Google AdsMetaTikTokGA4IntermediateGuide

New Client Ad Audit Checklist: 42 Things to Check in the First Week (2026)

The first-week audit for any ad account you inherit: access and context, tracking on every platform, money and pacing, structure and targeting, creative, landing pages, and the baseline you report against. Every item has a threshold, so an agent can run it.

Updated September 25, 202615 min read

every agency has a "new client audit." most of them are a slide deck that repeats the previous agency's mistakes with a new logo, because they start from the campaigns instead of from the data those campaigns were trained on. this is the first-week checklist that starts where the damage starts: whether a single number in the account can be trusted.

what a first-week audit actually is

it is not a performance review. you cannot review performance until you know whether the conversion count is real, the spend is landing where the reports say, and the platforms are learning from the right signal. the first week has one job: establish what is true, write it down, and turn it into a plan with owners.

here is what most audit guides get wrong: they list 100 things with no thresholds and no order. a checklist an agent can run has a check, a flag condition, and a recommended action for every line. that is how this one is written. 42 items, seven groups, five days.

the data that says this matters: wordstream's google ads performance study, built on more than 250,000 grader reports across 23 industries, found the average account wastes $1,127.54 per month, 29% of accounts recorded zero conversions over 90 days while still averaging 12,667 impressions a month, and 25% had never added a single negative keyword (wordstream, google ads performance study, 2025). most of that is found in week one if you look in the right order.

the fundamental shift: audit the signal before the campaigns

before: audits opened ads manager, sorted campaigns by spend, and started forming opinions about bids and audiences.

after: every major platform runs on machine-learned bidding and delivery. google's smart bidding optimizes toward whatever is marked primary. meta's ranking models train on whatever the pixel and conversions api send. tiktok's delivery learns from the events api. an audit that judges campaigns before it verifies those inputs is reviewing the output of a model that may have been trained on garbage.

the order that works: access and context, then tracking, then money, then structure, then creative and pages, then the baseline. each group depends on the one before it.

what this broke in existing accounts

1. inherited "wins" that were double-counted

the previous team reported a 4.2 roas. the native purchase tag and the imported analytics conversion were both primary. real roas was 2.1. the client's expectations were set on a number that never existed.

2. months of bidding toward the wrong event

lead accounts where the primary conversion was a page view. ecommerce accounts where add-to-cart was primary because purchase "didn't have enough volume." smart bidding did exactly what it was told.

3. pixel-only meta accounts with poor match quality

meta rates event match quality as poor, ok, good, or great, and says to "aim for an event match quality rating of good or great" (meta business help center, about event match quality). accounts running pixel-only for years sit at poor or ok, and every lookalike, every optimization, and every attributed purchase was built on it.

4. duplicate purchase events inflating everything

pixel and conversions api both sending purchase without a shared event_id. meta deduplicates only on event_id plus event_name (meta for developers, handling duplicate pixel and conversions api events). every purchase counted twice, every roas doubled.

5. spend the client did not know about

experiments left running, a paused campaign with an active ad schedule, a performance max campaign cannibalizing brand search, a legacy ad account under a different business manager. the first week is when you find the money nobody is watching.

6. no baseline, so no way to prove improvement

the agency changed twelve things in week one, performance moved, and nobody could say which change did it or whether the previous period was even measured the same way.

the checklist

each item: what to check, what flags it. thresholds are the ones the platforms document or, where none exists, the ones we use.

day 1: access and context (6)

  1. admin access on every ad account, pixel or dataset, tag manager container, analytics property, and merchant feed. flag: any read-only or missing access. you cannot audit what you cannot open.
  2. ownership. flag: any pixel, tag manager container, or ad account owned by a previous agency's business manager or personal login. ownership transfer goes on the plan before anything else.
  3. the business model and the money event. subscription, one-time purchase, lead to sales call, marketplace. flag: no written definition of what a conversion is worth. write it now; every threshold below depends on it.
  4. target economics. target cpa or roas, margin, payback period, ltv if known. flag: targets that were set from platform-reported roas rather than from margin.
  5. geography, seasonality, and promotions calendar. flag: an audit window that overlaps a sale or a holiday without noting it.
  6. the previous agency's change history. google ads change history and meta's activity log for 90 days. flag: more than 3 structural changes per week in the last month, which means the learning phase never finished.

days 1 to 2: tracking, every platform (12)

google ads and ga4

  1. conversion action status. flag: any primary action showing unverified, tag inactive, needs attention, or no recent conversions, which google defines as no conversions in the last 7 days (google ads help, troubleshoot conversion tracking status).
  2. one primary per outcome. flag: a goal with zero primary actions, or a native tag and a ga4 import both primary for the same outcome. see ga4 and google ads conversion import.
  3. enhanced conversions. flag: diagnostics status other than excellent on the primary purchase or lead action, or the customer data terms not accepted on the client account. see enhanced conversions.
  4. auto-tagging, the ga4 link, and utm hygiene. flag: auto-tagging off, no link, or utm_source and utm_medium values outside the allowed list. see the utm standard.

