A Meta ads audit checks ten things: account structure, naming, budgets and pacing, bidding, audiences, creative fatigue, placements, tracking, attribution settings and wasted spend. You can run a Facebook ads audit from Claude through Whathead’s read-only MCP address (whathead.com/mcp/read), which can read your account but cannot change anything, then turn the findings into a dry-run fix plan.
- Connect https://www.whathead.com/mcp/read in Claude for the audit. It lists only read tools and refuses any write before it reaches Meta.
- If numbers look wrong, start with tracking and attribution: a broken event makes every CPA in the account meaningless.
- For each finding, write down the evidence, the impact and the fix, then request the fix as a dry run on the full connector.
- The same checklist works on TikTok and Snapchat with different field names.
Whathead wrote this guide and makes the Meta Ads MCP server used in it. Meta facts come from Meta’s developer documentation and help pages, checked on 19 November 2026; sources are listed at the end. Meta renames settings often, so if something looks different in your account, check the linked page.
I run audits in the same order every time: can I trust the data, is the structure helping, and where is the money going. What an AI assistant changes is the reading. Claude can pull every campaign, ad set and ad with its settings and insights in a few calls, so you spend your time judging findings instead of exporting reports. Below is the checklist, what bad looks like for each item, and prompts to run it through Whathead’s Meta Ads MCP without giving the assistant any way to change the account.
- Definition Meta ads audit
- A Meta ads audit (or Facebook ads audit) is a structured review of a Meta ad account: how campaigns are built, how budget is spent, whether tracking can be trusted and where money is wasted. On a multi-channel account it is the Meta part of a wider PPC audit.
What does a Meta ads audit checklist cover?
Ten areas, each with its own section below. If you only have an hour, do tracking (8), attribution (9) and wasted spend (10) first.
Meta ads audit checklist at a glance
| What to check | What bad looks like | |
|---|---|---|
| 1. Structure | Budget level, Advantage+ levers, learning status | Many small ad sets splitting one audience |
| 2. Naming | Names against your convention, UTMs | Names you can’t parse |
| 3. Budgets and pacing | Spend vs plan, budget type, end dates | Flights that end early or under-deliver |
| 4. Bidding | Bid strategy and optimization goal | Caps that block delivery |
| 5. Audiences | Inclusions, exclusions, Advantage+ audience | Ad sets chasing the same people |
| 6. Creative fatigue | Frequency, CTR and cost trend per ad | Frequency up, CTR down, cost up |
| 7. Placements | Results by platform and position | Spend where the creative doesn’t fit |
| 8. Tracking | Pixel, Conversions API, deduplication, EMQ | Double counting, EMQ under 6.0 |
| 9. Attribution | Setting per ad set, removed windows | Ad sets compared on different windows |
| 10. Wasted spend | Spend with no results, stale ads | Weeks of spend, zero conversions |
1. Is the account structure helping delivery or splitting your data?
Meta learns per ad set, so many small ad sets with the same goal each get too little data. Meta’s learning phase guidance explains how many optimization events an ad set needs after a significant edit; the API reports the state in learning_stage_info.
- Check: budget level (Meta allows campaign or ad set budgets, not both); each ad set’s learning_stage_info and issues_info (ad set reference); each campaign’s advantage_state_info. Meta only treats a campaign as Advantage+ sales, app or leads when Advantage+ campaign budget, audience and placements are all on (Meta docs). Also look for old Advantage+ shopping or app campaigns: Meta stopped API edits for them and said the rest would be paused with Graph API v26.0 (Meta developer blog).
- Bad looks like: five ad sets with the same audience and goal, none leaving learning; a campaign everyone calls Advantage+ that reports DISABLED because one lever is off.
2. Can you read every campaign, ad set and ad name?
Names are the only metadata that survive into every export and client report. If a name doesn’t give market, objective, audience and creative, nobody can group it later, and neither can Claude. Designing the pattern is covered in our ad naming conventions guide.
- Check: every active name against the convention, duplicate names on live entities, and whether UTMs (url_tags) match the names.
- Bad looks like: “Copy of”, “New ad set”, three date formats, or two live campaigns with the same name in different markets.
