127 Google Ads Mistakes That Waste Money (And How to Find Each One in Your Account)

The short answer

A list of 127 mistakes reads like 127 separate problems. It is not. Almost everything below is a symptom of one of three root causes, and naming the cause is what turns a checklist into a diagnosis.

The first cause is conversion data nobody verified. The account reports outcomes, the reports look plausible, and every decision — manual or automated — is made from that number. If the number is wrong, the account can be managed impeccably and still lose money, faster the better it is managed.

The second is a mismatch between the intent behind a query and what the account actually offers when someone arrives. This shows up as the gap between the keyword list and the search terms report, and again as the gap between the ad and the landing page. Both gaps are visible in the interface and both are routinely unread.

The third is unchecked automation. Automated bidding, broad match, auto-applied recommendations and campaign types that select their own inventory all work by steering toward a target. Given a correct target they are useful; given a wrong one they are efficient at producing the wrong outcome, and their efficiency is what makes the damage compound.

Start from the symptom, not the checklist

Reading 127 items top to bottom is the slowest way to use this. Start from what the account is actually doing wrong, use the table to get to the right group, and work the relevant section properly rather than skimming all eleven.

The groups below are ordered roughly the way I work an unfamiliar account: structure and settings first because they explain what everything else is measuring, then the query layer, then measurement, then the levers. Numbers run continuously from 1 to 127 across the eleven sections.

Group 1 — Account structure and campaign settings (1 to 15)

Structural mistakes share a signature: the reported performance of one thing is quietly funded or contaminated by another. A campaign line in a report is only meaningful if everything inside it shares a goal, an audience and an economic profile. When it does not, the average hides the failure, and the failure is what you were trying to find. These show up as campaigns whose results look stable while their composition shifts underneath.

1. Search and Display running inside one campaign because the Display Network opt-in was left on, so Display impressions consume budget under a line the report labels Search.
2. Search Partners left enabled without ever segmenting the network column, so the account has no idea what that inventory contributes separately from Google search itself.
3. One campaign holding every product or service line, letting budget flow to whichever theme wins the auction cheapest rather than the one the business needs volume in.
4. Ad groups holding a dozen loosely related keywords against one set of ads, guaranteeing that no ad closely matches most of the queries that trigger it.
5. No campaign naming convention, which makes filtered reporting and automated rules impossible later and turns every handover into archaeology.
6. The same keyword live in two campaigns with no coordination, so the account competes against itself and the manager reads the split as two underperforming lines.
7. Brand and non-brand traffic combined in one campaign, where brand conversions subsidize the reported performance of everything else and hide which is working.
8. New campaigns launched from a duplicated template nobody re-read, inheriting network, location and audience settings that made sense for the original and not for this one.
9. A campaign per keyword built out of habit, spreading conversion data so thin across the account that no automated strategy has enough signal in any one place to model.
10. Performance Max running over the same products as a Search campaign with no exclusions, so the overlap is invisible in both reports and each takes credit for the other's demand.
11. Shopping and Performance Max competing for identical inventory with no priority scheme, which turns the question of which one serves into an auction artifact rather than a decision.
12. Ad rotation left on optimize while a controlled creative test is running, so the platform picks a winner before the variants have comparable delivery.
13. Tracking templates or final URL suffixes applied inconsistently across campaigns, so a subset of traffic arrives untagged and shows up as direct in every downstream report.
14. Adopting each new campaign type as it is offered without deciding what it is supposed to replace, ending with overlapping types that all claim the same conversions.
15. Removing campaigns and ad groups rather than pausing them, destroying the record that would tell the next manager this was already tried and why it stopped.

Group 2 — Keywords and match types (16 to 30)

The keyword list is a hypothesis about what people will type. The search terms report is what they actually typed. Every mistake in this group is a version of the same failure: managing the hypothesis and never reading the evidence. The tell is a manager who can describe the keyword strategy in detail and has not opened the search terms report at ad group level this month.