meta

  1. pixel plus conversions api on the same dataset. flag: pixel-only, or conversions api events with no browser events to deduplicate against.
  2. event match quality on the purchase or lead event. flag: rated poor or ok. meta says aim for good or great.
  3. deduplication. flag: purchase events arriving from both sources without matching event_id, visible in events manager diagnostics. see meta attribution setup.
  4. attribution setting consistency. flag: active ad sets using different attribution windows, or a comparison window that differs from what the client's reports assume.

tiktok

  1. pixel plus events api with deduplication. flag: single-source tracking or duplicate events in events manager. see tiktok attribution setup.
  2. the optimization event. flag: campaigns optimizing for complete payment when the events api has not delivered that event in the last 7 days.

ga4 and the site

  1. key events and hostname filters. flag: key events firing on non-production hostnames, or a purchase key event whose count differs from the store's order count by more than 15% for the same period.
  2. consent mode and regional coverage. flag: eea or uk traffic with no consent mode signals, or google ads modeled conversions at zero for those regions.

days 2 to 3: money and pacing (6)

  1. every active account and every active campaign, listed with 30-day spend. flag: any spend the client did not know about, and any account not in the client's own business manager or manager account.
  2. budget pacing. flag: any campaign more than 20% over or under its monthly pace at mid-month, or daily budgets that have been changed more than twice in 7 days.
  3. experiments and drafts. flag: any experiment running past its end date, or a draft that was applied without documentation.
  4. brand search spend. flag: brand terms with no dedicated campaign, or performance max serving on brand queries without brand exclusions.
  5. billing and thresholds. flag: a card on a personal account, a spend limit that will cap mid-month, or an unpaid balance.
  6. the 90-day search terms report by cost. flag: more than 10% of search spend on terms with zero conversions and no negative keyword coverage. wordstream found a quarter of accounts have no negatives at all.

days 3 to 4: structure and targeting (6)

  1. campaign count per objective. flag: more than 3 active campaigns for the same objective and geography. concentration is what the ranking models need. see meta andromeda.
  2. ad set and ad group signal density. flag: any ad set or ad group with fewer than 50 optimization events in the trailing 7 days while still active.
  3. audience overlap. flag: prospecting and remarketing audiences with no mutual exclusion, or lookalike and interest stacks that share more than half their reach.
  4. match types and keyword sprawl. flag: broad match without a smart bidding strategy, or ad groups with more than 20 keywords sharing one ad.
  5. placements and networks. flag: search campaigns opted into display network, or meta placements manually pruned without a documented reason.
  6. bid strategy versus signal. flag: target roas or target cpa on a campaign with fewer than 30 conversions in 30 days, or manual cpc on a campaign with hundreds.

day 4: creative (4)

  1. creative count and diversity per ad set. flag: fewer than 8 active ads, or more than half sharing the same format and visual style.
  2. fatigue. flag: frequency above 3.5 on a prospecting ad set, or ctr down more than 20% week over week on a top-spend ad.
  3. ad strength and disapprovals. flag: any disapproved or limited ad on a top-10 spend campaign, or responsive search ads rated poor.
  4. the funnel in the creative. flag: an ad set with no problem-aware hook, or no proof or offer ad. see the algorithms behind meta ads for why sequences matter.

days 4 to 5: landing pages and funnel (4)

  1. click to page. flag: any final url returning a redirect chain longer than 2 hops, a 4xx, or dropping query strings on the way.
  2. speed and core web vitals on paid landing pages. flag: largest contentful paint above 2.5 seconds on mobile for a top-5 spend landing page.
  3. page-level conversion rate. flag: a landing page receiving more than 10% of paid sessions with a conversion rate under half the site average.
  4. message match. flag: an ad promising an offer, price, or product the landing page does not show above the fold.

day 5: reporting and the baseline (4)

  1. one source of truth per number. flag: a client dashboard that mixes platform-reported conversions with analytics conversions in one column.
  2. the baseline period. record 90 days of spend, conversions by the platform's "by conversion time" column where available, revenue from the store or crm, and the gap between them. flag: no baseline stored before any change ships.
  3. the plan with owners. every flag above becomes a line: problem, evidence, impact, recommendation, effort, priority, owner, deadline. flag: any critical tracking item scheduled after any campaign change.
  4. the reporting cadence. weekly for the first month, with the baseline gap re-checked each week. flag: a first report scheduled before tracking fixes have 7 days of data.

what changed in 2026 specifically

google ads: enhanced conversions became a single toggle with multiple simultaneous data paths (april and june 2026), and offline uploads moved to data manager. an inherited account on a 2023 setup is likely under-sending user data and may still be on a retired upload path.

meta: andromeda completed rollout in october 2025 and the ranking stack (gem, lattice, the adaptive ranking model) keeps changing through 2026. structure findings that were acceptable in 2024, such as many small ad sets, are now defects.

ga4: per-conversion attribution settings (january 2026), custom conversion windows (august 2026), and hostname allowlists (september 2026). the ga4 side of an inherited account has settings the previous team never had access to.

apple: safari 26 (september 2025) made advanced fingerprinting protection default. click ids on apple traffic are unreliable, which raises the weight of items 10 and 18 above.

what actually works now

1. run the groups in order and do not skip ahead

context before tracking, tracking before money, money before structure. a structure finding written before the tracking group is a guess.