3. Are budgets pacing to the plan?
Most budget problems are arithmetic: a lifetime budget that runs out before the flight ends, or a daily budget too small for the ad set to learn.
- Check: daily vs lifetime budgets, spend to date against the media plan, budget_remaining on lifetime ad sets, and end dates in the account time zone.
- Bad looks like: a flight projected to finish early; one ad set taking nearly all of a campaign budget when the plan needs a split. For continuous checks, Whathead’s budget pacing board (early access) compares live spend with the approved plan.
4. Do bid strategies match the business goal?
Meta’s API has four bid strategies for auction ad sets: LOWEST_COST_WITHOUT_CAP (highest volume), LOWEST_COST_WITH_BID_CAP (bid cap), COST_CAP (cost per result goal) and LOWEST_COST_WITH_MIN_ROAS (ROAS goal).
- Check: bid strategy, cap or goal, optimization_goal and the promoted_object event per ad set, next to actual cost per result and spend vs budget.
- Bad looks like: a cost cap below any CPA the account has achieved, so the ad set barely spends; a sales campaign optimizing for link clicks; a value goal on events with no purchase value.
5. Are your audiences overlapping or too narrow?
Overlap matters when ad sets with the same goal bid for the same people. Ads Manager has an overlap view for saved and custom audiences; Whathead doesn’t expose it, so Claude compares targeting specs instead.
- Check: included and excluded custom audiences, lookalikes, locations and ages per ad set; whether Advantage+ audience is on; whether prospecting excludes purchasers where the plan says so.
- Bad looks like: three ad sets on the same lookalike; retargeting audiences inside prospecting; a very narrow audience with Advantage+ audience off on a conversion goal.
6. Is creative fatigue raising your costs?
Meta’s own CREATIVE_FATIGUE and CREATIVE_LIMITED recommendations say cost per result “may be higher than ads you ran in the past because its image or video has been shown to parts of your audience too many times” (Meta docs). Frequency, the average number of times each person saw your ad, is an estimate.
- Check: weekly frequency, CTR and cost per result per ad; quality, engagement rate and conversion rate rankings; how many ads each ad set rotates; fatigue items in the account’s recommendations.
- Bad looks like: frequency up while CTR falls on the same ad; one ad taking almost all spend for months. There is no universal fatigue frequency: compare each ad with its own history.
7. Where is the money going by placement?
Break insights down by publisher_platform (Facebook, Instagram, Audience Network) and platform_position (feed, Stories, Reels and so on), both standard Insights breakdowns (Meta docs).
- Check: results by placement; whether Advantage+ placements is on (any placement exclusion turns it off); applied publisher block lists; asset ratios for the placements each ad serves.
- Bad looks like: heavy spend on a placement with almost no results; a 16:9 video cropped into Reels; old exclusions nobody can explain.
8. Can you trust the tracking?
Check tracking before you judge any CPA or ROAS. General UTM and event checks are in our tracking QA guide; these are the Meta-specific ones.
- Optimization event: each ad set’s promoted_object uses the right pixel and event, and the event still fires. Through Whathead, Claude can read a pixel’s event stats and last-fired times.
- Deduplication: Meta deduplicates pixel and Conversions API events when the pixel’s eventID matches the server event_id and the event names match, within 48 hours (Meta docs). Server events without event_id get counted twice.
- Event Match Quality: Meta scores EMQ for web events from 1 to 10 (setup guide) and suggests aiming for 6.0 or higher (Meta docs). Events Manager is the reference view; the Dataset Quality API returns similar data, and Whathead can read the pixel’s integration-quality and diagnostics endpoints where your permissions allow.
- Aggregated Event Measurement: AEM is how Meta measures web events from iOS users who opted out of tracking. If Events Manager still shows configured web events, make sure your optimization event is listed.
- Bad looks like: a purchase campaign optimizing on a custom conversion that stopped firing; EMQ in the low single digits on Purchase.