16. Broad match introduced to a campaign whose conversion signal has not been verified, which hands query selection to a system pointed at the wrong target.
17. Phrase and broad variants of the same keyword sitting in one ad group with no routing logic, so which one serves is decided by the auction rather than by intent.
18. Treating exact match as literal. It matches close variants, including reordered words and same-meaning phrasings, and the search terms report is the only place that becomes visible.
19. Keywords selected from a research tool and pushed live without ever reading the queries they attract in this specific account, in this specific market.
20. Bidding the informational head term for a considered purchase and sending it to a page written for someone ready to buy, then concluding the term does not convert.
21. Competitor terms added without checking whether any page on the site actually addresses the comparison a person searching that term is trying to make.
22. Brand terms bid without any assessment of whether organic already holds that click, and without a defensive reason such as a competitor bidding on it.
23. Keeping keywords with no impressions indefinitely, which clutters the account and makes it harder to see the ones that are genuinely doing something.
24. Long-tail keywords split so granularly that no single one accumulates enough data to be judged, and the account is left managing noise.
25. Pausing keywords on cost alone, without checking whether they appear earlier in converting paths under a model other than last click.
26. The same keyword in multiple ad groups pointing at different landing pages, so which page a searcher gets is decided by ad rank rather than by anything you intended.
27. High-intent modifiers and research modifiers in one ad group, so a single ad and a single page have to serve two people at opposite ends of the decision.
28. Keywords in a language or script the ads and landing pages do not serve, usually inherited from a market test nobody unwound.
29. Misspellings and plural variants added manually when close variant matching already covers them, splitting data across near-duplicates for no gain.
30. Reviewing search terms only at account level, where ad-group-specific drift averages out and stays invisible until it is large.

Group 3 — Negative keywords (31 to 42)

Negative keyword mistakes divide cleanly into two kinds: not blocking what you should, and blocking more than you meant to. The second is the one that goes undetected, because suppressed traffic leaves no trace in the interface — there is no report of the queries you stopped being eligible for. That asymmetry is why negatives deserve the same review discipline as the keyword list itself.

31. Adding negatives at campaign level when the waste is coming from one ad group, which suppresses queries you wanted elsewhere in the same campaign.
32. Negative broad match on a term that also appears inside converting queries, quietly removing eligibility for a whole class of good traffic.
33. Negative lists applied to the campaigns that existed when the list was built and never attached to the ones created since.
34. Adding negatives as exact match when the waste is a recurring theme rather than a single query, so the same pattern keeps arriving in slightly different words.
35. Blocking a generic word that also appears in the brand name or product name, cutting off branded search without anyone noticing the drop is self-inflicted.
36. No negatives for the standing off-target categories in the vertical — job seekers, free and DIY variants, academic and definition queries — on a lead generation account.
37. No cross-campaign negatives to keep brand queries out of generic campaigns, so the generic campaign harvests brand demand and reports it as its own performance.
38. Negatives that conflict with active keywords in the same campaign, silently making those keywords ineligible while they remain enabled and apparently healthy.
39. Building the negative list only from queries with cost and no conversions, never from queries that converted into leads the sales team disqualified.
40. Never revisiting the list after the business added or dropped a service, so a term that is now core is still blocked by a decision made under old circumstances.
41. Confusing placement exclusions with negative keywords on Display and video inventory, and expecting a keyword-level block to control where an ad appears.
42. Adding negatives with no record of why, so the next manager removes what looks arbitrary and the account pays to relearn the same lesson.

Group 4 — Ad copy and assets (43 to 54)

Copy mistakes are hard to see from inside a reporting view because the metrics they damage are downstream. Ads that describe the category rather than the offer still earn clicks; they just earn the wrong ones. The diagnostic is not a metric at all — it is reading the ad next to the query it served against and asking whether a person who typed that would find the answer here.

