An AI tool can build a complete Google Ads account in minutes. The platform’s own guided setup will do it from your website URL, and a growing list of third-party builders will do it from a form. The output is usually a defensible first draft: campaigns themed sensibly, keywords pulled from language already on your site, ad copy that reads like a person wrote it. If you built your account this way, you did not make a mistake.
What went missing is the reviewer. A human media buyer produces a draft too. The difference is that the draft then gets read by someone who knows which parts of Google Ads punish you quietly: an insertion tag with no fallback, a conversion action pointed at the wrong event, broad match with nobody watching the search terms. The AI produced a build. Nobody performed the second pass. Weeks later the spend is running against defects that never announce themselves, because none of them are errors. Every one is a legal configuration the platform will happily serve.
The defects also repeat. Builders share templates and habits, so the same shapes turn up account after account. In a 23-account fleet QA sweep dated 2026-07 we found live defects of this kind in 9 accounts. Ours was one of the 9: our own account was carrying an ad reading “Better Tacking & Conversions” that was still serving-eligible when we pulled it on 2026-07-17. That sweep is where the classes below got their names.
Eight of them follow, each with how to spot it and what it costs you. Then a pass you can run yourself, and a rule for deciding whether the account is worth repairing.
Dynamic keyword insertion with nothing behind it
Keyword insertion writes the matched keyword into the headline. The syntax is {KeyWord:Default}, and the part after the colon is what serves when the keyword will not fit or would break the ad. Builders like insertion because it produces relevant headlines cheaply. They are careless with the default, and they reuse the same tag across several headline assets.
How to spot it.Open a high-spend ad group, edit the responsive search ad, and read the headline fields as raw text rather than in the preview. Two things: an insertion tag whose default is empty or is unedited filler such as “Services” or your bare category, and the same tag appearing in three or four headlines at once. Then step through the preview combinations and watch for the same phrase showing up twice in one impression.
Why it costs you. When substitution fails, the fallback is your ad. If the fallback is filler, you pay top-of-page prices for a headline that says nothing specific, at the moment the searcher is deciding. The duplication problem is quieter: repeated insertion collapses asset variety, so the system has fewer distinct combinations to test and some of what it assembles reads as a stutter. Ad strength drops, and the ad stops learning.
A negative keyword list that was written once
Almost every AI-built account we open has negatives. That is not the problem. The problem is that they were written once, at build time, from a generic list: free, jobs, cheap, salary, DIY. Nothing has been added since.
How to spot it. Open a campaign, then negative keywords, and read the list. Generic and short is the tell. Then open the shared library: if there are no negative lists, every campaign is carrying its own private copy of the same twelve words. Last, search the non-brand campaigns for your own company name as a negative. In most inherited accounts it is absent.
Why it costs you. Negatives are not a setup task. They are a standing structure fed weekly by the search terms report, which is where waste actually shows up. A one-time list handles the waste someone imagined in advance, never the waste you are buying now. Without shared sets, a fix in one campaign does not protect the others, so the account re-buys the same junk through a different door. With no brand guard, non-brand campaigns compete for people typing your name, which corrupts the numbers on both sides at once.
One ad group per keyword, several hundred times
This one is easy to admire at first. Every keyword gets its own ad group and its own ads, so relevance looks perfect and the account looks meticulous. It is the shape a generator produces when it optimizes for tidiness rather than for data.
How to spot it. Go to the ad groups view for the whole account, read the row count, then add the keyword count column. Hundreds of ad groups holding one keyword each is the signature. Sort by impressions over the last 30 days: most rows will show single digits or nothing, and ad strength will be thin or unrated across the board.
Why it costs you. Every learning system in the account needs volume in one place. Smart bidding needs conversions per campaign; responsive search ads need impressions per ad group to decide which assets work. Splitting one modest budget across four hundred containers starves all of them, and nothing accumulates enough evidence to be judged. The single-keyword ad group was a workaround for a matching behavior Google no longer has, and it now costs more than it returns. Consolidate until each ad group can carry its own honest ad test, and no further.
Broad match on everything, with nobody reading the queries
Broad match is a demand discovery instrument, and it is fine provided somebody reads what it discovered and prunes. Builders default to it because it fills volume fast. The pruning half never gets assigned.
How to spot it. Add the match type column to the keyword table. If every row is broad, note it. Then open the search terms report for the last 30 days across all campaigns, sorted by cost descending, and read the top 30 terms out loud. You are not judging whether they relate to your industry. You are asking whether a person typing that could ever buy what you sell. If insertion is also in use, those ugly queries are the ones that got printed into your headlines.
Why it costs you. Broad match without a feedback loop drifts toward whatever is cheap and adjacent, and cheap and adjacent is rarely a buyer. Paired with insertion, it does something worse than waste money: it shows the searcher their own junk query back inside your ad, next to your company name.
Brand and non-brand living in one campaign
People searching your company name already decided. People searching what you sell have not. Those are two purchases at two prices, and an account that mixes them cannot report on either.
How to spot it. Open the search terms report filtered to a single campaign. If your company name and generic category terms both appear, they are sharing a budget. The other tell is a campaign with a click-through rate and cost per conversion that look suspiciously good next to everything else in the account. That is usually brand traffic carrying the average.
