Build a Local Google Ads Campaign With AI: The Prompts I Use

This week I built two Google Ads search campaigns with an AI agent doing the clicking. Around 100 keywords each, a negative keyword list, two ads per campaign set up as a clean test, location and budget dialed in. It took an afternoon instead of days.

Here's how an AI Google Ads campaign actually gets built when you do it right: the steps, the prompts I use for each one, and the decisions I never hand to the agent.

Who decides what in an AI Google Ads campaign?

Get this straight before you open Google Ads. The agent is fast at the work. It has no business making the calls.

One more thing first: make sure conversion tracking works. A campaign without it can't tell you which keywords bring leads.

Step 1: Pull keywords from Keyword Planner

Keyword Planner lives inside Google Ads. Google's Keyword Planner help page says you need to finish account setup, including billing, to use it. It shows estimated monthly searches and average costs for keyword ideas.

I have the agent pull ideas from a handful of seed terms, then build a list of about 100 phrase match keywords. Phrase match, written in quotes, means ads may show on searches that include the meaning of your keyword, per Google's keyword matching guide. It's a middle ground between too loose and too narrow.

Why 100? Because it's easier to trim a list later than to guess what you missed. After a few weeks, the search terms report shows which keywords bring leads and which just spend.

Step 2: Build the negative keyword list

Negative keywords let you exclude search terms, so you don't pay for people who were never going to hire you. Google's negative keywords page covers how they work.

Say you run a made-up garage door company, Gravois Garage Door Co. You'd want to block searches about garage door technician jobs, how to replace a spring yourself, opener manuals and replacement parts. None of those people are calling you for a repair.

Step 3: Write two ads with only Headline 1 pinned

Responsive search ads let you give Google up to 15 headlines and 4 descriptions, and Google mixes them. According to Google's responsive search ads guide, headlines max out at 30 characters, descriptions at 90, and the two display path fields at 15 each.

Here's my setup. Two ads per campaign. Everything is identical except Headline 1, which I pin to position 1 in each ad. One ad leads with a speed message. The other leads with a different angle, like a guarantee you actually offer. That one pinned headline is the A/B test.

Google says pinning reduces the combinations it can try and can affect ad strength. I accept that trade on one headline, because I want to learn which message wins. Everything else stays unpinned so Google can still mix.

Step 4: Rotation, location and budget

Ad rotation

For a clean test, I set ad rotation to "Do not optimize," which rotates ads indefinitely so both get a fair shot. Google's ad rotation page says that isn't recommended for most advertisers, and notes that Smart Bidding strategies use the Optimize setting automatically. Check your bid strategy before you count on an even split.

Location

I targeted the St. Louis DMA. A DMA, or Designated Market Area, is the TV market region Nielsen draws around a metro, and Google lists Nielsen DMA regions as a US location target. For a business that serves the whole metro, it's a clean way to cover it without listing dozens of towns. If your area is tighter, use a radius or zip codes.

Also check the location option for presence. You usually want people in or regularly in your area, not anyone anywhere who searched about it.

Budget

These campaigns run at $10 a day. That's a starting budget for testing, not a recommendation for you. Know that Google's overdelivery rules let a campaign spend up to twice its average daily budget on a given day, but no more than 30.4 times the daily budget in a month.

The review pass: what I caught

The agent did the clicking well. Not perfectly.

In one ad, it pinned the test headline to position 2 instead of position 1. It reported the step as done, and in its mind it was. But a headline pinned to position 2 isn't the test I designed. If I hadn't opened the ad and looked, I'd have spent weeks comparing two ads that weren't testing what I thought.

Here's what I check on every agent-built campaign before it goes live:

• Every pin. Open each ad and confirm which headlines are pinned, and to which position.
• Locations. Right region, right presence option, no stray targets.
• Networks. Search only, unless you meant otherwise.
• Budget and bid strategy. Exactly what you approved, to the dollar.
• Negatives attached. A list that exists but isn't applied to the campaign does nothing.
• Keywords. Right match types, nothing added that you didn't approve.
• Final URLs. Every ad points to the right page, and the page loads on a phone.

That's why the build prompt ends with a read-back of what's on screen. The agent catches some of its own mistakes that way. I catch the rest.

Why the decisions stay human

An agent can build a technically perfect campaign for a business that shouldn't be running ads yet. It won't ask whether your phones are covered, whether your landing page makes people want to call, or whether you can even take on more work next month.

I covered the costliest versions of those misses in 5 Google Ads mistakes, and how to think about the number itself in how much a small business should spend on marketing.

The agent made the work fast. It didn't make it right. Right came from deciding what to build and checking that it got built.

What happens after launch

Launch is the start of the work, not the end. After a couple of weeks, open the search terms report, add negatives for anything irrelevant, pause keywords that spend without bringing leads, and look at which Headline 1 is pulling ahead.

That's how the list of 100 shrinks to the keywords that actually earn their spot. The agent can pull the report and suggest changes. You still approve every one of them.