Costly Google Ads Mistakes Small Businesses Make in 2026 (and How to Audit for Them)

Almost every account we take over is losing money in the same handful of places. Not exotic places. The same settings, the same structural shortcuts, the same habits, account after account.

This is a rebuild of an article I first wrote in 2016, and the reason it needed rebuilding is that the leaks have moved. In 2016 the expensive mistakes were match types and negative keywords. In 2026 the expensive mistakes are automation you were not ready for, a budget pacing rule Google quietly changed on 1 June, and blended metrics that hide a thirty-fold difference in what you are paying.

To make sure I was not just writing from my own accounts, I pulled the transcripts and comment threads from the eight highest-engagement Google Ads teardowns published in the last twelve months, by practitioners who each manage or coach at meaningful scale: Darren Taylor, Ben Heath, Aaron Young at Define Digital Academy, Jyll Saskin Gales, and the team at Grow My Ads. Where they agree, I have said so. Where our own account data says something different, I have said that too. Every dollar figure below comes from accounts we manage: $1,168,221 of spend, 478,270 clicks and 24,020 conversions over the 12 months to September 2026.

At the end there is a section of copy-paste AI prompts, with the guard rails that stop the AI confidently telling you something wrong. That part matters more than it sounds. Most people who need an audit cannot tell a good audit from a bad one, which is exactly the condition an unguarded language model exploits.

The change most small advertisers missed this year

Start here, because it is the newest and it moved real money.

Until June, if you ran a campaign with an ad schedule that turned certain days off, Google paced your budget against the days you were actually eligible to serve. From 1 June 2026, Google paces toward the full monthly limit of 30.4 times your daily budget regardless of how many days the schedule allows. The daily cap of twice your daily budget has not changed.

The arithmetic is worth doing slowly, because the effect is larger than it sounds.

That is roughly 38% more monthly spend on an untouched campaign. It compounds with an over-funded budget, because Maximize Conversions is designed to spend the daily budget, so with no target CPA in place the extra headroom goes straight into the same auctions you were already winning.

Two details people get wrong. Hour-of-day scheduling is not affected, only whole days turned off. And the pacing window is the calendar month, not a rolling window, so if you edit the budget mid-month the cap resets to spend so far plus the new daily budget times the remaining calendar days.

What to do about it: if you run any campaign with days switched off, divide your intended monthly spend by 30.4 and set that as the daily budget. For $1,100 a month on a weekdays-only campaign, the daily budget is about $36, not $50.

The mistakes, ranked by what they cost

Before the detail, here is the shape of it. The ranking is by what we typically recover in the first ninety days of an account takeover, not by how common the mistake is.

The rest of this article works through them in four groups, because that is how they actually cluster: foundations, settings leaks, structure, and behaviour.

Foundation mistakes

Conversion tracking that is wrong, partial, or measuring the wrong thing

Every practitioner in this research put this first, and one of them called it the number one silent killer. It is silent because the account still reports numbers. They are just not numbers about your business.

The failure modes we find most often, in order of frequency:

How to check it in ten minutes. In the campaigns report, not the overview, add a segment for conversion action. You will immediately see what is actually inside the conversions column, whether a campaign has its own goal, and whether anything in there is a page view. Then compare the last 30 days of conversions in Google Ads against the same 30 days of real enquiries in your inbox, phone log or CRM. If those two numbers are not close, stop optimising and fix the tracking first.

The fix that separates the accounts that work. Getting conversions tracked is the floor, not the goal. The accounts that consistently beat their category are the ones sending signed clients back to Google as the optimisation target, not enquiries. We wrote about what happens when you skip that step in why most businesses waste 70% of their paid leads, and about the measurement side of it in online lead generation.

Trusting the numbers in the interface without questioning them

Data that is too good to be true usually is. So is data that is too bad to be true.

A 50% purchase conversion rate is a tracking fault rather than a great campaign. Even 5% is an excellent ecommerce rate. A campaign reporting zero conversions for 30 days while your reception has been answering calls is untracked rather than failing.

