Small business? Stop wasting money following Google’s recommendations

Google has a lot of recommendations for your ad account.

• Increase your budget.
• Use broad match.
• Add more assets.
• Expand your targeting.
• Turn on automation.
• Try another campaign type.
• Give Google AI more room to optimise.

Google is very enthusiastic about helping you spend money.

Coincidentally, Google also owns the place where the money is being spent.

I’m not suggesting there’s a tiny villain in Mountain View rubbing its hands together every time an electrician clicks Apply recommendation.

But I am suggesting something much less dramatic: Google’s goals and your goals are not always identical.

Google wants its advertising system to have enough reach, enough data and enough flexibility to optimise. You want profitable customers.

Sometimes those things align beautifully. Sometimes they absolutely do not.

“Recommended by Google” does not mean “recommended for your business.”

This matters even more as Google pushes its advertising products further towards AI-led optimisation.

AI Max for Search is a perfect example. It can be powerful. It can also be completely inappropriate for an account that simply doesn’t have the budget, traffic or conversion volume to give the machine a decent education.

First, what is AI Max?

AI Max is an optimisation layer for existing Search campaigns rather than a completely separate campaign type.

When enabled, it can expand how Google matches your ads to searches and automate more of the creative and landing-page decisions.

Two major components are search term matching — going beyond existing keywords using broad match and keywordless technology — and asset optimisation, which can tailor ad text and potentially route users to pages Google believes are most relevant.

That sounds excellent. And in the right account, it might be.

AI needs something to learn from

Google’s Smart Bidding systems optimise using conversion information and contextual signals. More useful conversion data gives the system more evidence. That isn’t controversial. It’s machine learning.

If Google has seen hundreds of good leads come through your account, it has a much richer picture of what a likely customer looks like.

If it has seen three form submissions, one accidental phone call and your mum testing the contact form, we’re asking a lot of the robot.

Imagine two businesses

Business A: large national retailer

It spends $150,000 a month. Thousands of people visit the website. Hundreds of purchases happen every week. Products, prices, conversion values and purchase behaviour are well tracked.

There are thousands of signals for Google to learn from. Giving AI more freedom here can make enormous sense. Google can test, observe, learn, correct and do it again.

Business B: local electrician

Our electrician spends $1,500 a month. Maybe he receives 12 genuine enquiries. Six turn into jobs. Two are irrelevant. One is a customer asking whether he repairs washing machines. He doesn’t.

Now Google suggests expanding beyond his keywords. What could possibly go wrong? Plenty.

Small businesses have a different maths problem

For a large advertiser, an inefficient learning period can be absorbed. For a small business, ten bad clicks might represent a meaningful chunk of the week’s budget.

That’s the difference. Small business advertising isn’t simply large-scale advertising with fewer zeros. The economics behave differently.

A local dentist might only need ten strong new patients in a month. A photographer might need four weddings. A builder might only need two serious enquiries.

So when Google says “We found opportunities to capture more conversions,” the correct response isn’t “Fantastic. Turn everything on.” It’s: “What kind of conversions?”

The conversion problem nobody talks about enough

Google can only optimise towards the signals you give it. And many small-business accounts have terrible signals.

They treat all of these as equal conversions:

• Clicking the phone number.
• Submitting a form.
• Opening the contact page.
• Requesting a quote.
• Booking an appointment.
• Downloading a brochure.
• A qualified sales lead.
• A $15,000 customer.

These are not the same thing.

If you tell Google: “Please find me more people who click the contact button,” Google may become extremely good at finding people who click the contact button. Congratulations. You have optimised for clicking.

What you actually wanted was revenue.

Before handing more decision-making to AI, small businesses need to get much better at feeding Google meaningful information: qualified enquiries, booked appointments, sales, revenue, offline conversions, lead quality and customer value.

Automation does not fix bad measurement. It scales it.

So is AI Max bad?

No. That’s not the argument.

AI Max can extend keyword coverage and use existing keywords, creative and website content to identify additional relevant searches.

For an account with reliable conversion tracking, enough conversion volume, enough budget, a strong website, clear landing pages, good exclusions and controls, and a defined commercial objective, it deserves testing.

But that last word matters: testing. Not blindly enabling. Not clicking “Apply all”. Not assuming Google’s recommendation score is your business strategy.

Google itself gives us a clue

Google’s own AI Max documentation notes that the feature won’t be effective if the campaign is limited by budget.