43. Headlines that restate the keyword three ways and never state the offer, the differentiator or the reason to click this result over the one above it.
44. Responsive search ads pinned so extensively that no meaningful combination testing can occur, while the report still presents combinations as if they were tested.
45. Nothing pinned in cases where a legal, pricing or eligibility qualifier must appear in every served combination, letting the system assemble a claim you did not make.
46. One generic ad set reused across every ad group, which breaks the query-to-ad match that the whole ad group structure exists to create.
47. Ad copy promising a specific thing — a price point, a free trial, next-day service — that the landing page does not confirm within the first screen.
48. Sitelinks pointing at pages that duplicate the main landing page, adding surface area without adding a single new answer for the searcher.
49. Callout assets repeating the same claims as the headlines, spending the extra real estate on redundancy instead of a second reason to click.
50. Structured snippets using a category header that does not fit the values listed under it, which reads as machine-assembled and undermines the rest of the ad.
51. Call assets running during hours when nobody answers the phone, paying for a click whose only intended outcome is impossible at that moment.
52. Location assets pulling from an unmanaged Business Profile with wrong hours or an old address, so the ad is accurate and the extension contradicts it.
53. Image assets uploaded once and never checked in the served crops, where the important part of the image is frequently outside the frame.
54. Ads left untouched while the offer, pricing page or seasonal context changed underneath them, so the account advertises a version of the business that no longer exists.

Group 5 — Conversion tracking and measurement (55 to 69)

This is the highest-value group in the list, and it is the one most often skipped because the account appears to be working. Tracking errors do not announce themselves — they produce a number, the number is used, and the error propagates into bidding, budgeting and every report anyone reads. The first thing I check in any account is whether the conversion actions mean what the names on them imply.

55. The conversion tag firing on load of a thank-you page that can be reached by direct navigation, by refresh or by a back button, so non-events are counted as outcomes.
56. Counting set to every rather than one for a lead action, where a single person refreshing the confirmation page multiplies into several recorded conversions.
57. The native Google Ads tag and an imported GA4 key event both marked primary for the same action, double counting every conversion the account produces.
58. Micro-conversions marked primary alongside real ones, so the bidding strategy optimizes toward whichever is cheapest to produce rather than whichever matters.
59. No conversion value set at all, which leaves every value-based strategy with nothing to work with and forces the account to treat all outcomes as equal.
60. A single flat value applied to every conversion in a business where deal sizes differ by an order of magnitude, encoding a claim about the business that is not true.
61. A conversion window shorter than the real sales cycle, so long-consideration conversions fall outside it and the campaigns that produce them look like failures.
62. Phone call conversions counted from the click on the number rather than from a connected call of a minimum duration, counting misdials as leads.
63. A tag bound to the form submit button rather than to a validated success response, so every failed and re-attempted submission is recorded as a conversion.
64. Consent handling not configured deliberately, so conversions are lost or modeled in ways nobody in the account can explain or reconcile.
65. Enhanced conversions assumed to be active because they were switched on once, without anyone opening the diagnostics to confirm data is being received and matched.
66. Offline conversion import built during onboarding and silently failing since — click identifiers not stored on the lead record, or uploaded after they expired.
67. Cross-domain measurement missing where the form lives on a booking or payment subdomain, so the session breaks and the conversion is attributed to a referral.
68. GA4 and Google Ads conversion counts placed side by side in a report as though they were the same measurement, when the definitions, windows and models all differ.
69. Never reconciling reported conversions against records in the system where money is actually recorded, for one complete sales cycle.

That last one is the whole game, and it is a full exercise rather than a checkbox. The sequence I use for it — reconciling events, checking container hygiene, mapping conversion actions to revenue events, then reconciling against the CRM — is written up in full in the conversion tracking audit, and it is the one piece of work that should happen before any budget decision rather than after one goes wrong.

Group 6 — Bidding and budget (70 to 81)

Bidding mistakes cluster around two behaviors: giving an automated strategy a target it cannot reach with the data it has, and intervening often enough that it never stabilizes. Both feel like management. Both produce an account permanently in a transitional state, where every reading is taken during a period the strategy itself considers unreliable.

