PPC Audit Template: Run Google Ads Audits for New Clients With AI

A prospect PPC audit should cover five sections in this order: conversion tracking, account structure, search queries, ads, and budgets with impression share. AI is useful for summarising the exports and drafting the one-page prospect summary, while a person must check the numbers and the tracking.

The order matters more than most templates admit. If conversion tracking is wrong, every number in the later sections is wrong too, and an audit that praises a "strong CPA" built on inflated conversions embarrasses you on the first call. So the tracking section comes first and can overrule everything after it. The rest is the template itself, the prompts that handle the tedious reading, and a filled-in sample for an imaginary garden centre with the summary you would send.

Follow me on Instagram@sagnikteaches

Before you open the account: access and exports

Ask for read-only access to the Google Ads account (Google's access-levels help page lists Billing, Read-only, Standard and Admin). Read-only is enough for everything here, and it is a better first ask than standard access, because the prospect can say yes without worrying that you will change anything.

Connect on LinkedInSagnik Bhattacharya

Then export, for the last 90 days unless the account is seasonal, in which case compare with the same period last year:

Subscribe on YouTube@sagnikteaches

1. Campaigns report, with cost, conversions, conversion value, impression share, lost impression share (budget) and (rank).
2. Search terms report, with cost and conversions, sorted by cost.
3. Keywords report, with match type and Quality Score columns.
4. Ads report, with ad strength and clicks.
5. Conversion actions list, with each action's status, source, counting option and whether it is primary or secondary.
6. Change history for the last 60 days.

Save them as CSV in a folder named for the prospect and the date. If you want the full onboarding questions that sit around this, the onboarding questionnaire tutorial covers them.

Section 1: Tracking, the section that can end the audit

Open the conversion actions list first and ask four questions of each action.

• Does it measure something the business wants? A purchase, an enquiry or a booked call, yes. A page view, a scroll or "added to basket", usually no, unless it is deliberately secondary.
• Is it set as primary? Google Ads distinguishes primary conversion actions, which are reported in the Conversions column and used for bidding when their goal is, from secondary ones, which appear only in All conversions. A weak action set as primary distorts both reporting and bidding.
• Is it counted correctly? The counting option is either Every conversion or One conversion per ad click. Google's own help page suggests Every for sales and One for leads, so a lead form set to Every is a finding.
• Is it firing? The status column shows recording problems. A tag that last fired six weeks ago means missing data.

Two further checks: whether the same event is counted twice (an imported analytics goal alongside a Google Ads tag for the same form), and whether conversion values are real prices or a fixed default.

AI does very little here. You can paste the conversion actions table into a chat tool and ask it to flag oddities, which is a quick second opinion, but you need to check the tag itself in the browser, with a tag-debugging tool, on the real site. Treat any AI "looks fine" on tracking as unverified.

Section 2: Structure

Structure tells you how much thought has gone into the account. Look for these in order.

• Campaign split by intent. Brand search, non-brand search and any automated campaign types should be separate, with their own budgets. A brand campaign hidden inside a non-brand one makes the average CPA look better than it is.
• Ad group themes. One ad group with 140 mixed keywords is a warning. Tight themes allow tailored ads and cleaner queries.
• Match type mix. Note the proportion of broad, phrase and exact, and whether broad match is paired with sensible automated bidding and a negative keyword list.
• Settings. Locations (Presence, or the default Presence or Interest), languages, networks (Search partners and the Display network on a Search campaign), ad schedule and device adjustments.
• Automated types. Whether Performance Max campaigns or AI Max for Search campaigns are running, and whether the team can say what each is meant to do.

The settings check is where free templates earn their keep. Google's advanced location options page says targeting includes both physical presence and locations of interest by default, so a campaign left on Presence or Interest can serve ads to searchers who are only reading about the area, which is a common leak for local businesses. A garden centre that sells click-and-collect only usually wants Presence.

Section 3: Queries, where the money leaks

Sort the search terms report by cost and read the top 50 yourself first. Then bring in AI for the long tail, where a person would get bored. This is the step AI does best.