9. Which attribution setting is each ad set using?
Each ad set has its own attribution_spec, so two ad sets on different settings can’t be compared directly. Since 12 January 2026 the Ads Insights API no longer returns 7-day view and 28-day view; 1-day click, 7-day click, 28-day click, 1-day engaged view and 1-day view remain. The same change limits unique-count fields such as unique_actions to 13 months of history (Meta developer blog).
- Check: the setting of every active ad set; dashboards still requesting removed windows; Meta-reported conversions against GA4 or your MMP.
- Bad looks like: a 7-day click and 1-day view ad set ranked against a 1-day click one; a report showing zero view-through because it asks for 28d_view.
10. Where is spend being wasted?
Define “wasted” before you ask, for example “spent more than twice the target CPA with no conversion in 14 days”, or Claude will invent a definition.
- Check: ad sets and ads above that rule; ads still running after the offer ended; placements with spend and no results; destination URLs that break or drop UTMs.
- Bad looks like: a month of spend without a purchase; last season’s offer still live; near-identical ads competing in one ad set.
How do you run a Meta ads audit from Claude without risk?
Use Whathead’s read-only address. https://www.whathead.com/mcp/read signs in like the full server but lists only read tools. Its Meta tool, meta_native_read, runs the 390 read actions of the Meta gateway; anything that would create, change or delete is refused with READ_ONLY_TOOL_REJECTED_WRITE before a request leaves Whathead (tool catalog). The same address covers the read side of TikTok, Snapchat, GA4 and the other connected platforms.
Run a read-only Meta ads audit from Claude
Connect Claude to Meta Ads through Whathead’s read-only MCP address and audit an ad account without write tools.
- 01
Connect Meta in Whathead
On a plan with MCP access, open Integrations, connect Facebook with a profile that can see the ad accounts, and select the accounts to share.
- 02
Add the read-only connector
In Claude, open Customize > Connectors, click + Add, then Add custom connector. Enter https://www.whathead.com/mcp/read, sign in with Whathead and approve.
- 03
Turn it on in the chat
Click + in the message box, choose Connectors and switch it on. Leave the full Whathead connector off in the audit chat.
- 04
Set the scope
Name the ad account, date range, KPI and attribution setting, so every number uses the same basis.
On Claude Team and Enterprise, an owner adds the connector under Organization settings > Connectors (Claude help). ChatGPT can use the same address. Steps per client are in the connection guide and on Connect Claude to Meta Ads.
Prompts you can copy
Replace the IDs and thresholds with yours. Each prompt only reads.
Scope and structure
- “Use only the read-only Whathead tools. For Meta ad account act_XXXX, list every active campaign, ad set and ad with objective, budget, bid strategy, optimization goal, attribution setting and last-30-day spend, results and cost per result. Show the table, no conclusions yet.”
- “For each active campaign, read advantage_state_info and tell me which Advantage+ levers are on. Then list ad sets that are learning or learning limited.”
- “Our naming convention is Market_Objective_Audience_Launchdate. List every active name that doesn’t fit, and any duplicates.”
Money, audiences and creative
- “Here is the media plan [paste]. Project each campaign’s spend to its end date at the current rate and flag anything more than 10% off plan.”
- “For prospecting ad sets, list included and excluded custom audiences, and flag any that don’t exclude purchasers.”
- “For ads with spend in the last 28 days, show weekly frequency, CTR and cost per result, then list any creative fatigue items in the account’s recommendations.”
Tracking and waste
- “For the pixel our purchase ad sets use, show when each standard event last fired and any signal diagnostic issues. List ad sets that use a different pixel or event.”
- “Find ad sets that spent more than $200 in the last 14 days with zero purchases, with placement and audience.”
- “Write the audit as a table: finding, evidence with the numbers you used, impact, fix, priority. Mark anything you couldn’t verify.”
How do you turn audit findings into a dry-run fix plan?
Audits usually fail at the hand-off: the findings sit in a document and nothing changes. Close the loop in three steps.
- Order fixes by risk. Tracking and naming first (no delivery impact), then budgets and bids, then pauses and restructures.
- Switch connectors for changes. In a new chat, add https://www.whathead.com/mcp and ask for each fix as a dry run. Whathead’s Meta write actions are previews unless the call sets dry_run to false, and the preview shows the exact Marketing API request. Deletes also need confirmed: true.
