Search is where American advertising gets expensive. Legal, insurance, home services and medical categories run into double-digit dollars per click, and at those prices the errors that merely leak money elsewhere become urgent. This is what an AI tool genuinely needs to do for US Google Ads, and how to check before you pay for it.
The two-minute test that eliminates half the market
Ask whether keyword selection uses live Google Keyword Planner data through the official Google Ads API, or whether a language model is asked which keywords are popular.
The second approach is extremely common, because Planner access requires an approved Google Ads developer token and a live account while guessing requires neither. The output reads beautifully and contains terms with no real search volume. The resulting campaign gets almost no impressions, and nothing in the interface explains why.
💡 Check it yourself. Take five keywords the tool suggests, paste them into Keyword Planner with the United States selected, and look at the volume column. If several return a dash or near-zero, the list was generated rather than researched. This single test tells you more than any demo.
Negatives are where the money is saved
The part no vendor demo shows. Every keyword is a commitment to pay for clicks that match it; every negative is money kept. For a US business, several categories are worth building before spending a dollar.
- 1Job seekers. "attorney jobs", "plumber hiring", "careers", "salary", "how to become". A large share of search volume around any profession is people wanting to enter it.
- 2Free and DIY intent. "free", "how to", "DIY", "template", "yourself", "cheap".
- 3Research and definitional. "what is", "meaning", "vs", "reviews of" — sometimes valuable, frequently not, and worth separating rather than mixing.
- 4Out-of-area terms. Location targeting helps but does not stop keyword-level leakage; a Phoenix business can pay for "Chicago" terms.
- 5Wrong-product terms specific to you. A roofing contractor does not want "roof rack". No tool guesses these — you supply them, and they are often the most valuable negatives on the account.
A tool that generates keywords but not negatives, and does not keep extending the list from the search terms report every week, has done the easy half and left you the expensive half.
Match types, and why broad is sold to you
Exact
Least reach, least waste. The right starting point below roughly a few thousand dollars a month, and in expensive categories at any budget.
Phrase
Sensible middle once negatives are built and you are reading search terms weekly.
Broad
Genuinely powerful at scale with strong conversion signal feeding it. Dangerous before that. A tool defaulting you to broad on a small budget is optimising for Google's revenue, not your return.
A note on Performance Max
Google pushes it hard and it can work well, particularly for ecommerce with a healthy feed and real conversion volume. Two honest caveats for a smaller advertiser.
It gives up control and visibility — you see far less about where spend went and why. And it needs conversion data to optimise against; run it before you have reliable conversion tracking and it will spend efficiently toward nothing. For a service business with few conversions a month, a tightly built Search campaign is usually the better use of the same dollars.
Conversion tracking, or none of it means anything
Worth stating bluntly because it is the most common silent failure in the whole discipline. A campaign optimising toward a conversion that never fires is optimising toward nothing: the bidding has no signal, the reporting has no truth, and any tool sitting on top has nothing real to act on.
💡 We ran a campaign optimised for content views into category pages where that event did not fire. Dozens of clicks, zero recorded conversions, and a platform with nothing to learn from. The ads were fine. The measurement was not, and on a conversion objective those are the same thing. Verify the event fires on the exact page your ad points at — not your homepage — before you scale anything.
For US service businesses specifically, most enquiries are phone calls. If your tracking counts only form submissions you are measuring the minority and optimising toward it. Call conversion tracking and call extensions are not optional extras here.
Search and social in one place
Most AI ad tools cover one platform. That is tolerable until you want to know which channel produced your customers, at which point two dashboards with two attribution models becomes a genuine problem. There is also a budget argument: search captures existing demand, social creates it, and the right split shifts as you grow. Making that call across two products with two bills is harder than it needs to be.
What good looks like
Real volumes
Keywords researched against US Keyword Planner data, with geo targeting verified against the API rather than assumed.
Negatives from day one
Jobs, free intent, research terms and wrong-product terms in place before launch, extended weekly.
Conservative match types
Exact and phrase until conversion signal justifies broadening.
Calls counted
Call conversions tracked, not just forms. Otherwise you are optimising toward the wrong half of your leads.
Acts after launch
A keyword optimiser working on live performance, not a report you have to interpret yourself.
Keywords from real search volumes, not guesses
Adyft builds US Search campaigns on live Keyword Planner data through the official API, and runs them alongside Meta from one place. Fourteen-day free trial.
Start free — no card needed