How to Find Wasted.

You audit your ads with Claude by exporting your campaign data from Google and Meta, pasting it in, and asking Claude to grade each platform and rank where money goes out with nothing back. Industry analysis puts wasted paid-search budget at 20 to 30%, so an hour of reading usually finds real money. The prompts below run in any Claude, free, and the installed /ads-audit skill turns the one-off read into a graded baseline that later audits chart against.

The reason this is worth your hour is that the gap is large and quiet. It does not announce itself. Roughly 20 to 30% of the average paid-search budget is wasted (industry analysis, 2025), and most of that waste is spread across ad sets, search terms, and audiences that each look fine on their own. No single line in the account screams "stop." The job of an audit is to read the whole account at once, weight every finding by the dollars behind it, and tell you which fix is worth doing first. For the version that saves a baseline and tracks the gap month over month, see Claude as a business operating system.

This guide covers where ad budget gaps, the four or five gap types Claude looks for, the copy-paste prompts that find them, how to read the grades it returns, what the audit can and cannot do, how the one-off prompt differs from the installed /ads-audit skill, and the common mistakes that make an audit lie to you. Everything here works from manual exports. You do not need to connect an API or grant access to anything.

• Roughly 20 to 30% of the average paid-search budget is wasted (industry analysis, 2025).
• Most waste hides in four or five places at once: ad sets with spend and zero conversions, audience overlap, creative fatigue, broad match buying the wrong clicks, and broken conversion tracking.
• Claude reads Google and Meta data from your own exports, writes a per-platform grade plus an action plan plus ledger lines, and runs standalone with no account access.
• Weighting every finding by spend is what tells you which gap to fix first, instead of a flat list of problems.
• The /ads-audit skill reads the exports, scores each platform, and saves a baseline; the next audit leads with the delta.

Where Does Ad Budget Actually Gap?

Rarely from one big mistake. A single botched campaign is easy to spot and easy to pause. The 20 to 30% that quietly disappears gaps from several small places at once, none of which is obviously broken when you glance at it. That is exactly why it survives: each gap is below the threshold where you would stop and investigate, but added together they are a quarter of the budget.

The gap also moves. An audience that converted in March fatigues by May. A search term that was fine last quarter starts matching junk after Google loosens a match type. A new ad set launches, overlaps an old one, and the two start bidding against each other without anyone noticing. Because the account drifts, a one-time cleanup is not enough. You need a read you can repeat, with a record of what the account looked like last time, so you can see what changed. That repeatability is the whole reason the installed skill saves a baseline instead of just printing an answer and forgetting it.

Claude reads the account the way a careful media buyer would: it pulls every export into one view, weights each finding by the spend behind it, and ranks the gaps by money rather than by how alarming they sound. A 5% efficiency loss on your biggest ad set is worth more than a 90% loss on one that spends ten dollars a month, and a good audit says so plainly.

What Are The Four Or Five Gap Types?

Almost every wasted dollar falls into one of five buckets. Naming them is half the audit, because once you know the categories you know which exports to pull and which question to ask of each one.

1. Spend with nothing back. The clearest gap: an ad set or campaign that has real spend over the window and zero conversions, or a cost per acquisition so far above your target that it loses money on every sale. These are the lines you pause first. Claude finds them by reading spend against conversions and flagging anything below break-even, biggest spender first.

2. Audience overlap. Two ad sets targeting the same people, or two campaigns bidding on overlapping keywords, means you are in an auction against yourself. You pay more for the same click and split the learning data across both. This is common after an account has run for a year and nobody pruned the old targeting. Claude names the likely overlaps from the campaign and ad-set structure.

3. Creative fatigue. An ad that worked is shown to the same audience until the audience stops responding. Frequency climbs, click-through rate drops, and cost per result drifts up week over week. The fix is rotation, not a higher bid. Claude reads the trend over a longer window (90 days, not just the last 30) and flags creatives whose performance is decaying rather than holding.

4. The wrong clicks. Broad match and loose targeting buy traffic for terms you would never choose. The search-term report is where this hides: queries that are tangential or plainly irrelevant, eating spend with no chance of converting. Claude reads your campaign themes, then lists the search-term patterns most likely wasting money and the negative keywords to add first.

5. Broken tracking, the gap that hides the others. If conversions are not recorded correctly, every number above is wrong, and you optimize toward noise. This is the one that makes the other four invisible, so it gets its own section below. A real audit refuses to trust the data until it has checked that the account can actually see a conversion.

On Google specifically, there is a sixth place worth a hard look: Performance Max. PMax hides spend inside a single black-box campaign, so the audit has to go down to the asset-group level to see which groups, audiences, and placements are carrying the budget and which are quietly burning it. Without that asset-group detail, a PMax campaign looks like one number; with it, you can see the gap inside.

Which Prompts Find The Waste?

Export your campaigns first. From Google, pull the campaign report plus the search-term report and, if you run Performance Max, the asset-group report. From Meta, pull the campaign and ad-set report with spend, results, cost per result, and frequency. Set the window to the last 30 days for current performance and the last 90 for fatigue and trend. Then paste the relevant export under each prompt below. Each one is written to weight findings by spend, so the answer ranks by money.

Run them in order and you have walked the same path the installed skill walks: inventory the data, find the dead spend, find the overlap, find the fatigue, find the wrong clicks, look inside PMax, then collapse it all into a short ranked list of what to do this week.

Why Is Broken Tracking The Real Problem?

Because every other finding sits on top of it. If conversions are not tracked correctly, the spend-with-nothing-back finder will flag winners as losers, the overlap check will misread which ad set to keep, and the stop-doing summary will tell you to pause the wrong thing. Bad tracking does not just hide the waste. It actively points you at the wrong fix.

