Most people running ads on Google are watching the wrong number. They count clicks, check CTR, cheer the rising lead count — and at the end of the month they don't know whether they made any money at all. I'll show you how to set up a Google Ads account so that only one thing counts: how much money actually comes back. I'll walk you the whole way — choosing keywords, structuring campaigns, generating ads and landing pages in bulk, and finally measuring the return — and you'll do all of it in Claude Code, the tool that lets one non-technical person run an account like a specialist.
First, a few terms that will keep coming up. Google Ads is the advertising system in Google's search engine — you pay for your ad to be shown to someone who's searching for something. A keyword isn't a single word but a whole phrase someone types into the search box. A landing page is the specific page a person reaches after clicking your ad. And Claude Code is the tool where you assign tasks in plain language — treat it like an experienced specialist you've hired: you're the boss, it does the work.
Start from the money, not from the clicks
The first thing to get straight in your head: Google isn't your advisor. When you open the recommendations tab or take a call from your account rep, you hear advice that almost always leads to one thing — getting you to spend more on the platform. That doesn't mean Google is cheating you. It only means that its interest (your budget) and your interest (your profit) aren't the same thing, and it's your job to look after yours.
So watch the right number from the start. A click is nice, but it doesn't pay the bills. A lead — a form submission — is also nice, but some submissions are spam, mistakes, or people who'll never buy. The only thing that really counts is ROAS — return on ad spend, that is, how much revenue every unit of money put into advertising brings back. This distinction isn't philosophy, it's the mechanics that decide your profit: a campaign optimized for clicks pulls in cheap, random traffic; a campaign optimized for ROAS learns to recognize the people who actually pay — and it's this, not a bigger budget, that turns advertising into a business rather than a cost.
There's one more reason measurement matters most. Picture a business that acquires customers from several sources at once: Google Ads, SEO, social media, industry portals. If you don't know which source brought in the paying customer, you're making decisions blind — you might pour budget into a channel that delivers nothing and cut the one that keeps you afloat. The whole point of what you're about to build is to set up measurement so you know exactly where the people who pay are coming from.
And Claude Code changes one fundamental thing here: work that once required an agency or weeks of manual fiddling is now done by a single person who can't program. You don't have to be a Google Ads specialist — you have to be able to say what you want and understand what's happening underneath. The second part is what I'll teach you here.
Keywords that bring in customers
You won't win in Google Ads if you pick the wrong keywords. This is the foundation and there's no way around it. If you ask the AI alone "which keywords are best," you'll get an answer that sounds sensible and, in practice, burns through your budget — because the model doesn't know how many people actually search a given phrase or how much it costs. That data has to be pulled from the source.
First, tell the "money" keywords apart from the junk ones. Say you run a plumbing business. Someone types "plumber" — and some of those searches will never turn into a customer. "Plumber jobs" is someone looking for employment. "Plumbing course" is someone who wants to get trained. "Plumbing wholesaler" is someone after parts, not a service. You'll pay for a click on each of those phrases anyway if you choose your keywords badly. Your job is to fish out the phrases with buying intent — "emergency plumber," "plumber [your city]," "pipe repair [your city]" — and discard the rest.
The second element is match types — how closely the searched phrase has to match your keyword for the ad to show. Picture a dartboard. Dead center is the perfect phrase — exactly what you're after. But there are few bullseyes, so you have to widen the field. And here you have three options:
- Broad match — Google adds phrases that are "similar in meaning" on its own. You want "emergency plumber London," and the algorithm will show your ad against "plumber salary" or "how to unclog a drain yourself" too. Those are darts thrown at the wall beside the board. As a rule, avoid it — you'll burn budget on people who'll never buy.
- Phrase match — the search has to contain your phrase. "Best emergency plumber London, same day," "emergency plumber central London" — you catch all of these variants. This is the winner, because the same thing can be said a thousand ways, and you want to catch every variant without losing relevance.
- Exact match — only that one phrase exactly. Accurate, but too narrow: most people won't type it word for word, so you lose traffic.
Where do you find these phrases together with the data? In the Google Keyword Planner — a tool built into the Google Ads account. You go to "Tools," then "Keyword Planner," choose "Discover new keywords," and type something broad, e.g. "plumber." Immediately narrow the results to your city — by default Google covers the whole country, and you probably don't want to pay for clicks from the other end of it. You'll get a list of phrases, and beside each, three things the language model doesn't know on its own: the average monthly search volume, the level of competition, and a rough cost per click. Competition isn't a drawback here — where others are paying, there's money to be made.
One more technique that works: the city × service matrix. You take each of your services and cross it with each location you serve — "emergency plumber" for the city and every district or suburban township, then "drain unblocking" the same way, and so on. Each combination gets its own landing page. Note: not every combination makes sense. A big city brings traffic; a small township next door may have so few searches that it isn't worth targeting. The Planner shows you that right away.
