There are two completely different things people mean when they say they are using AI for Google Ads, and confusing the two is why most advice on the subject is useless.
The first is Google's own AI. Smart Bidding, Performance Max, automatically created assets. That is AI you switch on inside the account, and Google decides what it does with it.
The second is bringing your own AI, something like Claude or ChatGPT, and pointing it at the thinking that sits around the account. Research, ad copy, working out what a customer is worth, reading your own data back to you, working out what has gone wrong. That is AI you direct, and you decide what it does.
Most businesses have tried the first without meaning to, because Google turns a lot of it on by default. Almost nobody has properly tried the second, and that is where the leverage actually is.
What Google's AI already does in your account
Three things, mostly, and they are already running whether you chose them or not.
- Smart Bidding sets a bid for every single auction using signals you will never see. It genuinely beats manual bidding, but only once it has conversion data to learn from. Switch it on in a brand new account and it is guessing.
- Performance Max spends across Search, Shopping, YouTube, Display, Gmail and Discover from one campaign. It works when you have volume and clean conversion data. It quietly burns money when you do not, because you cannot see where the money went.
- Automatically created assets let Google write headlines and descriptions by reading your landing page. Sometimes useful. Sometimes it writes something you would never say about your own business.
The pattern underneath all three is the same. Google's AI is only ever as good as the definition of success you hand it. If your conversion tracking counts every form submission including the spam, the AI will go and find you more spam, very efficiently. That is not the AI failing. That is it doing exactly what you asked.
What bringing your own AI does
Completely different job. It never touches the auction. It does the work that surrounds it, which is the work most small teams skip because it is slow and nobody enjoys it.
- Research. What people actually type, turned into ad groups that make sense, with the negative keywords written before you spend rather than after.
- Copy and creative. Headlines and descriptions at volume, and ad concepts you can actually test rather than one you settled for.
- The numbers. What a customer is worth, what you can afford to pay for a lead, what your bid cap should be, with the arithmetic shown.
- Diagnosis. Why a campaign is not serving. Why cost per lead has climbed. Which search terms wasted money last fortnight.
- Reporting. Your own account data read back to you in plain English, saying what happened and what to do about it.
None of that requires a new platform. It requires the AI to be able to see your account, and for you to ask it properly. Here is a real example of what asking properly looks like.
I can afford [£X] per lead and roughly [X]% of clicks turn into leads on my landing page. Work out what my maximum cost per click should be, show me the arithmetic, and tell me what it means if my conversion rate turns out to be half what I assumed.
Note what that prompt does. It hands over the two numbers only you know, asks for the working rather than the answer, and asks what happens if the assumption is wrong. Most people ask an AI a vague question, get a confident vague answer, and conclude AI is not much use for this.
Where AI should not be trusted
We run this setup on client accounts every day, so this list is written from things that have actually gone wrong rather than from caution.
- It will invent a figure rather than admit it does not have one. Give it your data. Never ask it to estimate your conversion rate, your margins or your competitors' spend.
- It does not know your business. It has no idea that one of your products has a 12% margin and another has 60%, unless you tell it, and it will happily optimise you towards the wrong one.
- It is confidently out of date on platform features. Ad platforms change monthly and models are trained on older documentation. Check anything it tells you about a setting against the platform itself.
- Do not let it change budgets or bids unsupervised. Being wrong is cheap when it is writing headlines and expensive when it is spending.
Where to start
In this order, because each step makes the next one work.
- 1Fix your conversion tracking, and prove it works by testing it yourself rather than trusting the dashboard. Everything else rests on this.
- 2Work out what a customer is actually worth to you, using gross profit rather than revenue. One number, written down.
- 3Use AI for research and copy first. It is the lowest risk place to start and the difference is obvious immediately.
- 4Then use it for reporting, so you are reading your account rather than staring at it.
- 5Only then turn Google's own automation up, because now it has a clean signal to learn from.
Most people do this list backwards. They switch on Performance Max in week one, feed it a broken conversion signal, spend three months and conclude that Google Ads does not work for their business.