70. Switching bidding strategy while the previous change is still recalibrating, restarting the learning period and guaranteeing no clean read of either.
71. Setting a target cost per acquisition based on what the business wants rather than on what the account has demonstrated it can produce, so delivery collapses to almost nothing.
72. Applying a target return-on-spend strategy to a campaign that has no conversion values, which asks the system to optimize a ratio whose numerator does not exist.
73. Maximize clicks used as a launch default and then left in place long after there is enough conversion data to bid toward outcomes instead.
74. A budget capped so far below what the strategy needs that the campaign never leaves the limited-by-budget state, and every other diagnosis is confounded by it.
75. Raising budgets in single large jumps rather than stepped increases, resetting learning at the exact moment the account is trying to scale.
76. Shared budgets spanning campaigns with different goals, where the least disciplined campaign drains the pool and the disciplined one is starved without any visible cause.
77. Portfolio strategies grouping campaigns whose economics genuinely differ, so the strategy optimizes an average that describes none of them.
78. Manual bid adjustments layered on top of a smart bidding strategy that ignores most of them, producing a settings screen that does not describe what the account is doing.
79. Reacting to daily fluctuation as though it were signal, when day-to-day movement in most accounts is auction noise and weekday effect.
80. Never checking pacing mid-period, so the account underspends for weeks and then absorbs the remainder in a rushed final stretch at the worst prices.
81. Applying seasonality adjustments to ordinary weekly variation rather than to known, short, genuinely anomalous events, which teaches the strategy to expect the wrong pattern.

The underlying discipline here is not a bidding technique, it is a written rule about when a change may be made at all. I have covered the version I use — data thresholds before changes, holdouts for strategy tests, a logged reason for every edit — in the bid governance framework, and it exists mainly to stop diligent managers from destroying their own signal.

Group 7 — Geographic and schedule targeting (82 to 91)

Targeting mistakes are the cheapest to fix and among the most common, because the settings are configured once at launch and then live outside anyone attention. They produce a distinctive symptom: spend that is unimprovable by any amount of copy or bidding work, because the click could never have converted regardless of what happened after it.

82. Location targeting left on presence or interest when the business can only serve people physically in the area, paying for searchers who are merely curious about it.
83. Country-level targeting on a business whose actual service area is one city, with the difference absorbed as generalized underperformance.
84. Radius targeting drawn around a pin without excluding the adjacent area the business does not serve, because a circle does not respect a service boundary.
85. No excluded locations ever added, even after the location report has shown a region producing clicks and unusable leads for months.
86. Location bid adjustments set at launch from assumption and never revisited against what the location report has since shown.
87. An ad schedule mirroring office hours on an account whose primary conversion is a form that works perfectly well overnight.
88. The ad schedule configured in a time zone that is not the audience time zone, shifting every intended window by a fixed offset nobody notices.
89. Hour-of-day bid adjustments built on data too thin to support them, encoding a pattern that is mostly small-sample variance.
90. Campaigns running through periods when the business cannot fulfill — holidays, closures, stock-outs — paying for demand it will disappoint.
91. Language targeting set to the language of the ads while the audience browses with different language settings, quietly restricting eligibility.

Group 8 — Audiences and remarketing (92 to 99)

Audience mistakes tend to be errors of staleness rather than errors of design. A list that was correct when it was built keeps running against a business that has changed, and because it continues to deliver, nothing prompts anyone to check whether it should. The diagnostic question for every list in an account is simply: when was this last rebuilt, and against what definition.

92. Remarketing lists that still include people who already converted, paying to re-persuade customers who have nothing left to buy in that flow.
93. Membership duration set far longer than the real consideration window, keeping people in a list long after the intent that put them there expired.
94. Observation and targeting confused, with a list set to targeting and quietly restricting a campaign to a fraction of its intended reach.
95. Customer match lists uploaded once at setup and never refreshed, so the audience gradually becomes a snapshot of an old customer base.
96. Similar-audience expansion seeded from a list built on a conversion action that turned out to be the wrong one, scaling a definition error.
97. Remarketing creative identical to prospecting creative, spending a returning-visitor impression on a message that person has already seen and not acted on.
98. No frequency control on display or video remarketing, where the same person absorbs impressions well past the point of diminishing effect.
99. Audience lists too small to serve, sitting in the account looking configured while delivering nothing at all.