That is an important caveat for small businesses because they are often operating with tightly controlled budgets.

If your campaign already has more useful searches available than your daily budget can capture, expanding reach even further may not be the first problem worth solving. You may be better off becoming more selective.

The electrician test

Imagine an electrician in Brisbane spending $50 a day. His profitable work is switchboard upgrades, EV charger installation, renovation electrical work and emergency fault finding.

His current Search campaign is already capturing strong queries like “electrician north Brisbane”, “switchboard upgrade Brisbane”, “EV charger electrician” and “emergency electrician near me”.

AI Max could discover valuable searches he hasn’t considered, such as “home EV charging installation” or “old fuse box replacement”. Excellent.

But it might also explore adjacent searches that look semantically relevant but aren’t commercially useful. That’s where enough volume, budget and feedback matter.

At $1,500 a month, every experiment costs something. You don’t necessarily want your entire budget becoming Google’s training set.

What I would do instead

Step 1: capture the demand that already exists

Start with people clearly looking for the thing you sell: strong Search campaigns, clear geographic targeting, high-intent queries, relevant landing pages, negative keywords and accurate conversion tracking. Boring is underrated.

Step 2: work out what a conversion is actually worth

If one service generates $400 jobs and another generates $8,000 jobs, Google should ideally know the difference.

Step 3: collect enough evidence

Understand which queries convert, which services are profitable, which suburbs generate work, which days matter, which landing pages work and what a qualified lead actually looks like.

Step 4: expand deliberately

Test broader matching or AI Max against your existing structure. Ask whether it generated more qualified business at an acceptable cost, not merely more clicks.

And then there’s Demand Gen

Demand Gen is designed to reach people across visual Google surfaces including YouTube, Shorts, Discover, Gmail, Maps and the Google Display Network.

That is enormous reach. Wonderful. Your Brisbane electrician does not need three billion people. He needs approximately 30 kilometres of them.

This doesn’t make Demand Gen useless. It means we need to understand what job it does.

Search captures demand. Demand Gen helps create it.

Someone searching “emergency electrician Brisbane” already has intent. Something is broken. Possibly smoking. This person does not need a beautiful cinematic brand story. They need Kevin and his voltage tester.

Search is strong here because it intercepts an existing need.

Demand Gen works higher up the journey. Someone watching renovation videos on YouTube might not currently need an electrician, but could be planning a renovation. A well-targeted video introducing an electrical renovation specialist may create awareness before the search happens.

That can be useful. But awareness requires scale, creative, frequency, testing, time and, unsurprisingly, money.

Google’s own Demand Gen guidance is revealing

Google’s current Demand Gen performance guidance recommends a daily budget of at least 10 times the target CPA and says a campaign needs to generate at least 50 conversions to build a good foundational understanding of which impressions are successful.

The important strategic point is that these systems like data. They like room to learn. They like volume.

A $1,500-a-month local campaign is playing a very different game from an advertiser spending tens of thousands.

Demand Gen is also absorbing Display

Google has been moving traditional Display activity towards Demand Gen. This matters because the advertising ecosystem is moving towards broader, AI-assisted campaign systems operating across more inventory.

The direction is more automation, more surfaces, more algorithmic optimisation and less manual micromanagement.

For sophisticated advertisers with good data, this can be extremely powerful. For advertisers with poor measurement, it can simply make it easier to spend inefficiently at scale.

Automation changes the job of the advertiser

Ten years ago, a Google Ads specialist could spend an impressive amount of time changing keyword bids by 17 cents. AI is making that increasingly pointless. Good. Nobody should build a career around manually changing 17 cents.

But the value doesn’t disappear. It moves.

• Setting the right business objective.
• Structuring useful conversion signals.
• Understanding margins.
• Knowing which customers matter.
• Producing strong creative.
• Building useful landing pages.
• Feeding first-party data back into platforms.
• Recognising when automation is making a stupid decision.
• Knowing when not to follow a recommendation.

The dentist test

Imagine a dentist offering general check-ups, emergency dentistry, Invisalign, implants and cosmetic work. Google sees one conversion: Contact form submitted.

But commercially these leads are very different. A check-up enquiry might be worth a few hundred dollars. An implant patient may be worth thousands.

If the dentist gives Google increasingly broad control without connecting lead quality or value back into the platform, the system can optimise towards what is easiest to measure rather than what is most valuable.