There is a variant of this that shows up constantly in comment threads from real advertisers: Google Ads reports several clicks, analytics reports one user, and the advertiser concludes the platform is fraudulent. Usually it is neither. A click counts at the moment of the click. A session counts when the page loads and the tag fires. Slow pages, bounces before load, consent banners and ad blockers all sit in that gap. The gap is normal. The gap being enormous is a tracking problem worth chasing.

Judging the account on blended averages

This is the mistake I would most like to delete from the industry, because it makes intelligent people pause the wrong campaigns.

Here is one personal injury account we run, over the 12 months to September 2026.

Thirty-four times between the top and the bottom. The blended $3.96 is a real calculation and a useless fact. If you set a budget off it you will fund about an eighth of the non-brand clicks you thought you were buying. If you judge non-brand search against it you will pause the only campaign creating new demand and keep the one harvesting people who already knew the firm's name.

The same distortion appears geographically. In our family law client's account the identical service costs $13.45 a click in Brisbane and $21.91 on the Gold Coast. There is a fuller version of this argument, with all sixteen industries we have data for, in how much Google Ads actually costs.

The check: segment by network, then by campaign type, then by brand versus non-brand. If you cannot separate brand traffic from non-brand traffic, that is the first structural change to make, ahead of everything else in this article.

Only ever looking inside Google Ads

Attribution has been degraded for years by privacy changes and cookie restrictions, and it is not coming back. Google Ads will under-report. So will everything else, differently.

The specific trap is Performance Max and Demand Gen, where a lot of the value shows up as assisted or gets attributed elsewhere. A PMax campaign can look dreadful in the Google Ads column and be the only thing that changed in a month where total revenue moved. Before you pause it, look at analytics, look at your CRM, and look at the total.

Settings leaks

These are the cheapest mistakes to fix and the most embarrassing to find. Experienced practitioners miss them too, usually on an account they inherited and never re-checked.

Two of those deserve more than a table row.

Auto-apply recommendations is on by default on new accounts. Every practitioner in this research independently named it. It is the single setting most likely to undo work you did deliberately, because it will take an exact match campaign you built carefully and move it toward broad, or shift a bidding strategy you chose for a reason. Turn it off because Google does not have the context of your business, particularly on target CPA and target ROAS, where a goal set too tight acts as a brake on the whole account and a goal loosened automatically acts as an accelerator you did not press.

Google Ads reps. This one is uncomfortable to write, because it should not be true of a platform's own support. The consistent experience across practitioners, and ours, is that unless you are dealing with a senior account manager who can give you beta access and genuine competitor insight, the recommendations you get from the lower tiers are optimised for platform spend and feature adoption rather than your cost per acquisition. The most common damage we see is a small, controlled, profitable search campaign having AI Max or broad match switched on at a rep's suggestion, and the search terms report going wild within a fortnight.

Structure and strategy mistakes

Turning on automation before you have data to steer it

Broad match, Performance Max and AI Max are hungry rather than bad. Each of them replaces a control you used to hold with a prediction made from your conversion data, which means the quality of your conversion data is the quality of the campaign.

The failure pattern is consistent: a small advertiser with under $1,000 a month and a handful of conversions turns on broad match, gets scale immediately, and cannot tell which part of the new traffic is real because there was never enough signal to learn from. One advertiser in the research described running a search campaign with AI Max, getting 51 conversions for about 578 euros, and finding more than half were junk. That argues for having a working search campaign first and something in your tracking that distinguishes a lead from a qualified lead.

The order that works, and it is an order rather than a menu: exact match search, then phrase, then broad in a single contained ad group, then Performance Max once search is genuinely performing, then AI Max as an expansion test on top of that. Skip a rung and you are paying to find out what the previous rung would have told you for less.

If you are on a small budget and wondering whether any of this applies to you, we went through the specific dynamics of that in why a small Google Ads budget is harder to manage.

Budgeting without reference to your own cost per click

A daily budget is an arithmetic decision rather than a comfort one. Ask whether the budget buys enough clicks for the campaign to produce a readable result, not what you can afford to lose.

Here is what $1,500 of media buys across industries we actually run, which is the fastest way to see why the same budget is generous in one market and pointless in another.