You are auditing a Google Ads search terms export for a garden centre.
The business sells plants, tools, compost and garden furniture, and
offers click-and-collect. It does NOT offer landscaping, garden waste
collection or jobs.
Classify each search term as Relevant, Irrelevant or Unsure.
For Irrelevant, give a one-word reason (jobs, free, service, competitor,
other) and suggest a negative keyword and match type.
Output a table: term | cost | conversions | class | reason | suggested negative.
Do not change the numbers. If a term is ambiguous, say Unsure.

Illustrative output (five rows):

What you fix: read every Irrelevant and Unsure row. A tool might suggest a broad-match negative "free", which would also block "free delivery garden centre", a term that converts. Always check suggested negatives against your converting terms before you recommend them. For automated campaign types, the AI Max negative keywords tutorial goes deeper.

Do the arithmetic in a spreadsheet, not in the chat. Sum the cost of Irrelevant rows with a formula, because language models do arithmetic unreliably, and a prospect call is no place to find that your waste figure was $90 out.

Section 4: Ads

• Number of active responsive search ads per ad group. Google's responsive search ads help page recommends at least two responsive search ads with Good or Excellent Ad Strength per ad group. Count how many ad groups have only one.
• Ad strength and asset use. Ad Strength runs from Incomplete through Poor, Average and Good to Excellent. Look at how many headlines and descriptions are filled and how many are pinned: Google's Ad Strength page says pinning limits the system's ability to test combinations and suggests pinning only one or two essential headlines.
• Message match. Does the headline reflect the search? Does the landing page match the headline?
• Quality Score components. Google's help page lists three: expected click-through rate, ad relevance and landing page experience. Note which one is "Below average" most often, because that points at ads, keywords or the page.
• Assets. Sitelink, callout, call and location assets: note which are set up, which are missing and whether sitelinks point at real, distinct pages.

Use AI to draft sample improved headlines for the audit summary, with the rule that every claim in them must come from the prospect's own site. Our responsive search ad generator tutorial covers the character limits and the review step. In an audit you only need two or three examples to show the gap.

Section 5: Budgets and impression share

Google Ads reports impression share: the share of eligible impressions your ads actually received. Two columns explain the rest. Search lost IS (budget) is the share of time ads did not show because the budget ran out, and it is available at campaign level only; Search lost IS (rank) is the share lost to poor Ad Rank. The split tells you what to do.

• High lost impression share (budget) on a profitable campaign: the budget is the limit. Raising it is the cheapest growth available.
• High lost impression share (rank) on a campaign: bids or quality are the limit. More budget will not help.
• Budget spread: compare budget share against conversion share by campaign. A campaign taking 40% of spend and delivering 10% of real conversions is the first place to look.

Check also the bidding strategy: is a target set that the data can meet, and has it been changed repeatedly? The change history will show that, which is why you exported it.

A filled-in sample: a garden centre with $6,000 a month

Everything in this sample is illustrative. Say a garden centre spends $6,000 a month across three campaigns and reports 106 conversions, an apparent cost per conversion of about $57.

Tracking finding. The conversion actions list shows three primary actions: Purchase (52), Enquiry form (16) and Add to basket (38). Add to basket is primary and counted every time. Removing it from the primary set leaves 68 real conversions, so the true cost per conversion is $6,000 / 68 = about $88, not $57. The bidding algorithm has also been optimising towards basket additions.

Query finding. In the non-brand campaign the AI classification plus a manual check puts $510 of the $3,900 on irrelevant terms with no conversions. That is 13% of non-brand spend, roughly $6,100 a year at the current rate.

Budget finding. The brand campaign loses 18% of impressions to budget, while it is the cheapest source of conversions. A modest budget increase there is likely to be worth more than the same money anywhere else, though a person should confirm it against the real conversion figures.

Ad finding. The non-brand ad group has one responsive search ad with three pinned headlines and a "Poor" strength rating, against Google's recommendation of at least two ads rated Good or Excellent.

The prompts that assemble the audit

With the five sections done, one prompt compiles your own notes into the working document. Give it your findings and not the raw exports, so it formats and does not invent.