That isn’t necessarily AI failing. That’s the business giving AI the wrong exam question.

The photographer test

Now imagine a photographer shooting family sessions, corporate headshots, weddings and commercial campaigns. Her biggest margin comes from weddings and commercial work, but family-session forms convert more easily.

If every form submission counts equally, automation can naturally favour volume. Again: more conversions does not automatically mean more business value.

Before turning on more Google AI, fix these seven things

1. Conversion tracking

Can you confidently say which campaigns generate genuine customers? If not, stop here.

2. Lead quality

Can you distinguish a qualified enquiry from spam, irrelevant calls and tyre-kickers?

3. Conversion value

Can you tell the system that different customers have different commercial value?

4. Geographic controls

Are you spending money only where you can genuinely serve customers?

5. Landing pages

If Google sends a user to a page, does that page answer the query properly?

6. Budget

Do you have enough budget to allow experimentation without starving your highest-intent activity?

7. Creative

Particularly for Demand Gen, do you actually have strong images and video? Putting terrible creative across billions of potential eyeballs does not magically make it good. It just makes the embarrassment scalable.

When Demand Gen actually makes sense for a small business

Demand Gen can make sense when a business has a strong visual offer, enough budget beyond core Search, a meaningful remarketing audience, existing customer data, a service people research over time, good video and image assets, enough conversion volume to learn from and a desire to grow awareness rather than merely harvest existing demand.

A home builder might use it to stay visible during a six-month consideration journey. A cosmetic dentist might educate people about treatment options before they search for a provider. A wedding photographer might build familiarity months before couples begin requesting quotes.

But if the business has $1,000 a month and hasn’t even captured existing high-intent Search demand properly? I’d probably solve that first.

A simple small-business Google Ads hierarchy

Layer 1: measurement

Know what generates revenue.

Layer 2: capture obvious demand

Reach people actively searching for your services.

Layer 3: improve conversion

Make the website, landing pages, calls and forms work harder.

Layer 4: improve data quality

Feed qualified leads, sales and value back into the platform.

Layer 5: expand intelligently

Test broader matching, AI Max, remarketing and adjacent demand.

Layer 6: create demand

Use Demand Gen, YouTube and other channels to reach customers earlier.

Most small businesses don’t need to begin at Layer 6. Yet advertising platforms have a remarkable tendency to suggest exactly that.

Why Google recommendations can be dangerous

Recommendations aren’t malicious. Many are genuinely useful. The problem is context.

Google can see an advertising account. It cannot fully understand your cash flow, staffing limitations, margins, whether you actually want more customers, which service you hate delivering, whether your best technician is away, or whether one conversion became a $30,000 job and another was Dave asking whether you sell extension leads.

A recommendation can be mathematically logical inside the advertising platform while commercially stupid outside it. That’s why human judgement still matters.

One rule I wish every small business would follow

Before clicking Apply, ask: What business problem is this recommendation solving?

Not: What will this do to my optimisation score? Not: Google says performance could improve, should we do it? What actual problem?

If you don’t know, don’t apply it yet.

AI isn’t the problem

This isn’t really an anti-AI argument. Quite the opposite. AI can make advertising dramatically more effective. It can process signals humans can’t, identify patterns we would miss and make millions of auction-time decisions.

But automation works best when humans give it good goals, good data, good creative, good constraints and enough volume to learn.

The future of Google Ads isn’t humans versus AI. It’s humans deciding what matters, and AI doing more of the optimisation.

Which means the worst possible strategy is probably: “Google recommended it, so we switched it on.”

The bottom line

AI Max isn’t rubbish. Demand Gen isn’t a scam. Broad match isn’t evil. Automation isn’t coming to steal your credit card while you sleep.

But sophisticated tools need appropriate conditions.

For a large advertiser with hundreds or thousands of conversions, strong first-party data, significant budgets and experienced people overseeing the account, Google’s increasingly automated ecosystem can be incredibly powerful.

For a small local business generating ten leads a month? You need to be much more selective.

Capture the obvious demand first. Fix your tracking. Know what a qualified lead is. Understand what different customers are worth. Build enough useful conversion history. Then give the machines more room.

Because Google has one enormous advantage when you’re experimenting with Google Ads.

It’s experimenting with your money.

Maybe make sure the experiment is worth running.