One search, one group, one page
Now the most important structural rule — in the jargon it's called SKAG, a single keyword ad group: an ad group built around one keyword. It sounds technical, but it comes down to one idea: what someone typed into the search box should match your ad, the ad should match the landing page, the page the email, and the email the sales conversation. One coherent experience from start to finish.
See it from the customer's point of view. They type "plumber London." If the ad reads "galvanized pipe replacement," they bounce — that's not what they were looking for. If they see "plumber London," they click. And if, after clicking, they land on a page whose headline says "trusted plumbers in London," they'll stay and send a request, because everything confirms they've reached the right place. The most common mistake I see is one generic ad leading to one generic page — it looks like a labor saving, and in practice it lowers your conversion rate and raises the cost of every customer.
That same coherence has a second, hard benefit: cheaper clicks. Google gives each keyword a quality score — a rating of how well your ad and page match what someone searched for. The higher the score, the less you pay for the same placement. And the quality score rests mainly on the relevance of the ad and landing page to the phrase — which is exactly what the SKAG structure builds. The coherence you're building anyway gives you a lower cost on its own — as a side effect.
Connect Claude Code to your account and build a campaign
You connect Claude Code to your Google Ads account once. Setup takes a few minutes and after that you never come back to it. I won't walk you through every click in the Google panel here — that changes, and Claude itself is there for the detailed steps, guiding you through the process as you go. What matters is that you understand the high level: you set up a Google Ads account, plus a so-called manager account (a separate account that manages the first one), you generate an access key (a token) in the admin panel, and in the Google Cloud Console you create a project and the authentication keys that let Claude Code talk to your account. You keep keys and passwords in a separate, secured file — you never paste them into an ordinary document. Once everything's in place, you tell Claude "connect me to my Google Ads account," confirm the login, and you're done.
From here the enjoyable part begins. Google Ads has three levels — it's best to picture them as nested folders on a drive:
- Campaign — the top folder. Here you set the daily budget, the days and hours your ads run, and the locations. I usually group campaigns around a single service.
- Ad group — the folder inside. Under the SKAG strategy it holds exactly one keyword, e.g. "emergency plumber London."
- Ads — the files inside. One keyword produces ads tailored to that keyword, which lead to one page tailored to the same keyword.
When you have Claude build the first campaign (you give it your website address so the tool can research your business and tailor the copy), it's worth knowing what it's configuring for you — because these settings decide whether the money is spent sensibly.
Search network, not "junk" traffic. Google offers several types: search, Performance Max, display, video, shopping, demand. For a service business, what counts is the search network — these are people searching for something right now, with intent. The rest is mostly random traffic: banners on other people's sites, ads in the inbox, in apps. Few people click a banner like that on purpose. So you set the campaign to the search network and leave the rest alone (with one exception — remarketing, which I'll come back to).
Location — presence, not interest. This is the setting that quietly burns the most budget. Target your city and the area around it (for a business that travels to the customer, usually a radius of a few dozen kilometers around the center; for a business the customer comes to, much less). But there's a trap: by default Google sets "presence or interest" — and will show your ad to someone on the other side of the world who is "interested in" your city. Switch it to presence only. And add location exclusions — it's worth excluding most countries besides your own, because otherwise you catch bot traffic and VPN clicks from across the globe that you pay for and that will never buy.
Bidding strategy. You pay per click in an auction system — whoever bids more usually shows higher (and more expensively). You start the account on the maximize conversions strategy — teaching Google to look for people who fill in the form, not just click. Once you've gathered data, you move to optimizing for ROAS, that is, for customers who actually pay. It's the same logic I started with: a click and a lead are stages, but the goal is money.
Negative keywords. These are phrases for which you deliberately block your ad from showing. Some are universal to any service business — job seekers, people wanting to do something themselves, schools, courses, certificates, technical support, existing customers. Claude will build you a shared exclusion list to attach to all your campaigns and reuse over and over. The other part — catching specific junk phrases from the reports — is ongoing optimization, which I'll come back to at the end.
Once the structure is in order, new campaigns and groups come together very fast in Claude Code. This is work that would take weeks by hand — configuring campaigns, groups, ads, exclusion lists — and here one person does it in a fraction of that time, even with no Google Ads experience.
Bulk ads and pages that convert
You find the winning ad by testing many variants — you put out dozens of versions, one turns out clearly better than the rest, and you stick with it for a long time. The trouble is that creating dozens or hundreds of good ads by hand is hours of dull work. In Claude Code you generate them in bulk in a few minutes.
A single search ad isn't one piece of text but a set of elements: a dozen-odd headlines, several descriptions, plus extensions — callouts (e.g. "24/7 service," "no call-out fee," "licensed and insured"), structured snippets (a list of services), and sitelinks. Claude generates all of this at once, with the extensions filled in, error-free. The extensions play a double role: they add specifics and take up more space on screen — an ad that fills a large area catches the eye more often than a small block that's easy to skip.
