by Austin LeClear
on August 16, 2026
TLDR: A client took his Google Ads account back off us, asked ChatGPT and Claude what to change, then applied Google’s recommendations on top. He switched to Maximize Conversions with no CPA ceiling, added broad match keywords in batches, and deleted the negative keyword list we had built. Over the next two months conversions fell 56.76% and cost per conversion climbed 52.14%, from £96.84 to £147.33. He is paying roughly double for less than half. We still run his business partner’s other account, untouched, and it is performing fine.
We had a client take his account back and run it himself.
Two business partners, two Google Ads accounts, serving two different countries. We managed both. Then one partner decided to answer a question a lot of business owners are quietly asking right now: do we even need an agency if AI can just tell us what to do?
So he took over his account, asked ChatGPT or Claude what to change, applied Google’s recommendations on top of it, and let it run.
He got his answer. It cost him roughly half his conversions.
This is not a hypothetical, and it’s not a made-up story to sell agency services. This is an account we used to manage, with a change history you can read.
Prefer video? Check out A Client Let AI Run His Ads – It Got Destroyed on YouTube:
Still with us? Good, let’s dig in.
How We Found Out He Had Taken Over the Account
The first sign wasn’t a dashboard. It was a Slack message from the strategist managing the account.
That’s the whole story in one sentence. Bad recommendations from a general-purpose chatbot, stacked on top of every recommendation Google was pushing in the account.
I don’t know exactly what he asked or which model he used. I do know what landed in the account, because Google Ads keeps a receipt for every change.
The Before: What the Account Looked Like Under Management
Here’s the account from January 1 through April 19, 2026, the window covering our management plus the first few days of the takeover.
Under our structure, this account ran 50 to 70 conversions per week at roughly a £100 cost per acquisition. It was not a big growth story or a rescue project, just a well-tuned account doing predictable work every week, the kind of steady result that pays a business’s bills.
Look at those two lines. Flat and calm for three and a half months, then at the right edge they cross: CPA climbs steeply while conversions drop just as fast. That is where it breaks.
April is when he took over.
The After: 13 Conversions at a £351.69 CPA
Hover that last week on the graph and Google hands you the number.
Weekly conversions dropped from 50 to 70 down to 13, and the cost per conversion went from about £100 to £351.69.
The following week was worse, at 3 conversions around £700 each. By May 3 it recovered slightly to about 15 conversions at £244, which tells you the account had stopped getting worse, not that the problem was fixed.
Notice the detail most people skip. Between April 19 and May 3, the account spent another £5,765.95 and added 18 conversions. That works out to about £320 for each extra conversion, against a £96 baseline. Every additional pound going through the account was buying worse results than the pound before it.
The Damage: Conversions Down 56%, Cost Per Conversion Up 52%
Here is the comparison that matters. April 1 through May 31, the AI-run period, against the previous period, which was our setup and our management.
Read those two numbers together, because separately they undersell it:
- Conversions: down 315, or 56.76%
- Cost per conversion: £96.84 to £147.33, up 52.14%
He is now paying roughly double to get less than half.
That’s not a dip or a learning period. This account went from buying a customer for £97 to buying one for £147, while the number of customers it delivered got cut in half. If this were a lead gen business with a sales team, half the sales pipeline disappeared while the ad invoice went up.
The math doesn’t work. It can’t work.
What AI and Google Actually Told Him to Do
The change history separates changes Google recommended from changes the client made himself, and both columns are ugly.
First, look at the applied recommendations. These are Google’s suggestions, accepted with a click.
Then his own manual changes, marked “Web client (manual)” with no recommendation attached.
Strip out the noise and there are four real changes:
- He switched the bid strategy to Maximize Conversions, with no target CPA ceiling.
- He added an enormous volume of broad match keywords, in batches of 15, 10, 10, 10, 10, 9, 8 and 7. Some came from Google’s recommendations, most were his own.
- He removed essentially all of the negative keywords we had built. Google’s “redundant keyword” recommendations stripped out phrase match terms too.
- He made a pile of ad copy changes.
Number four is fine, honestly. It was mostly ad copy, and I don’t think it produced any meaningful change in either direction.
One through three is how you destroy an account.
Why Those Three Changes Broke the Account
Individually these changes are survivable. Together they’re fatal.
Broad match tells Google it can show your ad for anything it decides is related to your keyword. Maximize Conversions tells Google to spend the entire budget chasing volume with no cost ceiling. Negative keywords are the limits that stop Google spending on searches that will never buy.
He expanded the targeting as far as it goes, removed every limit, then told the system to spend freely.