Below are my audit findings for a prospect, by section, with figures I have checked.
Write a one-page summary for the business owner, with:
- A two-sentence overview
- The three most important findings, each with one sentence of evidence
- One recommended first step per finding
- A closing line offering to walk through the full audit on a call
Plain language, no jargon, no promises of results, no criticism of
whoever set the account up. Use only the figures I have given you.

The line "use only the figures I have given you" is the safeguard. Without it, a model may round, compare to "typical" benchmarks it has made up, or promise a percentage improvement. Read the summary against your own notes line by line. Once that is done, the human fixes are usually small: tightening a sentence, removing an adjective, checking the $88 against the spreadsheet.

What the one-page summary looks like (illustrative)

Your Google Ads account: what we found

Overview: Your account spends $6,000 a month and reports 106 conversions.
On closer inspection about 38 of those are basket additions, not sales or
enquiries, so the real cost per result is nearer $88 than $57.

1. Tracking counts basket additions as conversions.
   Evidence: 38 of 106 reported conversions. First step: set it as secondary.
2. About 13% of non-brand spend goes on searches unrelated to your shop.
   Evidence: $510 last month on terms such as jobs and waste collection.
   First step: add a negative keyword list.
3. Your brand campaign is limited by budget.
   Evidence: it misses 18% of possible impressions. First step: test a
   higher daily budget for 30 days.

Happy to walk through the full audit on a call.

Scoring each section so findings rank themselves

A prospect cannot act on twenty observations, and you cannot decide which three to lead with by feel. Score each section red, amber or green using the same short rules every time, then lead with the reddest sections that have money attached.

The thresholds are rules of thumb for ranking your own findings, not industry standards, so adjust them as you learn what matters for your clients. The value of a fixed rubric is consistency: two people on your team auditing different prospects will describe a similar account in the same terms, and your prospects can compare one audit with another.

A lead-generation account behaves differently

The garden centre sells online, so conversions are mostly purchases. A lead-generation business, such as an optician taking appointment requests or a veterinary practice taking bookings, changes three parts of the template.

• Tracking: form submissions and calls are the conversions, and calls need their own tracking. Check call length thresholds for call conversions and whether calls from ads are counted at all. Count each form once per click, never every time.
• Quality of leads: a lead is not a booking. Ask whether bookings or attended appointments are fed back to Google Ads. If not, bidding optimises for form fills that include spam and wrong-service enquiries. You can only mention this as a gap, since the data sits in the client's own systems.
• Queries: irrelevant searches look different: "free eye test" for a private practice, or "free vet advice" for a paid one. The classification prompt needs the business's real exclusions, so write them into the prompt each time.

An illustrative case: a veterinary practice spends $2,400 a month and reports 60 enquiries, a cost of $40. A review of the form shows 9 spam submissions and 11 asking for a service the practice does not offer. The tidy number is $2,400 / 40 = $60 per usable enquiry. The reported figure was flattering by a third, which is exactly the kind of finding that makes a prospect trust the rest of the audit.

What AI gets wrong in an audit

• It trusts the labels. If an action is named "Purchase" but fires on a page view, a model reading the list will believe the name. Only the tag check finds that.
• It treats benchmarks as facts. Ask for "average click-through rates for the industry" and you will get plausible numbers with no source. Keep benchmarks out of the prompts and out of the summary.
• It over-recommends negatives. Single words such as "free", "cheap" or "near" sound like waste but can sit inside converting searches, so each suggestion needs a check.
• It writes confident summaries of thin data. On a $300 a month account, 90 days of data holds a handful of conversions, and any finding about conversion rate by campaign is noise. Add a rule to your template: below roughly 30 conversions in the period, state that the data is too thin for conclusions about performance, and limit the audit to tracking, structure and waste.