† This account is several years old and was constrained by an advertiser verification problem, so its conversion rate reflects that constraint rather than the category. Treat it as an illustration rather than a benchmark.

Personal injury sits off the bottom of that table. At $33.09 a click on non-brand search, $1,500 buys about 45 clicks and roughly four enquiries, which is not enough to conclude anything in a month.

Two different failures come from getting this wrong. Under-budgeting means your daily budget cannot buy enough clicks in an expensive auction, so the campaign starves and you conclude the channel does not work. Over-budgeting is less obvious and just as expensive: give a campaign more budget than the available demand supports and Google has to honour the budget, so it bids higher for the same traffic you were already getting. You do not get incremental volume, you get an inflated cost per click.

That second one connects to something worth saying plainly, because it is the most common false hope in small business advertising. Opportunity is not infinite. If you already hold high impression share on your core terms, more budget will not find more customers, because there are only so many people searching. Out of a hundred people searching for a plumber near them today, a large share will not contact anyone today, and no budget changes that. Growth past that point comes from expanding the targeting, not from raising the number.

Ad groups that are not segmented by intent

This is the structural mistake with the widest blast radius, because it silently raises the price of every click in the account through Quality Score.

The principle is simple. An ad group should contain keywords that can honestly share one ad. A cleaning company offering residential cleaning and office cleaning has two different buyers, two different deal sizes, and two different ads to write, so that is two ad groups and possibly two campaigns so the budgets and bid strategies can differ. Residential cleaning and home cleaning are synonyms for the same job, so they belong together.

Roughly five to fifteen keywords per ad group is a sane working range rather than a rule. Three is fine. Eighteen is fine. The test is whether one ad can speak honestly to all of them.

The old practice of single keyword ad groups is over. Splitting "car service company", "car service garage" and "car service near me" into three ad groups gives you three ads saying the same thing and thirds of the data you need. Over-segmentation is now as expensive as under-segmentation.

There is a fuller treatment of the Quality Score mechanics, and why this is the cheapest lever most accounts never pull, in the Quality Score guide.

While we are here: Quality Score does matter, and the argument that it does not is a misreading. The 1 to 10 number in your keyword report is a diagnostic and not an auction input. Your actual ad quality, which Google calculates and does not show you, absolutely is an auction input and determines your Ad Rank. Those are two different things wearing similar names. Improving ad relevance, click-through rate and landing page experience lowers what you pay per click. Chasing the number itself from a seven to a nine is where the returns run out.

Writing ads that look exactly like your competitors' ads

Look at any commercial search results page in your market. The ads are close to interchangeable. Same claims, same structure, often the same words, and increasingly the same words because everyone is generating copy from the same models trained on the same winners.

If your ad is indistinguishable from the other five, the click comes down to position and luck, and position is something you are paying for.

The practical version of this: write from your actual differentiators and the problems your customers describe, not from the keyword. Keyword relevance is one pillar of a good ad, not the whole building. Then deliberately keep one headline in five as a higher risk option, something angled differently from everything else on the page. Most of those will lose, and Google will stop serving them cheaply enough. The occasional one that wins moves the account more than any bid adjustment will.

Use your assets properly while you are in there. Callouts, structured snippets, sitelinks and images take up more of the results page, which lifts click-through rate, which lifts ad quality, which lowers your cost per click. The most commonly ignored recommendation in Google Ads accounts is the one asking you to make headlines and descriptions more unique, which tells you something about where the industry is.

Sending every click to the homepage

If your ad promises something specific and the landing page is a general homepage, you have paid for a click and then asked the visitor to do the work of finding what you promised. Message match between ad and page is one of the most reliable conversion rate improvements available, and it costs a page rather than a budget increase.

This also sets a boundary on what advertising can do. Ads amplify what already works. A weak offer, thin credibility, poor reviews or a 10% gross margin are not problems Google Ads can solve, and paying to put them in front of more people gets you to the answer faster and more expensively. We wrote about the landing page half of this in detail in lessons from optimising 400 Google Ads landing pages.

Behavioural mistakes

Optimising every single day

This is the one that surprises people, because it feels like diligence.

