Google Ads captures existing intent. It does not create demand. Every judgment below follows from that, including the ones that cost us money to say.
So the first question isn't how to run Google Ads. It's whether the intent you'd be buying exists in a form you can buy profitably. Nine checks decide it. One missing usually means fix that thing first. Two or three missing means the right recommendation is not to run Google Ads yet, and we'd rather say so before you spend than after.
Here's the gate, yours to run whether you ever talk to us or not. There's also an interactive version that does the gate-2 arithmetic for you.
The nine checks
1. Enough high-intent search volume. Not total volume. Searches from people who could realistically buy. A few dozen relevant searches a month in a service area means there may be no inventory to win, and demand generation should come first.
2. Economics that survive the CPCs. Work backwards from gross profit per customer to the maximum you can afford per click. Can you buy a customer for substantially less than that customer creates? If not at realistic click costs and conversion rates, keep the money.
3. A budget that buys enough attempts to learn. Not a dollar minimum, because the same budget is generous in one market and trivial in another. The sanity check is roughly 100 or more genuinely relevant clicks a month, preferably enough volume for 10 to 20 meaningful conversions. Below that you're arguing about noise.
4. A definition of a good lead. If every form fill enters Google as equally valuable, Google gets very good at finding whoever fills out forms most cheaply, who is not necessarily the buyer. Lead-gen accounts need the whole chain: click, lead, qualified lead, opportunity, customer, revenue.
5. Tracking proportional to your sales cycle. A six-month cycle isn't disqualifying. A six-month cycle where the only signal Google sees is "contact form submitted" means optimizing against a weak proxy. Long cycles need CRM discipline and offline conversion data.
6. A sales process that isn't already the bottleneck. Ads won't fix unanswered phones, untouched forms, or a two-week quote. Buying more leads to solve a sales-operations problem is a faster way to waste money.
7. A website that can close the click. Paid traffic makes a bad conversion experience more expensive rather than curing it. Sometimes the right move is spending the first month's ad budget on the landing page.
8. Non-brand that can stand on its own. If the only campaign that works is your own company name, that's brand defense. Legitimate, but it isn't a scalable acquisition channel and nobody should describe it as one.
9. Tolerance for testing losses. Search terms need testing, ads need to fail, landing pages need comparing. If losing $2,000 learning that an offer doesn't convert would hurt the business, keep the $2,000.
Compressed to one line: enough existing search intent, margins that support the likely acquisition cost, a budget that buys enough qualified traffic to learn, and the ability to measure what happens after the click.
Why your brand ROAS is lying to you
This is the part that costs us money to publish, so here it is plainly: reported brand ROAS overstates what paid search contributed.
Someone who searched your company name was already coming. The campaign collects credit for the last click on a customer who was arriving anyway. A 20x return on brand search is a measure of capture efficiency, not proof of incremental growth.
It's the same failure retargeting has, for the same reason: attribution rewarding the last touch on a customer already in motion. Treat brand numbers as directional until a holdout test says otherwise. Keep brand and non-brand separated in the first report rather than the fourth. And ask the question Google will never ask you: what happens if we stop paying for this traffic?
What the first two weeks should look like
The account has to be trustworthy before it's optimizable. An inherited account doesn't get rebuilt on day one unless something is plainly broken.
Observe, baseline, fix measurement, remove obvious waste, then make structural changes with evidence behind them. Two weeks of that produces a findings document worth more than two weeks of changes: what the measurement actually captures, where the spend actually goes, and a prioritized testing roadmap. If an agency rebuilds your account in week one, ask what evidence the rebuild was based on.
Don't inherit my settings. Inherit my reasoning.
Every account accumulates scar tissue. Broad match disabled because it produced bad leads six months ago. A state split into its own campaign because it closes at twice the national rate. An odd negative keyword that exists because one irrelevant query spent $4,000. Performance Max deliberately excluded from brand.
Someone can arrive, apply every current best practice, produce a beautifully clean structure, and destroy six months of learned context.
