AI Google Ads Audits: What They Get Right, What They Get Wrong & Why You Still Need a Human

AI just audited a Google Ads account with two years of data, and it got every observation right and almost every conclusion wrong. 

In this episode of the Google Ads Unleashed Podcast, host Jeremy Young breaks down a real AI-generated Google Ads audit a client submitted, walking through exactly what the tool caught, what it missed, and why correlating data points without business context leads to dangerously confident but fundamentally flawed conclusions. 

If you manage Google Ads for yourself or a client, this episode will sharpen how you use AI tools without letting them do your thinking for you.

Get your free 30 minute strategy session with Jeremy here: https://www.younganddigital.marketing/

Scale your store with 1:1 coaching: https://www.younganddigital.marketing/1-2-1-coaching

Transcript

This week something happened which I didn't think I'm going to see for a little while but it is finally here. A client of mine sent me a forensic audit of their own ad account. 2 years of data change history root cause analysis and everything. AI has supposedly taken our jobs. And for a while I was scared because when smart shop shopping came out, everyone was scared. When automated bid strategies were rolled out, everyone was scared. But this week has solidified me much more that for at least the next few years, you will need someone who actually knows marketing, who can think and who has absolute pathological knowledge of Google Ads as a platform. What has happened exactly and what you should be doing when you want to do your own AI? audits or get faced with such audits is shown. I'll talk about this in this week's episode. Welcome back to Google Ads Unleash guys. Hope everyone is doing fabulously this Monday. Client has sent me a full audit AI generated of their ad account this week and wanted me to talk about it. And this is going to happen much more because everyone has now a tool at their disposal where it's claw chat GVT if you do a bit of hooking up etc etc that can read your ad account write reports do maybe even changes suggestions etc etc and a lot of Google ads experts are worried about this I see it as an opportunity at least for the next few years probably AI is going to get better and better and better and we'll do more of the doing further down the line but for now it sort of solidifies one kind of takeaway for me and that is you still have to do the thinking. Today I'm going to tell you about exactly what happened, what the audit got right, what the audit got wrong, and what you should be doing when you're faced with this or if you're generating one, how you should interpret it. So to be fair, the audit that I actually got was really carefully done. Uh it flagged its own confidence levels. It said what the data couldn't access. Even admitted that some of its uh um some of the APIs it was connected with was either with the wrong account or it had several accounts in there. Uh but here is where things go wrong. Like for instance, this client is super super clever. I love him a lot. He actually very good friends of ours as well. However, the issue is is that he starts to have overconfidence in the AI and potentially well I think he's very aware of that but um it is some times than extremely hard to argue against that when people are so deep in the narrative because that is what the AI does really really good. Uh it paints a fantastic sort of um narrative based on the data that it can see because what the AI is really really good is it caught a couple of things. So for instance it caught a conversion tracking inversion. Google over reported some lead data for bump between 5 to 15% for over about 3 years, which in the grand scheme of things wasn't actually that bad. It was just one conversion action that was duplicated. But since we've actually fixed the conversion tracking or had an expert looking over this a few months ago, this has actually now flipped to minus 19% under reporting, which of course means that we are now not feeding the algorithm enough data. So this is a great prompt. So to us, this is something that uh a human reviewing monthly numbers could never really catch. To be honest, it's really really hard. You only see this if you plot the ratio across 30 plus months or if you have any sort of suspicion that anything was even done wrongly. We hired a tracking guy, paid him for this and we would have thought that it would have fixed the data. That being said, we used to also track different conversion actions and also had other campaigns live. For instance, a brand campaign which we don't even have live anymore. Meaning that we used to have a lot more conversions that we track because they were branded uh conversions for instance. Now we don't. And this is where it gets right. It sees this and then it prompts you to ask the right questions but it doesn't have the context that I have and understands that a campaign is instance branded versus non-branded. And this could also be the cause for the under uh reporting of the data. It's also uh super super great at um talking like uh getting sto like stories and changes uh uh right. So it's more or less like a research assistant that builds a timeline um and sort of tells you a sort of consecutive um s sort of story of different events that we happened and undertook in the ad account. So overall though It has one big flaw though that it can only access the data and do something with the data that it doesn't actually uh well and it can only do a story with the data that it sees because what it got massively wrong is that it used all of those account artifacts as actual proof of the market. So for instance, one thing that it said is that because impressions were up 