- Apply in small batches and re-read. After each live change, have Claude read the entity back. Our bulk editing guide covers batch size and rollback.
Read every preview before you confirm, don’t click “Allow always” on write tools, and pause instead of deleting.
Hand-off prompt: “Using the audit above, prepare dry-run changes for the P1 items only, show old and new values, and wait for my confirmation.” If the same problems keep coming back, Whathead’s anomaly detection (early access) checks live setup, pacing and performance between audits. For the wider model, see AI ad operations agents.
How is a TikTok or Snapchat audit different?
The checklist carries over; the names, automation and attribution defaults don’t.
Audit differences by platform
| Meta | TikTok | Snapchat | |
|---|---|---|---|
| Middle level | Ad set | Ad group | Ad squad |
| Automation to review | Advantage+ budget, audience and placements (advantage_state_info) | Upgraded Smart+ for Traffic, Sales, Lead Generation and App Promotion | Bid, placement and targeting settings per ad squad |
| Attribution | Per ad set; 7-day and 28-day view removed from the Insights API since 12 Jan 2026 | Per ad group: click 1, 7, 14 or 28 days; view off, 1 or 7 days | Stats API defaults: 28-day swipe, 1-day view |
| Match quality | EMQ from 1 to 10 in Events Manager | EMQ as a weighted average of match key coverage, plus a Diagnostics tab | Snap Pixel and Conversions API events per ad squad |
| Whathead read tool | meta_native_read (390 read actions) | tiktok_native_read (280) | snapchat_native_read (182) |
On TikTok, upgraded Smart+ puts manual and automated setup in one flow, so check what each campaign automates (TikTok help); attribution is set per ad group (TikTok help) and Events Manager’s Diagnostics tab lists event issues (TikTok help). On Snapchat, align swipe and view windows before comparing channels (Snap docs). Setup: TikTok Ads MCP and Claude for Snapchat Ads.
Meta ads audit: common questions
How often should you audit a Meta ads account?
When you take over an account, before you scale budget, and about once a quarter. Check tracking and pacing monthly, because those break between audits.
Can Claude audit my Facebook ads account without changing anything?
Yes. Connect Whathead’s read-only address, https://www.whathead.com/mcp/read. It lists only read tools and refuses any create, edit, pause or delete before a request reaches Meta.
Is a Meta ads audit the same as a PPC audit?
A PPC audit covers every paid channel, including search platforms such as Google Ads; a Meta ads audit is the Meta part. The order is the same (data, structure, money), plus Meta-specific checks such as Advantage+ levers and Event Match Quality.
What do I need to run the audit from Claude?
A Whathead plan with MCP access (MCP Actions $39 a month, Operator Bundle $65 or Agency $249, each with a 14-day trial), Meta connected with a profile that can see the ad accounts, and a Claude plan with custom connectors.
Can I use Meta’s own Ads MCP server for an audit?
Yes, for Meta-only reads; our Meta Ads MCP comparison covers the options. Whathead adds a read-only address and TikTok, Snapchat and other platforms in the same conversation.
What should I fix first after an audit?
Tracking and attribution first, because they distort every other number. Then wasted spend, then budgets and bidding, and structural changes last, since they reset learning.
Audit Meta Ads from Claude, read-only
Connect Whathead’s Meta Ads MCP at whathead.com/mcp/read: 390 Meta read actions and no write tools in the conversation.
Sources
- Meta structure and automation: Advantage+ campaign experience, ASC and AAC deprecation, ad set reference, learning phase
- Meta reporting: Insights API attribution and retention changes, Insights breakdowns, opportunity score and recommendations
- Meta tracking: deduplication, verifying your setup, Dataset Quality API, technical setup guide
- TikTok: Smart+ upgrades, ad group attribution, Events Manager diagnostics; Snapchat: measurement API
- Claude: custom connectors; Whathead: Meta Ads MCP, tool catalog, connection guide, pricing
- Related on this blog: Best Meta Ads MCP servers, Facebook Ads MCP
Written by Ahmed Shahin, founder of Whathead. Last reviewed 19 November 2026.