Time and cost of the audit

Plan the time before you offer audits as a sales tool, because a free audit that takes four hours per prospect is expensive. An illustrative breakdown for a mid-sized account:

The saving of about 75 minutes comes almost entirely from the reading-and-writing steps. The tracking and structure checks do not shrink, and should not. At an internal rate of $60 an hour, 1.25 hours is $75 per audit; across 12 audits a quarter that is $900 of time. The figures are illustrative, so time your own first few audits and replace them. Note also the pricing of the tool you use: if the team already has a business plan for a chat assistant, the extra cost is nil, so this belongs in your existing stack rather than a new subscription.

From audit to proposal

The audit's last job is to make the next conversation easy. End the call with a choice, not a pitch: fix the tracking first (a one-off piece of work), then clean up queries, then restructure. A scoped proposal follows from the red sections in order. If you are putting that into a document, the scope-of-work tutorial shows how to turn findings into scoped, priced steps. Keep the audit and the proposal separate documents: the audit is evidence, the proposal is a plan, and mixing them makes both weaker.

What not to put in a prospect audit

• Promised results. "We'll cut your costs by 30%" is a claim you cannot support before access and testing.
• Blame. The previous agency or in-house marketer may be on the call. Describe what you found, not who did it.
• Every finding. Three well-evidenced points beat twenty. Keep the long list internal.
• Raw AI output. Prospects can tell. A summary that reads like a template fails the first test of an audit, which is showing that someone looked at this account.

How to check the audit is sound before you send it

1. Recompute every figure in the summary from the exports, in a spreadsheet.
2. Check each suggested negative keyword against the converting search terms.
3. Confirm the tracking finding on the live site with a tag-debugging tool, not from the report alone.
4. Ask a colleague to read it as the prospect and point out any sentence they could not act on.

Over time, keep a log: date, prospect, hours spent, whether they replied. After ten audits you will know whether the hours convert into conversations, and whether the AI steps saved enough to be worth keeping. For the reporting side after you win the account, the client reporting tutorial follows on naturally. If you'd rather map this with someone, my AI implementation consultation covers it on a 1:1 call.

Questions agencies ask about AI-assisted PPC audits

How long should a prospect audit take?

In the illustrative breakdown here, a mid-sized account takes about four hours by hand and closer to three with AI drafting: half an hour collecting exports, about two hours on the five sections and 20 minutes on the one-page summary. A single small campaign takes far less. Time your first three audits and set your own standard.

What access do I need from a prospect?

Ask for read-only access to the Google Ads account, which Google Ads provides as an access level. That is enough to view settings and export reports, and it lets the prospect see you cannot change anything. Ask for the Analytics and tag manager views too if you want to check tracking.

Should I give the prospect the full audit?

Give them a one-page summary with the three biggest findings and the evidence for each. Keep the long working document for yourself, or for after they sign. The summary should prove you understand the account without handing over a free fix list.

Can I paste account exports into ChatGPT or Claude?

Only if the prospect has agreed and your tool plan keeps business data out of model training by default. Strip anything personal. Search term and campaign exports rarely hold personal data, but spend and conversion figures are commercially sensitive, so treat them as confidential.

Further reads

• Performance Max Reporting: Explain Asset Results to Clients With AI — Reading Performance Max asset reports once the audit is done.
• Can AI Manage Your Google Ads Without an Agency? — What the client could do in-house, useful for honest audit advice.
• AI Media Planning for Small Agencies: Channel Mix and Budget Splits — Where search sits within a wider channel plan.
• AI Brand Audit: A Template Branding Studios Can Sell — How another kind of agency structures an audit template.
• Best AI for Advertising: A Small Agency's Creative Testing Pipeline — Testing ad variations after the audit's ad findings.
• AI Tools for Agencies: What a Five-Person Team Pays Each Month — What your audit and reporting stack costs per month.
• Performance Max vs Standard Shopping for a Small Online Store — A small-store way to choose between Performance Max and standard Shopping: budget floor, six-week test, fair comparison and a decision rule.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google Ads Help pages on access levels, primary and secondary conversion actions, conversion counting options, Quality Score, impression share data, advanced location options, responsive search ads, Ad Strength, AI Max for Search and sitelink, callout, call and location assets, read 4 November 2026; the sample account is illustrative.