That’s not three mistakes. That’s one mistake made three times in the same direction.
Broad match without negatives means Google starts buying searches that are close to your business but not actually your business: the browsers, the researchers, the job seekers, the people looking for something free. Maximize Conversions with no target means it keeps buying them as long as budget exists. And that “Forecasting Set CPA Target” recommendation increased the budget while all of this was happening.
To be fair, a full broad strategy can work. I’ve consulted with companies spending millions per month who make very good money from broad reach. The difference is conversion volume, meaning enough weekly data for smart bidding to find the pattern instead of guessing at it. Even then, those advertisers keep a CPA target in place rather than letting it spend with no limit.
This client spends fine. He does not spend like that. Full broad works for maybe 1% of advertisers, and nothing in a chat window told him he wasn’t one of them.
When to Actually Use AI on a Google Ads Account
This is not an anti-AI post. I have Claude running right now, I use Claude Code every day, and AI is genuinely part of how we work.
The difference is that I know which question to ask and when, and my Claude references our own brain rather than the open internet: thousands of hours of Loom recordings, internal meetings, and trainings built up over five years. He asked an ads question without knowing what a good ads answer looks like, using a general subscription that pulls from the open internet and Reddit.
Most of that source material contradicts itself, so the answer can’t be right. It can only be confident. The answer is reassuring, well written, certain, and sometimes wrong. If you’re not an expert, you cannot tell a confident wrong answer from a correct one.
Let ChatGPT or Claude near your account if:
- You already know what a correct answer looks like, so you can catch a wrong one
- You’ve trained it on your own methodology, recordings, and account history
- You’re using it to pressure-test a strategy, not to decide which buttons to click
- Someone on your side can audit every recommendation against your own data
Keep it away from your account if:
- You can’t tell a good ads answer from a merely confident one
- It’s a general subscription pulling from the open internet and Reddit
- You’re about to change bid strategy, match types, or negatives on its advice
- You have no baseline to measure the damage against if it’s wrong
Treat Google’s recommendations the same way. Google’s incentive is more spend, yours is more profit. Every recommendation in that tab is a suggestion, not an instruction.
FAQs: Letting AI Run Google Ads
Can AI run a Google Ads account on its own?
Not today, and this case study shows why. ChatGPT and Claude can’t see your conversion volume, your margins, or your lead quality, so it can’t tell you whether a tactic fits your situation. It will still answer with total confidence.
What actually caused the 56% conversion drop?
Three changes made together: switching to Maximize Conversions with no CPA ceiling, adding a large volume of broad match keywords, and removing the account’s negative keywords. You can recover from any one of those. All three at once removes every limit on where the budget goes.
Is it safe to apply Google Ads recommendations?
Some are useful, but many push toward broader reach and higher spend. In this account, applied recommendations added broad match keywords, stripped out phrase match terms, and increased a budget. Review each one against your own strategy instead of accepting them in bulk.
When does a broad match strategy actually work?
When the account has enough weekly conversion volume for smart bidding to find the pattern rather than guess at it. That usually means high spend and high conversion counts, and those advertisers still tend to keep a CPA target in place rather than running Maximize Conversions with no limit.
What can ChatGPT not tell you about your Google Ads account?
Anything that depends on your actual data. It can’t see your conversion volume, your close rate, your margins, your lead quality, or your account history, so it can’t judge whether you have enough data to support a tactic. That is the gap that turned generic advice into a 56% conversion drop here.
Final Thoughts
The account is still running this way, and there are no recent changes in the history. That suggests it is simply being left to run now, with some hope that it fixes itself.
It won’t. Half the conversions at double the CPA is not a phase. We still run his business partner’s account in the other country, same business and same offer, and it is doing fine. That is about as clean a control group as this industry ever hands you.
The tools aren’t the problem here. AI is a serious multiplier if you already know what you’re doing and you’ve given it the right data to work from. Point it at the general internet, hand it an account it cannot see, and you get a very convincing explanation of why you are losing money.
Keep someone who knows the account between the recommendation and the button. That is the part you cannot automate.
Get a personalized, 1-on-1 Google Ads account review.
Not sure whether your bid strategy, match types, or the recommendations sitting in your account are quietly costing you conversions?
If you want our team to take a look, we’ll do a complete audit of your account at no cost. You’ll know exactly what to change to improve your results in the next 90 days. It’s a custom, in-depth analysis that reveals day-one improvements to your campaigns.
We don’t take every account that applies. We only work with businesses where we’re confident we can deliver real results, so we’ll need a few details on your business and goals before we book the call.
(If you qualify, it’s completely free.)
If that sounds like a fit, click below to apply.