Why it costs you. Two mechanisms, and they compound. The first is measurement: the blended cost per conversion averages cheap demand you already earned with expensive demand you are trying to buy, so the number reads acceptable while the non-brand half runs far above what the business can pay. The loss is real and invisible. The second is allocation: an automated bid strategy chases the cheapest conversions available, and those are brand, so a campaign you funded to find new customers spends its day re-buying people who already knew your name. Split them into separate campaigns, not separate ad groups, because budget is a campaign-level control.
The conversion action counts the wrong thing
This is the defect that makes the other seven harder to see. It is the one we care about most.
How to spot it.Open conversion settings and list every action with its category, its source, and whether it is primary. Two shapes turn up constantly. One is a page view counted as a conversion, sometimes a thank-you page but often any page or a session-length event, created automatically and never questioned. The other is every form submission counted as a lead, undifferentiated. Then do the arithmetic that settles it: compare last month’s conversion count against the number of real inquiries you or your team can name. If the account claims 60 and you can name 9, the signal is fiction.
Why it costs you. The bidder is obedient. Tell it to buy form submissions and it will find the people most likely to submit forms: spam traffic, job seekers, vendors, price shoppers. It is succeeding at the target it was given, and it gets better at that target every week. That is why a broken conversion action is worse than none at all. The account spends months learning an audience you do not want, and every later optimization builds on it.
Pinning errors in responsive search ads
A responsive search ad takes up to 15 headlines and 4 descriptions and assembles combinations from them. Pinning locks an asset to a position. Both over-pinning and under-pinning show up in generated accounts, for the same reason: the tool had no view about which messages are required.
How to spot it. Open the ad editor and look at the pin icons on each asset. Over-pinned looks like every position filled and locked, leaving one or two possible combinations. Under-pinned is harder and matters more: read the headline list and ask whether any message must appear in every impression. A license number, a service area, a qualifier such as commercial only or by appointment. If it is unpinned, it is optional, and it will be missing from most of what serves.
Why it costs you. Over-pinning turns off the only ad testing mechanism the platform still gives you. Under-pinning costs you in click quality: the qualifier that would have stopped the wrong person from clicking is missing from the combination that serves most often, so you pay for that click and for the time spent disqualifying them.
Auto-applied recommendations left switched on
The build was one event. Auto-apply is continuous. Many AI-built accounts are handed over with a broad set of automatic recommendations enabled, so the account keeps editing itself after everyone stopped looking.
How to spot it. Open the recommendations page, find the auto-apply settings, and read what is enabled. Then open change history for the last 90 days and look for changes attributed to automation rather than to a person. Added keywords, broadened match types, raised budgets, and new ad variations are the common ones.
Why it costs you. An account that changes itself cannot be diagnosed. You read it on Tuesday, act on Friday, and the thing you fixed has already been re-added. Automatic keyword expansion undoes negative work directly, and auto-created assets reintroduce the insertion and pinning problems above. Turn every auto-apply setting off before you begin. Not because the suggestions are always wrong, but because a moving account cannot be measured, and you can re-enable a specific one later on purpose.
How to inspect the account in one sitting
This takes 30 to 60 minutes and needs nothing but read access. Do it in this order, because each step tells you what to look for in the next. Keep one document open and write down the dollar figure attached to every finding. The dollar figures are what make the list actionable.
- Search terms, last 30 days, all campaigns, sorted by cost. Write down the terms that could never buy anything and the total spent on them. That number sets the stakes for everything else.
- Change history, last 90 days. Filter to automated changes. Write down whether the account is still editing itself and what it has been changing.
- Conversion settings. List every action, its category, and whether it is primary. Write the last-30-day count next to the number of real inquiries you can name. That gap is the most important line in the document.
- Ad previews for the two or three highest-spend ad groups. Read the headlines as raw text. Note bare insertion defaults, repeated insertion, and pins.
- Shared negative lists. Note how many exist, what is in them, and which campaigns they are actually applied to. Note whether brand terms are excluded from non-brand campaigns.
- Campaign settings. Networks first: search partners and display expansion are frequently left on. Then locations, where the default reaches people merely interested in your area. Then the bid strategy and its target, which is what the account has been told to want.
At the end you have a page or two of findings with money attached to them. That page is worth having whether you fix the account yourself, hand it to somebody, or do nothing for a month.
Rescue or rebuild
The instinct after reading a list like this is to burn the account down and start over. Sometimes that is right. Often it throws away the one asset that took real time to accumulate: conversion history the bidding system has learned from. Here is how we decide.
- Repair it when the structure is salvageable
- The campaign count is small enough to hold in your head, brand can be separated cleanly, the conversion action measures something real or can be repointed without invalidating what came before, and there is spend history worth keeping. Repair is faster, cheaper, and does not reset learning.
- Rebuild when the foundation is the defect
- Hundreds of thin ad groups nobody can maintain. No conversion history you would trust. Or the case that settles it on its own: the conversion action has been counting the wrong event since day one. Then the recorded history is not merely thin, it is wrong, and every automated decision on top of it was trained toward the wrong buyer. Keeping that history keeps the wrong training with it.
- Either way, the procedure is the same
- We run repair and rebuild under one 14-day fixed-price engagement at $2,500. Which one you get is a finding of the audit, not a quote given in advance, and the defect list with its evidence is yours to keep whichever path you take and whoever does the work.
One caution on rebuilds. A rebuild that carries the old conversion definition across is not a rebuild. Fix what the account is optimizing toward first, then decide what to do with the campaigns. That order changes the answer more often than anything else on this page.