The website first, the first ad three months later

Why the website alone was not enough

A new site is not found on its own. The keywords showed that people search for a concrete fault or a purchase, not for a company name.

So the work fell into two parts. One is traffic, because climbing the organic results takes years. The other is the site itself, because an ad only takes someone as far as the click.

Each of these without the other is money thrown away. Ads pointing at a weak site: bought visitors who leave. A good site without ads: nobody finds it.

The order: foundations first, ads after

The collaboration began in December 2024, and in the first three months not one forint went on advertising. That time went into the website, and into getting the business onto Google's Business Profile with the right search categories.

The paid campaign only started in March 2025. The order is deliberate: advertising amplifies what is there — with nothing to amplify, you buy visitors at a premium for a place they walk away from. The business profile also brings local search results for free, and paying for those with ads would be a waste.

What this means if your situation is similar

If you run a trade or service business, the question is rarely “to advertise or not”. It is rather what an enquiry costs, and whether it is worth that much. A few weeks of running shows the number, and from then on the decision is made with figures, not feelings.

The other question that came up before this campaign launched: if a hundred visitors arrived right now, what would happen to them? If the site is slow, shows no work, and carries no sign that others were satisfied, most of the ad budget is gone before anything happens.

And before paying for anything: check whether your Google Business Profile appears for the phrases your customers actually use. It is free, and for many trade businesses it brings enquiries on its own. With this client, the first three months of the collaboration were exactly that — with no ad spend at all.

The campaign

Narrow targeting, cost per click kept low

The search campaign is built on few, precisely matching keywords, and on the consistent exclusion of irrelevant searches. That is what keeps the cost per click around HUF 49–50 since the campaign started — while Google's optimisation score stands at 97.5%.

The low click price is not a trick here but a consequence: someone searching for exactly what the ad offers is more likely to click, and the system rewards that with cheaper impressions. Broad targeting achieves the opposite.

Keywords set up to tolerate imprecise searches

People do not search the way the trade talks about itself. They type something, then add another word: a city, a brand, an “urgent”. So the keywords are not set to exact match. This is what I wrote to the client when the campaign launched in March 2025:

“These keywords will be set up so that if the searcher adds something else to their query, the ad can still appear.” Kristóf Karner, handing over the keyword list, March 2025 — translated from Hungarian

The client received the full keyword list in a shared document, to see what the money goes on. In the first weeks a weekly report went out with results and suggestions. A campaign is cheapest to steer in its first month — every correction later costs more.

Budget-based, cyclical running — for the campaign

The campaign does not burn continuously but in cycles: the client tops up the budget when there is capacity for new work, and the campaign runs while the budget lasts. At the end of each cycle a report goes out, and the client decides about the next one.

In a seasonal trade this fits reality better than advertising on a flat monthly fee — in summer, at peak, there is no point paying for more enquiries than can be taken on. The clearest sign the model works: since March 2025, every cycle has ended in a renewal.

Website maintenance runs separately, on a contract with a monthly fee. Splitting the two is deliberate: the ads should burn when there is capacity for new work, but updates, backups and fixes should continue even while the campaign rests.

The June 2025 cycle also ended in a renewal; according to the client's feedback, the number of incoming orders had risen noticeably.

The website that receives the traffic

A custom block theme, no page builder

The website runs on a custom WordPress block theme with its own block patterns — not on a page-builder plugin. This decision is about loading speed: a visual builder generates code in layers, and with traffic arriving from ads that costs money directly, because visitors leave a slow-loading page.

The theme splits into 18 PHP modules, each with one job: image conversion, gallery, reviews, loading screen, header clean-up. A change touches one file, not the whole theme.

Google reviews that refresh on their own

The reviews come from the Google Places API Place Details endpoint, on a daily schedule. The site does not call the API on every load — it reads a local snapshot that the scheduled task refreshes. That is faster and cheaper at once: the Places API charges per call.

The original site did this with a third-party review widget. The custom solution replaced it, so that no third party stands between the most important element of trust and the visitor.

Automatic image conversion

JPG and PNG images uploaded to the media library convert to WebP automatically. For the existing images a custom command-line tool was built (wp cspk webp-backfill), so hundreds of older images switched over in a single run. On a trade site this is not a side issue: the gallery showing the work is the heaviest content on the page.

The parts that are not obvious

The business profile's category decides which searches you appear for

A few days after the business profile went live, the client reported this: searching with one name for the trade, he found his business — with the other, he did not. Both names mean the same work.

The primary and additional categories set in a Google Business Profile decide which searches the business enters the local results for. A single category is usually narrower than what the business actually does — and the missing categories cause not a weaker position but a complete absence for that search.

The only way to uncover this is real searches: trying the phrases customers actually type — not the ones the trade uses for itself.

Brand logos cannot simply be put on a website

Displaying the manufacturer logos of the products you install is trademark use, which needs the owner's permission — typically an official dealership or reseller agreement. So no such logos went onto the site, and the hero image comes from licensed stock.

In the short term this looks plainer than the competitors' walls of logos. In exchange there is no legal risk for a trademark owner to dig up later.

The Google Places API returns at most five reviews at a time

1. 01 Places API at most 5 reviews
2. 02 Merge by author + timestamp
3. 03 Snapshot stored locally
4. 04 Website calls no external API

daily · old reviews stay

The Place Details response carries at most five reviews. If the site simply printed the fresh response, every new review would push an old one out of sight — the trust the business earned over years would turn invisible.

The solution is a merging store: every fetch folds into the earlier snapshot, and duplicates are filtered by author identifier and timestamp. The shown review count can only grow, never shrink.

// a review's identity: author URL + timestamp
$key = trim( $r['author_url'] ) . '|' . $r['time'];
if ( ! isset( $by_key[ $key ] ) ) {
    $new_count++;
}
// fresh API data always wins — the author may edit the text
$by_key[ $key ] = $r;

$ find . -name "*.php" | wc -l ; find . \( -name "*.php" -o -name "*.css" -o -name "*.js" \) -exec wc -l {} + | tail -1
      18
    1795 total

The loading screen and lazy image loading rule each other out

A dark overlay covers the content until everything has loaded, then fades away. This only works if lazy loading is switched off meanwhile — otherwise the images would start loading after the overlay is gone, and the visitor would see exactly the flicker the loading screen is there to hide.

So the theme disables WordPress's native lazy loading, and the plugins', in a separate module. Two optimisations that make sense on their own and hurt together — the kind of thing you only notice in production. On two other projects lazy loading went two opposite ways: on one, turning it off helped; on the other, turning it back on did.

The scheduled task does not run reliably without visitors

WordPress's built-in scheduler only fires when someone visits the site. In a low-traffic period that means the daily review refresh can be skipped. So the system can also be run from the command line (wp cspk fetch-reviews) and driven by real server-side scheduling — the built-in way is just the convenient default.

Questions about this project

Summary

An air-conditioning company's online presence was built from December 2024 in a strict order: first a website and a Google Business Profile with categories matched to real searches, then — three months later, from March 2025 — a search campaign. The campaign runs on a narrow keyword set with consistent exclusions, in budget-based cycles; by July 2026 it had brought 87,740 impressions and 6,478 visits at an average of HUF 49 per click. The website is a custom WordPress block theme with no page builder, in 18 PHP files; Google reviews refresh daily from the Places API into a merging store, because the API returns at most five reviews per call. Images convert to WebP on upload, with a custom WP-CLI command for the back catalogue. Maintenance runs separately from the campaign, on a monthly contract.

← All case studies