Group 9 — Landing pages and post-click experience (100 to 111)

The account stops at the click, which is exactly why post-click mistakes survive longest. Nothing in the Google Ads interface reports that a page contradicts its ad, loads slowly on the devices most clicks come from, or asks for information a person is not ready to give. The only diagnostic is to open the page as a searcher would, on a phone, having just read the ad.

100. Every ad pointing to the homepage, which forces the searcher to re-navigate to what the ad already promised and loses a share of them at that step.
101. A landing page headline that does not restate the query in recognizable language, leaving the visitor to verify for themselves that they are in the right place.
102. A form asking for fields the business does not need at this stage, trading conversion volume for data nobody uses before the first conversation.
103. A page that is acceptable on desktop and slow on mobile, where the majority of clicks land and the tolerance for waiting is lowest.
104. The offer positioned below the fold, so the visitor has to scroll before seeing any confirmation that the page answers the ad.
105. Multiple competing calls to action on the page, which converts a decision into a choice and reliably reduces the number of people who make either.
106. No proof of any kind for a first-time visitor arriving from an ad — no evidence, no specifics, nothing that distinguishes the claim from every competing claim.
107. One thank-you page shared across every form on the site, so the tracking cannot attribute a conversion to the form that produced it.
108. Landing pages changed by another team without notifying whoever runs the account, so performance shifts and the cause sits outside the advertising platform entirely.
109. The phone number on the landing page not being a tracked number, which means every call from paid traffic is invisible to the account.
110. A chat widget capturing conversations that never reach the CRM, creating a class of leads that exists in the business and not in any report.
111. Redirect chains on the destination URL, where parameters are dropped along the way and the resulting sessions arrive without their tracking data.

Page speed and page structure are not marketing concerns that stop at the ad platform boundary. The same architectural discipline that makes a site work for search is what makes it hold up under paid traffic, and a landing page rebuilt in isolation from the rest of the site tends to inherit none of it.

Group 10 — Attribution and reporting (112 to 121)

Reporting mistakes are the ones that cause the wrong decision rather than the wasted click, which makes them more expensive per instance than anything above. They share a pattern: two numbers with different definitions compared as though they were the same number, or one number reported without the second number that would catch it lying.

112. Comparing platform-reported conversions with analytics sessions and treating the difference as an error, when the two are measuring different things by design.
113. Judging upper-funnel campaigns on a last-click view, which structurally assigns their contribution to whatever campaign closed the path.
114. Changing the attribution model partway through a reporting period and then comparing performance across the change as though the basis were constant.
115. Reporting on a window shorter than the typical conversion lag, so recent spend always looks worse than older spend and every fresh test appears to fail.
116. Treating impression share as a target to maximize rather than a diagnostic to interpret, which rewards buying inventory the account had good reason not to buy.
117. Using click-through rate as the headline account health metric, when it rises reliably whenever the account narrows toward its easiest, most branded traffic.
118. Reporting conversion count with no revenue alongside it, which is the single most common way an account looks like it is improving while it degrades.
119. Comparing periods without accounting for the number of days, weekday composition or seasonal context in each.
120. Auto-applied recommendations changing the account with no corresponding entry in anyone reporting narrative, so results move for reasons the report cannot explain.
121. Building the entire report from platform data and never reconciling it to the system where revenue is actually recorded.

The metric that lies

Conversion volume is the most dangerous number in a Google Ads account, because it is the one everybody looks at and the one that can rise for reasons that have nothing to do with the business improving. It rises when a cheap micro-conversion is marked primary. It rises when a tag starts double counting. It rises when an automated strategy discovers a segment that produces many low-quality outcomes at low cost. In all three cases the chart goes up and to the right and the bank balance does not follow.

The second number that catches it is revenue, or in a lead generation business, qualified pipeline recorded in the CRM for the same period. Put them on the same chart. If conversions are climbing and the second line is flat, you do not have a growing account — you have a conversion action that has drifted away from meaning anything, and the drift is being rewarded by the bidding strategy every day it goes unnoticed.