One thing you have to watch: pinning the main keyword. Google rotates the headlines, testing which works best — but you want the first position to always show exactly what the user searched for. If someone typed "emergency plumber London" and sees "a quote in 2 minutes, arrival today" at the top of the ad, they'll think "what even is this company?" and move on. So pin the core phrase at the top and let the offers rotate underneath. And keep in mind one condition of testing: to crown a winner, each version has to gather enough impressions. A hundred variants with negligible traffic will tell you nothing — you need enough volume for the data to be reliable.
The ad leads to a landing page, so for each ad you build a separate page — with a headline matched to the same phrase. Claude builds it in Claude Code; you can even point it to a visual reference (a screenshot of a page whose look you like, say) and it'll recreate a similar style. At first the page works only locally, on your computer — more on publishing it live in a moment.
What makes a page convert traffic into submissions? A few things that work regardless of industry:
- Social proof, ideally video testimonials. People want to know you can be trusted. The best evidence of how you'll do in the future is how you did before — and a customer testimonial shows that.
- A form on the page itself. Don't hide it on a subpage — every extra click lowers your conversion rate. The form belongs right where the person landed.
- A founder video. A short introduction. It's not about selling the service — someone looking for a plumber knows why they're there. It's about getting them to choose you over someone on the next street, because they've come to like and trust you.
- A fast response to the submission. This isn't a page element, but it's a lever of enormous power. The average business replies to a submission after many hours; calling back within a minute changes the sales numbers dramatically — you're doing exactly what you'd do anyway, just faster.
And most important: don't guess which version of the page is better. I can have my opinion, you yours, a friend at the table yet another — and we can all be wrong. The split test decides: you put out two versions, measure the data, and the data says what works.
Measure the return and optimize
The most important part of the whole puzzle. Setting up the account does nothing — the money is made on optimization. Start with remarketing, that is, reaching again the people who've already visited you. You have two audiences: cold (people who've never heard of your brand — less inclined to buy) and warm (those who've already been on your site). The more contact someone has with you, the greater the chance they'll reach for their wallet. Everything you've done so far concerned the cold audience; now you add the warm one.
For remarketing to work, the site needs a Google tag — a snippet of code that recognizes visitors and lets you show them an ad again later. Claude handles the technical side; you just need to understand what's happening. You can also ask Claude to check for itself whether the tag works correctly (via the Google Tag Assistant tool) instead of wading through raw data yourself. Of the two kinds of remarketing, the most reliable are search ads aimed at people who've already visited you (RLSA) — the same search-network ads, just shown to warm traffic, more expensive per click but with a higher conversion rate. Display is hit or miss, and I'd keep it for remarketing at most.
Next — a dashboard and account audit. Claude Code can build you a simple panel to review on, say, Monday morning: how much you spent, which campaigns are eating budget, what revenue and ROAS they deliver, what to pause and what to scale. You don't even have to build a panel — you can simply ask: "analyze my account and tell me what to improve." And here the negative-keywords topic returns: even with a perfect SKAG structure, the account will start catching phrases you didn't expect. Claude goes through the search-terms report and points out the junk phrases — and often catches things you wouldn't have noticed after years of running the account yourself (e.g. that a phrase you took for a customer query is actually job listings, or the wrong geography). You approve the report, and the tool adds those phrases to the exclusion list. Less junk traffic means a lower cost per paying customer — the simplest lever you have.
That leaves the hardest but most valuable piece: closing the full ROAS loop. I'll show the logic, because the details depend on which CRM you work in:
- URL parameters. You append information about where the user came from to the page address — the keyword, the campaign, the click ID, the source (paid search, not SEO).
- Hidden form fields. The form has fields the visitor doesn't see, which capture that data from the address and submit it along with the request.
- CRM. The submission lands in your database as a lead with a name, contact details, and information on which keyword, campaign, and click it came from.
- Payment signal. When the customer pays the invoice, a flow fires that marks the amount against the specific lead in the database.
- Uploading conversions back to Google. You export this data (e.g. as a CSV file) and upload it back into Google Ads.
That last step is the crux. Without it, Google doesn't know how much a given customer actually paid — it guesses that every lead is worth the same. Once you upload the real amounts, the system learns to look for people similar to those who actually paid, and optimizes the campaign for real profit, not an averaged guess. Only then do you see the return down to the last unit of currency — not CTR, not click count, just money.
Two final things to round out the workshop. The landing page was running locally — to get it onto the internet, you push the project to GitHub (like a cloud drive, but for code) and publish through Vercel, which serves it at a public address; then you swap the destination address in the campaign from the temporary one to the real one. And this whole repeatable process — generating ads, campaigns, exclusion lists — you wrap into a Claude Code skill: an on-demand workflow you invoke with a single command, so you don't have to explain the context from scratch every time.
If you take one rule away from this, let it be this: advertising isn't about doing more, it's about measuring the right thing. Clicks, CTR, and leads are intermediate metrics — what counts is the money that comes back. Once you build the account around return rather than traffic, Claude Code becomes your Google Ads specialist, available on every command. The next step is small and concrete: before you automate anything, sit down with the Keyword Planner and write out ten phrases where someone genuinely wants to buy what you sell. The rest of the puzzle will hold together around that list.