2. fix tracking before you touch a campaign

tracking fixes need 7 days of data before anything downstream can be judged. ship them on day 2 or 3 so the first campaign changes, in week two, land on clean signal.

3. write the baseline with the platform's own reconciliation columns

google ads "conversions (by conv. time)," meta with the attribution window the client's reports assume, and store or crm revenue for the same dates. record the gap. the gap is the baseline, not a bug.

4. turn every flag into a line with an owner

problem, evidence, impact, recommendation, effort, priority, owner, deadline. the deliverable is the task list, not the deck.

5. hold the first structural change until the learning phase math works

if item 26 flagged, the campaign restructure comes first and the bid strategy change comes after it has 50 events in 7 days.

6. re-run the checklist monthly

the platforms change quarterly and the previous agency's habits linger in the account. a monthly pass on items 7 through 24 catches drift before the client does.

what to stop doing immediately

  • stop opening the audit with campaign performance.
  • stop reporting the previous agency's roas as the baseline without reconciling it.
  • stop changing bids or budgets in week one before tracking is fixed.
  • stop treating an "ok" event match quality as fine.
  • stop accepting duplicate purchase events as "meta's attribution."
  • stop letting brand search run inside performance max without exclusions.
  • stop delivering findings without owners and deadlines.

the orbit agent instructions

{
  "guide": "new-client-onboarding-audit-checklist",
  "agent_tasks": {
    "audit_access_and_ownership": {
      "check": "enumerate every connected ad account, pixel or dataset, tag manager container, analytics property, and merchant feed, with the access level held and the owning business entity",
      "flag": "any asset with read-only or missing access, or owned by an entity other than the client",
      "recommend": "request admin access and initiate ownership transfer before any other audit item"
    },
    "audit_tracking_health_all_platforms": {
      "check": "for each platform, read conversion action or event status, primary versus secondary assignments, event match quality, deduplication diagnostics, attribution settings, and ga4 key event counts versus store or crm counts for the last 30 days",
      "flag": "any primary action in a non-active status, more than one primary per outcome, event match quality below good, duplicate purchase events without a shared event_id, inconsistent attribution windows, or a purchase count gap above 15 percent",
      "recommend": "produce the tracking fix list first, ship it before any campaign change, and wait 7 days before judging performance"
    },
    "audit_spend_and_pacing": {
      "check": "list all active campaigns across accounts with 30-day spend, pacing against monthly budget, running experiments, brand search coverage, billing method, and the 90-day search terms report by cost",
      "flag": "unknown spend, pacing more than 20 percent off at mid-month, expired experiments still running, brand queries served by performance max without exclusions, or more than 10 percent of search spend on zero-conversion terms with no negatives",
      "recommend": "stop unknown spend, add brand exclusions and negatives, and log each item on the plan with an owner"
    },
    "audit_structure_and_signal_density": {
      "check": "count campaigns per objective and geography, optimization events per ad set or ad group in the trailing 7 days, audience overlap, keyword sprawl, placement pruning, and bid strategy versus conversion volume",
      "flag": "more than 3 campaigns per objective, any active ad set under 50 events in 7 days, overlapping prospecting and remarketing audiences, or target-based bidding under 30 conversions in 30 days",
      "recommend": "consolidate structure first, then change bid strategies only once signal density supports them"
    },
    "audit_creative_and_landing_pages": {
      "check": "for top-spend ad sets, count active ads and their format and visual diversity, frequency, week-over-week ctr, disapprovals, and for top-spend landing pages measure redirects, query string survival, mobile largest contentful paint, and page conversion rate against site average",
      "flag": "fewer than 8 ads or over half sharing one style, frequency above 3.5, ctr down more than 20 percent, any disapproval, redirect chains over 2 hops, lcp above 2.5 seconds, or page conversion rate under half the site average",
      "recommend": "add creative diversity before pausing anything, and fix landing page technical issues before testing new pages"
    },
    "produce_baseline_and_plan": {
      "check": "assemble 90 days of spend, platform conversions by conversion time, store or crm revenue, and the gap between them, then compile every flagged item into a plan with problem, evidence, impact, recommendation, effort, priority, owner, and deadline",
      "flag": "no stored baseline, any critical tracking item scheduled after a campaign change, or a first client report scheduled inside 7 days of tracking fixes",
      "recommend": "store the baseline, order the plan tracking first, and set the first report for the week after tracking fixes have a full 7 days of data"
    }
  }
}

the bottom line

the first week decides whether every number you report for the next year can be trusted. verify access, fix the signal, find the money, then judge structure and creative against a baseline you recorded yourself.

42 items, in order, with thresholds. an agent can run it every month. a human should run it before the first campaign change.

built by orbit - the performance marketing intelligence layer. questions? jose@orbitllm.com