ and where data is too thin |
| Audiences | Targeting settings, exclusions and supported demographic or geographic reports | Whether delivery matches the intended customer and service area |
| Creative performance | Ad IDs, period-over-period results, frequency and the actual creative assets | Which ads may need a new hook, offer or format |
| Client outcomes | Qualified leads, booked calls, sales or net revenue alongside platform results | Whether cheaper platform results are useful to the business |
Do not rank a traffic campaign against a sales campaign using a generic ‘results’ column. Their results can mean different things. Similarly, compare prospecting and retargeting in context: people already familiar with a business are not equivalent to new prospects.
5. How do you compare Facebook and Instagram placements?
Ask for platform first, then placement detail. A combined account average can hide a weak Instagram placement or a Facebook placement that drives inexpensive leads with poor qualification. Other platforms can appear in Meta reporting; retain them so the totals reconcile. Meta’s official Insights field reference includes publisher_platform and platform_position as reporting breakdowns.
For [ACCOUNT_ID] and the two audit periods, compare publisher platforms, then supported platform/placement combinations. Show spend, the selected conversion count and cost per conversion; add purchase value and ROAS only when available. Label click metrics precisely. Do not call all-click CTR link CTR. Keep unsupported conversion breakdowns marked unavailable. Show the sample size behind every recommendation and distinguish a review candidate from a confirmed problem. Do not change placements or budgets.
A high cost per acquisition (CPA) with two purchases is a different decision from the same CPA with two hundred. Inspect spend, volume, conversion delays and the client's target together. The cheapest placement is not automatically the best place to move the next dollar; its performance at a larger budget is untested.
6. How do you check creative fatigue and audience problems?
Look for a pattern across comparable periods: frequency rising while response and conversion performance deteriorate. Frequency alone is not proof of fatigue, and there is no universal frequency number at which every client should pause an ad. An offer change, seasonality or a website problem can produce similar symptoms.
Compare each ad's performance across the two periods. Flag possible fatigue only where multiple signals support it: frequency, consistently defined CTR, conversion volume and cost per conversion. Separate ads that declined from ads that never performed well. For each flagged ad, show its ID and evidence. Inspect copy and creative assets only if they are available; otherwise ask me for the ad preview or image/video. Suggest one test and the evidence that would support keeping it. Create nothing.
A performance table does not show the visual hook, landing-page promise or video opening. Review those assets before recommending new creative. Likewise, demographic results can suggest a question to investigate, but they do not prove audience overlap or justify excluding a segment from a small sample.
For a recurring local-service review after the audit, see the personal-trainer Meta ads playbook. The wedding-photographer guide covers the different problem of leads that take months to become bookings.
7. What must you verify outside ChatGPT?
Check Meta Pixel and Conversions API event behavior in Events Manager if the campaign relies on website conversions. Where browser and server events represent the same conversion, review deduplication: Meta’s server-event reference explains how event_id and event_name identify matching events. A performance report alone cannot certify that tracking fires correctly or that the right event is configured.
Reconcile platform results with CRM or store records, but do not expect identical totals: attribution rules, conversion dates, reporting delays and business definitions can differ. A reported purchase ROAS is attributed purchase value divided by ad spend; it is not profit and does not prove how many purchases the ads caused.
If conversion counts are zero or absent, investigate tracking and reporting before calling all associated spending wasted. For lead campaigns, keep different action types separate rather than adding every action into one lead count.
8. What does a useful client audit finding look like?
This fictional example uses matched lead definitions and periods, with CRM qualification counts supplied by the client. It is not a LoomaScale customer result or an industry benchmark. All amounts are USD.
| Segment | Spend | Recorded leads | Qualified leads | Cost per lead | Cost per qualified lead |
|---|---|---|---|---|---|
| Facebook placements | $1,200 | 40 | 12 | $30 | $100 |
| Instagram placements | $900 | 45 | 3 | $20 | $300 |
| Combined | $2,100 | 85 | 15 | $24.71 | $140 |
Instagram appears cheaper at the recorded-lead stage, but qualification changes the decision. The combined cost per qualified lead is $2,100 divided by 15, or $140; it is not the average of $100 and $300.
A defensible finding is: ‘Instagram placements produced cheaper recorded leads but a higher cost per qualified lead in this period. Check offer, placement, qualification criteria and follow-up consistency before testing a budget change.’ It does not claim Instagram is inherently worse or predict guaranteed savings.
9. How do you turn the audit into a client report?
Keep the report short enough to discuss on a call, with evidence available underneath. The client should understand the decision and what remains uncertain.
Turn the verified findings into a client-facing Meta ads audit report. Include: 1. Scope: account, currency, time zone, dates, goal, conversion definition and attribution context. 2. Executive summary: up to three findings, each with a number and its source. 3. Performance table: both periods, absolute changes and percentage changes where the baseline is nonzero. 4. Priorities: measurement fixes first, then evidence-supported performance tests. 5. Action plan: finding, evidence, proposed action, owner, success metric and review date. 6. Limitations: missing fields, small samples, attribution differences and checks still needed. Use only verified numbers. Mark examples and hypotheses clearly. Keep proposed account changes pending client approval.
Close the audit by agreeing on the first action and its review date. Preserve the original report so the follow-up measures what changed against a documented baseline. If the client also runs paid search, use the separate Google Ads audit workflow, which covers search terms and keywords.
To run this workflow against a connected account, connect Meta Ads to ChatGPT through LoomaScale, confirm the client's account, and paste the audit prompt above.