So every account we manage carries an operating thesis held outside Google Ads: what you sell and what a customer is worth, which conversion Google optimizes toward, what counts as a qualified lead, which campaigns create demand and which capture it, what brand traffic is allowed to claim, target CAC ranges and the reasoning behind them, geographic differences that matter, known bad traffic, tests already run and how they turned out, things that look wrong and are intentional, what would justify increasing budget, what would justify cutting it, and what remains unknown.
And a decision log, not just Change History. Change History says target CPA moved from $120 to $145. It doesn't say it moved because sales reported this campaign's leads close at 31% against 12% elsewhere while impression share was constrained. The second sentence is what the next operator needs.
Four rules that keep it useful:
- Never optimize immature data. Know your conversion lag and set the date after which a cohort can be judged. A business that closes leads 21 days after the click can't have last week's campaigns declared broken on Monday.
- Don't make five changes and claim a lesson. Change bidding, keywords, budget, landing page and conversion definitions together, watch CPA improve, and you've learned nothing about which move worked.
- Define the failure condition before the test starts. Decide in advance what would stop it. That's what keeps a $2,000 test from becoming a $15,000 test on the argument that it just needs more data.
- Watch the marginal dollar, not average ROAS. A brand campaign reporting 20x may have nowhere useful to put another $1,000, while a non-brand campaign reporting 4x may absorb another $20,000 profitably. The question is always where the next dollar goes.
Where we refuse to compete
The standard isn't whether Google approves the ad. It's whether the tactic is accurate, useful to the searcher, and something we'd be comfortable explaining publicly.
On competitor bidding, the line sits between their customers and their identity. Comparison-shopping searches are fair game when the ad plainly identifies us and sends the visitor somewhere genuinely useful. Putting a competitor's name in a headline to manufacture relevance, implying endorsement, imitating their branding, or intercepting clear navigational intent is not. Google's policy approval is not legal clearance, so anything unusually aggressive goes to a lawyer rather than being declared safe because Google accepted the keyword.
The other refusals, stated so you can hold any agency to them:
- Brand campaigns don't disguise weak acquisition. Brand and non-brand stay separated in reporting, always.
- Conversions known to be garbage don't get optimized toward, however good the dashboard looks.
- Broad match, Performance Max and automated bidding don't get switched on because Google recommends them. Automation amplifies whatever signal it's given, so more freedom for Google is not automatically an upgrade.
- Optimization Score doesn't get chased. Nothing changes to turn a percentage green.
- Attribution doesn't get made harder to understand on purpose. No blending to flatter, no quiet attribution-window changes to rescue a month, no view-through conversions presented as incremental customers.
- Spending doesn't continue merely because stopping would reduce a management fee.
That last one is why we set a client's spend band at onboarding and review it quarterly rather than continuously. If recommending a cut immediately cut our own fee, the recommendation would never get made on the merits.
One note on platform change
Google is moving from keyword plus ad plus landing page toward business data plus website plus creative plus conversion signal, with AI assembling the result. The practical consequence for you: your website has become a targeting input, which means SEO changes can now change paid behavior, and the two need a shared change log.
The migration calendar underneath that shifts often enough that any date printed here has a short shelf life. As of September 2026 the AI Max transition is running on a published schedule, and the specifics are best read at Google's own documentation rather than taken from an article. We keep our own copy verified and dated, and we re-check it quarterly while the migration runs.
The test, one more time
Run the nine checks against your own numbers before anyone sells you a campaign. If two or three come back missing, the useful next move is fixing those, not launching. A business that fails the gate and runs anyway usually concludes that Google Ads doesn't work, when what happened is that it was never going to work yet.
We run paid media as the demand-capture layer beside organic, which is what covers months 1 and 2 while search compounds. Pricing is published, including the part where a recommendation to cut your spend doesn't cut our fee. And if you'd rather start with what AI assistants and Google already say about you, the free AI Visibility Audit takes about two minutes and needs no call.