There's a good argument doing the rounds at the moment: nobody visits websites anymore. AI answers the question, the click never happens, and the beautiful site you paid for has quietly become technical debt.

The people making it are mostly right about where things are heading.

They are wrong about why it matters, and they are wrong about what a small business should do on Monday morning.

Here is the harder version.

Most small business websites were never worth visiting

Websites didn't stop being worth visiting. Most of them never were, and the format used to cover for it.

Think about what the average NZ small business website actually contains. A hero image of a building. "About Us" with three paragraphs about passion and quality. A services page listing six things with two sentences each. A contact form. Maybe a gallery.

Now ask the question a search engine has always been asking: why would anyone click this?

There is no answer on that site. No prices. No process. No specifics about how the job actually runs, what it costs, what goes wrong, what you do about it. Nothing a real customer was actually wondering. It's a brochure that happens to live at a URL.

That site was never winning. It was getting traffic because ten blue links meant somebody had to be in position seven, and position seven got a few crumbs. The website wasn't earning the click. The format was handing it out.

AI removed the crumbs. That's the whole story.

This isn't your fault, but it is your problem

To be fair to every business owner reading this: you were sold that site.

An industry spent fifteen years selling websites as design projects. The deliverable was how it looked. The conversation was about hero images, colour palettes and whether the logo should be bigger. Almost nobody was paid to ask "what does your customer actually need to know before they'll ring you", because that's harder to sell and harder to invoice.

So we got a country full of good businesses behind beautiful, empty websites.

The machines reading the web now do not care how it looks. They care whether there is anything on it worth repeating. For a lot of sites, honestly, there isn't.

What we can actually prove

Here's where I'll break ranks with most of the posts on this topic, including the ones I agree with.

Nearly all of them are built on borrowed statistics, usually American, usually enterprise. Big percentages, global brands, numbers pulled from companies with a data team and a product feed. Fine numbers. They tell a Hamilton building company nothing.

We have Google Search Console access across our client sites. So instead of quoting anyone, we went and looked.

The bit that held up

On our own site, the pattern is unmistakable. Full conversational questions rank far better than the short commercial terms we have spent years chasing:

What someone searched	Where we rank
"is it better to outsource social media or hire internally in nz?"	1.3
"what should i ask before signing a contract with a marketing firm?"	2.7
"how much does professional web development cost for an auckland business?"	2.9
"what are the risks of hiring a cheap digital marketing agency?"	3.1
"marketing consultant"	12.6
"digital marketing hamilton"	15.0

Same website. Same domain. Same authority. The only thing that changes is the shape of the thing being asked.

Long, specific, human questions put us on the front page. The short, generic, high-volume terms leave us on page two, where nobody goes. And long conversational questions are exactly how people talk to ChatGPT, Claude, Perplexity and Google's AI Overviews.

That is the shift, visible in our own data, no borrowing required.

The bit that didn't hold up

Now the part you won't read in the other posts.

We tried to prove the scary half. The claim that AI is quietly stealing clicks from sites that still rank fine. If that's happening, it should show up as click-through rates falling at a fixed ranking position.

We took a client site with over a year of continuous Search Console history. First pass looked like a smoking gun: click-through rate in the top three positions fell from 8.6% to 5.8% across the year. A third of the clicks, gone.

Then we controlled for the fact that the mix of searches had changed, and compared only the queries that appeared in both periods. 118 of them.

Click-through rate went up 69%.

The apparent collapse was the site ranking for a different, broader set of searches, not AI eating the clicks. Of those 118 searches, exactly three showed the pattern we were looking for. That's noise, not a trend.

So: on our portfolio of NZ small business clients, right now, we cannot show you that AI is taking your clicks. It may well be coming. It is not yet visible in the data we hold.

We're telling you that because the alternative is repeating something we can't stand behind, and you have no way of checking. Most of what you're reading about AI search is written by people selling the cure.

The advice that follows is much cheaper

If the problem were really that people have stopped visiting websites altogether, the answer would be enormous. Re-platform, adopt agent protocols, build a machine-first data layer, hire engineers.

That is a fine plan if you are a listed retailer with a product feed and a team.

If you turn over two million dollars building houses in the Waikato, it's nonsense.

Because the actual problem isn't your website's architecture. It's that your website doesn't say anything. And that is a writing problem, not an engineering one.

The retrofit, in rough order of what it's worth:

Answer the questions you get asked on the phone. Every one of them, properly, on its own page. What it costs. How long it takes. What happens if it rains. What you do when the council knocks it back. If a customer asks it, it belongs on the site. This is the whole game and it costs you nothing but time.

Put real numbers on the page. Prices, ranges, timeframes. The reflex is to hide them so people have to ring. Hiding them means you have nothing worth citing, and AI will quote whoever was brave enough to publish. Being the only one in your trade who says what it costs is a genuine advantage.

Write like a person answering a question. Question as the heading, answer in the first two sentences. No throat-clearing. Both machines and humans skim exactly the same way now.

Add the plumbing. Schema markup, an llms.txt file, clean structure. It's a day's work and it makes everything above legible to the machines doing the reading. We wrote the practical guide to that here.

Then keep the nice design. This is where I properly part company with the technical debt argument. Your design isn't debt. When AI does send someone your way, they land on a page and decide in about two seconds whether you look like a real business. Aesthetics are what converts the humans who arrive. Gutting them to please a crawler is solving the wrong problem twice.

So what actually changed

Not much, and that's the uncomfortable bit.

Websites have always been judged on whether they deserved the visit. For twenty years the format was generous enough to hide it. Ten links, a few crumbs each, everybody ate.

AI is a harsher marker. One answer, drawn from whoever said something specific and worth repeating. If your site is a brochure, there is nothing to repeat, and there never was. You just used to get away with it.

The businesses that will do well here are not the ones with the best architecture. They are the ones willing to say more, in public, than their competitors are comfortable saying.

That's been true the whole time. It's just that now you can see who's doing it.

If you want the practical version of all this, start with our guide to AI SEO for NZ businesses. It covers llms.txt, schema, structure and the rest, step by step.

And if you'd rather talk it through, get in touch.

Your Website Was Never Worth Visiting. AI Just Made It Obvious.