Impression share is the same failure in a different costume. It is a useful diagnostic when a campaign is losing share to budget and you want to know how much demand is going unserved. It is a terrible target, because the fastest way to increase it is to buy the inventory the account was correctly declining, and the metric will improve at exactly the moment the economics get worse.

Group 11 — Process, governance and account hygiene (122 to 127)

The last group is not about the account at all. These are the conditions under which every other mistake becomes permanent — an account with no change log cannot diagnose a regression, and an account with no owner accumulates drift faster than any individual error costs. There are only six, and they are the ones that determine whether the previous 121 get found.

122. No change log, so the question "what changed before this got worse" has no answer and every regression is investigated from scratch.
123. Auto-apply recommendations left on, which means the account is being edited continuously by a party that does not attend the review meetings.
124. Account access shared through a personal login rather than individual users, making the change history unattributable and offboarding a security problem.
125. No written rule for when a change may be made, so changes are made whenever someone feels uneasy, which is the definition of reacting to noise.
126. No named owner between agencies or after a handover, during which settings drift and nobody is accountable for the drift because nobody is accountable for anything.
127. No scheduled cadence for reviewing search terms, negatives, assets and tracking, leaving all of it to whenever someone happens to look.

The eight that account for most of the waste

If you do nothing else with this list, do these. They are chosen not because they are the most sophisticated but because each one, when wrong, invalidates a large amount of other work — and because in accounts I am handed, these are the ones that are most often wrong.

• A conversion action that does not mean what its name implies — wrong trigger, wrong counting rule, or a micro-conversion marked primary. Everything downstream inherits this error.
• Conversions counted without values, so the account cannot tell a small outcome from a large one and optimizes toward whichever is cheapest.
• The search terms report unread at ad group level, which is where the largest single block of immediately recoverable spend usually sits.
• Brand and non-brand traffic combined in one campaign, hiding whether anything other than brand demand is working.
• Location targeting set to presence or interest on a business that can only serve local customers.
• An automated bidding strategy given a target the account has never demonstrated, keeping delivery suppressed while looking like a bidding problem.
• Ad copy promising something the landing page does not confirm in the first screen, which converts a good click into a wasted one after you have paid for it.
• No change log and auto-apply recommendations enabled together, which makes every future diagnosis guesswork.

Seven of those eight are measurement or matching problems, not bidding problems. That is the honest shape of the thing: the levers most managers spend their time on are the ones least likely to be the cause.

What this list is not

It is not a work order. An account with sixty of these present does not need sixty fixes — it needs the tracking corrected, then a fresh read, at which point a meaningful number of the remaining items will turn out to have been misdiagnoses caused by the bad data. Working a long checklist top to bottom is a way to be busy in an account without changing its economics.

It is also not a substitute for knowing the business. Several items here are correct in one context and wrong in another. Bidding on brand terms is waste for a business nobody is bidding against and cheap insurance for one under attack. A long conversion window is right for a considered purchase and misleading for an impulse one. The account cannot tell you which situation you are in; only the business can.

And it does not cover the case where the account is configured correctly and the offer is the problem. If tracking is verified, queries match, pages confirm the ad, and the outcomes still are not there, the remaining explanation is usually that the market is not persuaded — and no amount of account work fixes that. Recognizing that boundary early is worth more than another pass through the settings.

Where to start this week

Verify one conversion action end to end, from a real submission through to the record in the system where revenue is tracked. Then read the search terms report for your highest-spending ad group, sorted by cost, and act on what is there. Then open your highest-traffic landing page on a phone immediately after reading the ad that sends to it. Those three take an afternoon and will tell you which of the eleven groups above is actually your problem.

If Quality Score keeps coming up in these conversations, it is usually downstream of the structural and page-level items rather than a cause of its own — the four myths worth clearing up first covers what actually moves it. I write these diagnostics out of running accounts with real client money on them at Arcetis, and the full method these groups sit inside is documented on the Signal-to-Revenue Framework page.

Related reading