Daily bid changes, budget adjustments, pausing and unpausing the same keywords, flicking campaigns on and off. Each change of consequence puts smart bidding back into a learning phase, and it makes every change unmeasurable, because you can never isolate what the last one did. The consistent experience among practitioners is that accounts making half the changes outperform accounts making twice as many, and when an account has no consistency in its change history that is usually the whole diagnosis.

Adding negative keywords sits outside this. Negatives are low-disruption and should be part of a weekly rhythm. Bid, budget, targeting and structure changes are the ones that need a reason and a window to be judged in.

A sane cadence looks like this.

Set and forget

The mirror image, and equally common in accounts that were built by someone competent and then left alone. Search behaviour shifts, competitors enter and leave the auction, Google changes pacing rules in June without telling you, and the search terms report drifts. A campaign that was well-built eighteen months ago and untouched since has been ageing the whole time.

Expecting the first campaign to work

Most individual campaigns do not become profitable. That is the normal shape of the channel, not a failure of it, and understanding that is what separates advertisers who persist from the ones who spend $2,000, get nothing, and conclude the platform is a scam.

The economics are asymmetric. Ten campaigns at $500 each that go nowhere cost you $5,000. The eleventh, if it works at a 4x return and you scale it to $100,000 of spend over a year, returns $400,000. The losses are capped at what you chose to test with. The winner is not capped. This is the same structure as any portfolio of small bets, and the mistake is treating each test as a verdict on the channel rather than as one of the bets.

What this does not mean is that you should burn budget without a hypothesis. It means you should test small, deliberately, one variable at a time, and be willing to scale hard when something works rather than leaving it at the budget you tested with.

Not managing the expectations of whoever is watching

If you have a client, a boss or a business partner, part of the job is explaining conversion lag, why you do not pull reporting on the first of the month, and why the account was not touched this week. Every practitioner in this research raised it, and the comment threads are full of freelancers describing clients who want conversions in week one on a small budget and who pause campaigns monthly. That is a communication problem being expressed as a performance problem, and it will end the engagement before the account gets a fair run.

We wrote about the version of this that shows up on the sales side in stop blaming your marketing agency.

Running the audit with AI, accurately

Here is the part I was asked to add, and the part I want to be careful about.

A language model is genuinely good at this work. It reads a search terms export faster than you do, it does not get bored at row 400, and it will spot patterns in ad copy you have stopped seeing. It is also willing to be confidently wrong, and if you do not know Google Ads well enough to audit your account, you do not know it well enough to catch a plausible wrong answer.

So the method matters more than the prompt. Four rules make the difference.

Prompt 1: search terms triage

The highest value use, and the one that scales worst by hand. Run it in two steps.

Step one, context:

I run a Google Ads campaign for a business that [describe what you sell, who buys it, and where you serve]. I am pasting the exact keyword list I am actively bidding on below. Read it so you understand what I am targeting. Do not analyse anything yet and do not give me recommendations. Reply only with a one paragraph summary of what you believe my business does and who it serves, so I can confirm you have understood before I give you the next task.

[paste keyword list]

If the summary is wrong, stop and correct it. Everything after this inherits that understanding.

Step two, the triage:

Here is my search terms report export. For every search term, assign it to exactly one of three categories: RELEVANT, POSSIBLY IRRELEVANT, or CLEARLY IRRELEVANT. Judge relevance against the keyword list and business description above, not against general intuition.

Return a table with columns: search term, cost, conversions, category, and a one sentence reason for the category. Sort by cost descending within each category. Do not suggest negative keywords yet, do not recommend bid or budget changes, and do not summarise. If a term is ambiguous, place it in POSSIBLY IRRELEVANT rather than guessing.

[attach search terms CSV]

The three categories are the guard rail. Read the RELEVANT list first: if the model has put something in there that you know is wrong, its judgment on this account is unreliable and you should tighten the context before trusting the rest. The CLEARLY IRRELEVANT list with high cost and zero conversions is where your negatives come from. The middle bucket is the one that needs you.

One caution that comes straight from practitioners and matches what we see: if more than about 10% of your search terms need negating, negatives are the wrong fix. You have a keyword, match type or bid strategy problem underneath, and whacking moles will not reach it.

