Google Ads for HVAC companies: the build order

For HVAC companies the conversion that matters is a completed, invoiced job, and it lives in your job management system (ServiceTitan, Housecall Pro, Jobber) — not in the ad platform. Google can only optimize toward what you send it, so whatever you pick here decides what the algorithm spends the next ninety days chasing. Pick the wrong one and every later step in this playbook is executed perfectly in the wrong direction.

Google enables these on every new Search campaign, and none of them are keyword decisions — they are switches on the settings screen that decide where the money goes before a single keyword applies. Everything else in this playbook operates on whatever is left after they have taken their cut, so a reader who skips this step spends the rest of it optimizing a remainder.

One flat keyword list is why most accounts in this vertical cannot tell which half of the budget works. Sorting terms by the intent behind them means you can bid, write and judge each group on its own terms — a problem-aware search and a competitor search convert at rates that have nothing to do with each other, and averaging them hides both.

Exact match includes close variants — misspellings, plurals, reordered words, dropped function words. Phrase match now behaves roughly like the old broad-match-modified. Broad match reaches as far as Google judges useful, which with a good conversion signal is genuinely valuable and without one is an open tap. Start narrow, prove the signal, then widen — in that order, because widening is reversible and a mistrained bidding algorithm is not.

Every vertical has a set of searches that look commercial and are not. In HVAC companies those are job seekers, students, people looking for free help, and DIY researchers. Broad and phrase match will find all of them for you. Add this list at the account level on day zero — the first week of an unfiltered account is the most expensive week it will ever have.

Ad groups exist so the ad can match the search. The moment a group holds two intents, the ad has to be vague enough to cover both, and vague ads lose the auction on relevance before budget ever enters the picture. Build these groups, each with its own ads and its own landing page.

Google rewards assets that match the search and the page behind it. It penalizes — and increasingly disapproves — copy that promises something the page does not deliver. Every line below is written to be provable on arrival. Character counts are shown because assets over the limit are rejected at upload, which is where most copied-from-a-blog ad copy dies.

Position is not bought, it is earned and then paid for. Google ranks you on your bid multiplied by how good it expects your ad and landing page to be, plus your assets — so a relevant advertiser at half the bid outranks a vague one, and then pays less per click at the same position. This is the mechanism underneath everything above: tight ad groups and honest ad copy are not tidiness, they are the discount.

Maximize Conversions on completed jobs, with a budget schedule that ramps ahead of forecast heat. Dayparting matters more here than in most verticals — an emergency ad that runs when nobody answers the phone is money set on fire. Smart bidding is a data problem before it is a strategy problem: with thin or wrong conversion data it will confidently optimize toward noise, and it does that faster than a manual account can lose money.

Most accounts are killed or rescued on evidence that could not have meant anything yet. The bidding algorithm is fitting a model, and a model fitted on two weeks of data is fitted on noise. This is the shortest honest schedule: what to look at each stretch, and — more usefully — what to ignore.

An account left alone drifts toward the cheapest clicks it can find, because that is what the system is built to do when nobody corrects it. This is the shortest routine that keeps it honest. It takes about an hour and it is the difference between an account that compounds and one that plateaus at month three.