This is why a real audit checks tracking before it trusts a single number. The honest move, when conversion data looks missing or wrong, is to say so plainly: conversion tracking is not verifiable, treat every downstream number as suspect, fix this first. Claude will not invent a conversion figure to paper over the gap, and you should not let any audit do that either. Ask it directly: "Before you grade anything, does this data look like the account is recording conversions correctly, or are there signs the tracking is broken?" If the answer is the latter, that becomes your first fix, ahead of every gap below it.

How Do You Read The Grades?

The installed /ads-audit skill does not just list problems. It scores each active platform against a category checklist, turns that into a 0 to 100 number and a letter grade, then rolls the platforms into one account-wide grade weighted by how much each one spends. The grade is a way to compress a long audit into something you can act on and compare next month.

Read it in this order. Start with the headline account grade, because that is the budget-weighted summary. Then read each platform's grade, because a healthy Google account can hide a failing Meta one when they are averaged. Then drop into the findings under each platform, where every flag is marked pass, warn, fail, or not-applicable, with the export row that proves it and the fix attached. A check that cannot apply to your account (no Performance Max, no shopping feed) is removed from the math, never counted as a failure, so the grade reflects only what is actually running.

The single most useful habit is to read the math, not just the number. A good audit shows how it got to the grade: which categories scored low, how spend was weighted across platforms, which findings dragged the score down. A grade you cannot trace is a grade you cannot trust. When Claude shows the work, a B- becomes a to-do list ("fix tracking, prune two overlapping ad sets, add negatives") instead of a vague verdict.

What Can The Audit Do, And What Can It Not Do?

The most important thing to understand is that this is a read, not a remote control. Claude reads the data you export and tells you what to change. It does not reach into your ad accounts and change anything. That line matters, so here it is in full.

Read it as a fast, tireless media buyer who has read every row of your exports and never loses track of a number. The recommendations are only as good as the data you paste, and the changes still happen in your own ad accounts, by your own hand. That is a feature, not a limitation: nothing in your account changes without you deciding it should.

One-Off Prompt Or The Installed /ads-audit Skill?

The prompts above and the installed skill use the same model and ask the same questions. The difference is memory and structure. A pasted prompt reads one export and prints an answer that vanishes when you close the tab. The /ads-audit skill reads every platform you run, scores them, and saves a baseline file, so the next time you audit it opens with what changed since last time.

Use the prompts to feel the value in ten minutes. Install the skill when you want the gap tracked instead of re-discovered. The skill writes per-platform health lines to a ledger (Google health, Meta health, and the aggregate), so a re-run can say "Google up nine points since April, Meta down four because the pixel regressed" instead of starting the conversation over. The prompt is the taste. The skill is the record.

How Do You Run The /ads-audit Skill?

The skill ships inside Conversion OS, a free, open set of Claude skills. Installing the whole thing takes about five minutes and no coding. The /ads-audit skill itself is standalone, so it runs from your pasted exports with no further setup.

1. Open Claude Code or Cowork, then Plugins, then Marketplace, and add the Conversion OS repository.
2. Export your last 30 and 90 days from Google and Meta as described above, plus the PMax asset-group report if you run Performance Max.
3. Run /ads-audit and paste the exports when asked. It audits each platform in parallel and returns the graded report.
4. Read the headline grade, the quick wins, then the action plan. Make the changes in your ad accounts yourself.
5. Next month, export again and re-run. This time it opens with the delta against the baseline it saved.

If you also run the one-time /setup interview (about thirty minutes), the audit ties its findings to the targets you record, your cost-per-lead goal, your ROAS target, your real customer profile, so a flagged ad set is judged against what you actually want, not a generic benchmark.

What Are The Common Mistakes?

Trusting the numbers before checking tracking. If the account cannot see a conversion, every grade and ranking is built on sand. Ask Claude to verify tracking first, and treat a "not verifiable" answer as your number-one fix.

Auditing a window that is too short. Thirty days shows current performance but hides creative fatigue and seasonal drift. Pull 90 days too, or the audit misses the decay that is quietly raising your costs.

Ranking by alarm instead of by spend. A 90% loss on a ten-dollar ad set feels urgent and is not. Weight every finding by the dollars behind it, which the prompts above are written to do, so you fix the gap that actually moves the budget.

Treating Performance Max as one number. PMax hides the gap inside a single campaign. Without the asset-group report you are grading a black box. Always pull the asset-group detail before you judge a PMax campaign.

Auditing once and calling it done. The account drifts. A one-time cleanup decays within a quarter. Save a baseline and re-run, which is the entire reason the installed skill exists.

Asking Claude to change the campaigns. It reads exports; it does not touch your accounts. The plan is yours to execute. That separation is deliberate and keeps you in control of every change.

Methodology

The waste figure, roughly 20 to 30% of the average paid-search budget, is from 2025 PPC industry analysis and is the only outcome number this article claims. Conversion System publishes no client result numbers, so none appear here; the proof is the named research and the method shown. The prompts and the grading, baseline, and ledger behavior described above reflect what the open Conversion OS /ads-audit skill actually does, which you can read in the public repository: it reads Google and Meta data from exports, scores each platform against original category checklists, rolls them into one budget-weighted grade, writes an action plan and per-platform ledger lines, and saves a baseline that later audits chart against. It runs standalone from pasted exports with no account access, and it never invents a metric it was not given. The aim of this guide is to help you see the gap before you raise the budget.

Related: Claude as a business operating system · Run a free revenue audit with Claude · Run an SEO audit with Claude · Claude for lead generation and scoring