47% year on year It said that uh with high confidence by the way whilst also admitting that it had actually no search volume data about the market whether there's been a demand softening or a demand strengthening that as impressions were up 47% yearonear. It could not be a competition issue because if they were competitors in the market impressions would be down. Well, what actually happened is that a competitor entered the market in November causing significant pressure on first of all the uh SQL ratio that we've been tracking. So in fact since then already uh lead quality had dropped slightly which was back in November and in December also but also what had happened is with a maybe slight under reporting which I've just mentioned and this competitor pressure which if you would check the auction insights which it didn't you would see then you would probably realize that Pmax has put a lot of um which by the way I've done an episode on this. This is a very rare event where we're actually using Pmax in a lead gen account. Listen to it. There's a very good reason to it uh for this um where Pax is then pushing the budget into lower value placements because it's not being able to compete at the target CPA with the competitor that I've mentioned which causes impressions to rise. So in fact it made the wrong conclusion from just the data that it saw because it has no idea about auction inside and it doesn't have an idea about any of the outside data of the of the business and the real things that had happened. Um it also sort of uh miscorrelated a couple of things. So for instance we had a couple of campaigns that we launched on a new domain within the within the same ad account. They related domains is a very long story but We launched those on max clicks. So it correlated that with poor quality data or broad traffic that would go to bad leads. Although the one thing didn't really have anything to do with the other. And I think what all of this does, it just uh paints a story of correlation which actually isn't causation and it says that as matter of fact. So for instance, certain things that happened in the same month are just cor causated to have caused an outcome which doesn't actually really happen. Um or for instance some of the things that it said um that might cause a drop in lead quality or something like that. It acknowledged that this happened earlier. Nevertheless, it still said that you know increase in budget on PMAX for instance caused this. So what has sort of stood out to me when when when this happened is that AI is really really good for the timeline and the questions that you should be asking, but it's really really poor for the conclusions that you can draw from it. So when you get presented with something like that or if you're running it yourself, you should really ask yourself on every causal claim that it makes, is there something that the AI can't see? Is it, for instance, some data that it doesn't have? Maybe you can even still feed it the data afterwards and refine your audit, right? If you're worrying that, hey, this doesn't really make sense or maybe it doesn't see seasonality, which you could feed the data. You could maybe you use for instance Google Trends or something else or download Google Trends data and feed it that it could it may not see the offer. Uh for instance, the competitor had a quite compelling offer um and a different way of displaying um uh basically information to a lead that may or may not actually entice that lead to convert. It didn't really see that right. It also didn't has no idea about the sales team. Is the sales team good? Has there been an issue within the team? Has there been an issue within the market etc etc. So whilst it prompts you to ask good questions and the right questions and whilst you want to valid validate those and actually acknowledge them. You really have to be very wary of some of the oversimplifications that it makes because AI does the uh sort of grunt work. It repaints the story of all the changes and and everything and builds that into a narrative but you as the expert have to be the layer on top of the tool that actually makes sense of the data because that is where AI still is going wrong in my opinion and I think you can only fix that by adding more context or by being the piece of context that it needs in order to make sense from it. So what do I make of this uh audit? So like I said, it painted a really good picture and retraced a lot of things, but it got every correlation and causation completely wrong because it didn't actually know what was going on. Um it got every observation right, but almost every cause wrong. And This is kind of the essence of this whole podcast and this uh lesson. The AI narrows the search space and the strategist still has to do the thinking and activate his brain or her brain in order to make sense of what it reads. And this is how we operate at Y&D, right? Um you don't just sort of uh uh swallow the uh AI uh slop, but rather we make sense of what is actually going on in there. Um, we make sense of what's going on in your business. We make sense of you. We make sense of what's going on in the market. And this is the foundation that drives us to make the right decisions for ad accounts and for the businesses that we work with. And if you're unsure about um your situation right now, need a little help, just get in touch Jeremy on Google Ads. You can also go to um my website by doing a digital dot marketing which will be redone in in the next few weeks. So it'll have a new liquor paint on it which is nice. You can book a call with me there and you can also uh just send me an email at jeremy youngdigital.marketing. This has been Jeremy Young, your personal Google Ads expert and wish you a happy and productive week ahead.