Prompt 2: the settings leak check

I am going to paste my Google Ads campaign settings. Check them against this list and tell me, for each one, the current value and whether it is a likely problem: networks (search partners and display), location targeting method (presence versus presence or interest), bidding strategy and any target CPA or target ROAS value, daily budget, ad schedule, and campaign level conversion goals.

For each item, output: setting, current value, risk level as HIGH, MEDIUM or NONE, and one sentence on what it costs me if it is wrong. Flag it as HIGH only if it would directly cause wasted spend or spend outside my target area. Do not recommend changes to anything you were not given data about, and say explicitly if a setting I have listed is not present in what I pasted.

[paste settings]

The last instruction is the important one. Models fill gaps. Forcing it to declare missing data rather than assume it is what stops you acting on an invented finding.

Prompt 3: the cost per acquisition diagnosis

When your cost per acquisition jumps, there are only two possible causes: your cost per click went up, or your conversion rate went down. Everything else is downstream of one of those. This prompt forces that structure.

My cost per acquisition changed from $X to $Y between [date range A] and [date range B]. I am pasting campaign level data for both periods including impressions, clicks, cost, click-through rate, average CPC, conversions and conversion rate.

Step one: determine whether this was driven by a change in cost per click, a change in conversion rate, or both, and show the arithmetic. Do not proceed to step two until you have stated which it is.

Step two: if it was cost per click, list which of these six causes the data supports and which it rules out: bid or bid strategy changes, ad quality changes, competitor activity, search demand changes, changed search term matching, ad copy changes. If it was conversion rate, list which of these the data supports: landing page changes, traffic mix changes, offer or pricing changes, tracking changes, seasonality.

For each cause, state whether the data I gave you can confirm it, rule it out, or is insufficient. Do not speculate beyond the data. Tell me exactly what additional export would resolve each insufficient case.

[paste both periods]

The value here is the elimination rather than the answer. Being told that four of six causes are ruled out and two need a change history export is a far more useful morning than being handed a confident guess.

Prompt 4: the differentiation check

Below are my current Google Ads headlines and descriptions, followed by the ad copy of the [number] competitors appearing for my main keywords, which I collected from the search results page.

Identify every claim, phrase or angle that appears in both my ads and at least two competitors' ads. List them as a table: phrase, how many advertisers use it including me. Then list every claim in my ads that no competitor makes.

Do not write me new ad copy. I want to see the overlap first.

[paste your ads, then competitor ads]

Almost every business that runs this discovers their entire ad is in the shared column. That finding is worth more than any generated alternative, because it tells you the problem is your positioning rather than your copywriting, and no model can fix positioning for you.

Prompt 5: the budget pacing check, post June 2026

I have a Google Ads campaign with a daily budget of $X and an ad schedule that runs only on [days]. Since 1 June 2026 Google paces campaigns toward 30.4 times the daily budget per month regardless of how many days the schedule allows, with a hard cap of twice the daily budget on any single day.

Calculate: my current monthly pacing target, what it would have been under the old day-proportional pacing, the percentage difference, the implied spend per active day, and the daily budget I should set to land on a monthly spend of $Z. Show the arithmetic for each.

That one is pure arithmetic, which is exactly where a model is reliable and where a busy business owner is most likely to skip the step.

What not to ask it to do

Do not ask it to decide whether to pause a campaign. Do not ask it whether your budget should go up. Do not ask it to write your negative keyword list unsupervised and paste the output into the account. Those are judgment calls that depend on things the model cannot see: your margins, your capacity, what a client is worth to you over three years, whether you can handle more work next month.

The model's job is to read more rows than you can and hand you a structured shortlist. The decisions are still yours, and if you would rather they were not, that is the honest argument for hiring someone. Which is roughly where the next article picks up: the questions to ask before hiring a PPC agency or freelancer.

Where to start if you only do one thing

In order, and stop at the first one you fail.

If you would rather have someone go through all of it properly, that is what our Google Ads management work is. If you are a law firm, the practice-area specific version of this sits under legal marketing.