Skip to main content

Two audits, measured

Can a poorly built website hurt my chances of being found on AI search?

5 MINUTE READ

By Joseph de Silva, Founder, Visual State Studio

Joseph holds degrees in computer engineering, cyber security and graphic design, and did commercial photography and videography for over seven years before founding the agency.

Published 25 August 2026. Updated 25 August 2026.

OWNER QUESTION

Can a poorly built website hurt me on AI search?

BEST FOR

Owners who suspect their website is holding them back on AI search

Introduction

Yes, and not in a vague way. An AI answer is assembled from what a crawler can read on your pages, so a site that hides its text, says the wrong thing in its code, or has nothing worth quoting is invisible to the answer even when it looks fine to a person. We audited two Melbourne business websites this month and the failures were specific, measurable and fixable.

What an AI actually sees when it visits your site

When ChatGPT, Perplexity or Google’s AI summary reads a website, it isn’t looking at the design. It reads the page code: the headings, the sentences, the business details written into the structure, the links, and whether all of that arrives without a browser doing extra work. A page can look polished on your screen and be almost empty to a machine. That gap is the whole problem, and it’s why “poorly built” has a precise meaning here. It doesn’t mean ugly. It means the machine can’t read it, can’t trust it, or can’t find anything in it worth repeating.

Two Melbourne sites, measured

This month we audited two established businesses in Melbourne’s south east, a pool builder and a louvre roof company. Both had real reviews, real photography and years of real work behind them. Neither was being named by AI answers for the searches that win them jobs.

Neither business did anything wrong. Both paid people who claimed to know what they were doing. One site came from one of those website gurus who know it all, who sold the same template to several competing pool companies and never changed the business details inside it, so every one of those builders is now telling Google they are the same company in Queensland. The other was clicked together in a drag and drop page builder by someone who treated a website like a box of Lego: snap the blocks in, admire the picture on the screen, send the invoice. A website is not the picture on the screen. It is what a machine reads underneath, and neither of these people ever looked.

The pool builder’s two websites each scored 25 out of 100 on an AI readability check: findable, but poor on quotable, understandable and trustworthy. On the day we checked, Google’s AI answer for the premium search in their market named four builders, and neither of theirs was one of them. The louvre roof company was cited by all four AI engines we tested for its own brand, and by three of the four for repair questions, because a plain service page existed to be quoted. For the discovery question that brings in new work, who installs motorised opening louvre roofs in Melbourne, it was absent from all four. Same business, same site. The only difference was whether a page answered the question.

The five ways a site goes quiet

Everything we found fell into five patterns, and each one is checkable on your own site.

  1. The code says the wrong thing about who you are. Every website carries business details in its code that visitors never see: name, address, phone, the map location. The pool builder’s site had another company’s name, an address in Queensland and a Canberra phone number on every page, inherited from a template that had been copied across several dealer websites. The louvre roof company’s code put the business in the wrong state. An AI that’s told two different stories about where you are trusts neither.
  2. There is nothing to quote. An answer repeats sentences. The pool builder had 21 suburb pages that were 84 per cent identical to each other, three articles all published in one month nearly two years ago, and no project write-ups at all. The louvre roof company’s pages ran between about 60 and 393 words. Slogans and product specifications give an AI nothing to carry into an answer.
  3. The text isn’t really there. Answers written in images, content that only appears after a click, headings missing from the code. Seven of the louvre roof company’s eight marketing pages had no main heading at all. We caught a version of this on our own new home page this week, before it went live: the answers in our FAQ only rendered when someone clicked a question, which means a crawler would have seen four questions and no answers. The point is that a modern, well designed site can still hide its best content by accident.
  4. It’s broken on the phone the buyer is holding. The pool builder’s homepage took 8 to 9 seconds to show its main image on a phone and loaded 34 style files and 34 code files to do it, nearly half a megabyte of which was never used. The louvre roof site rendered every one of its 21 pages as a fixed 980 pixel desktop canvas, so every mobile visitor scrolled sideways. Google measures this directly, and slow, broken pages get crawled less and trusted less.
  5. The trust signals are missing or fake. No named people behind the business in the code, no published dates, and in one case a five star review typed into the website by whoever built it, dated 2013, which Google was showing in search results in place of 36 real reviews averaging 4.9. That one breaks Google’s rules outright.

The two minute test

Open your homepage, right click, choose View Page Source, and search the code for a sentence you know is on the page. If it isn’t there, a crawler doesn’t see it either. Then search the source for your suburb and your phone number, and check they’re yours. Then ask ChatGPT and Google who does what you do in your area and see whether you’re named. You’ll have most of the answer in two minutes, and the self-test walks through the rest.

What fixes it

The repairs are the same work that makes a site good for people: correct business details in the code, one page per real question with a plain answer in real sentences, project write-ups with the actual job in them, real reviews connected to the site, named people, and a build that loads fast on a phone. Where a site was built on a template that hides all of that, patching rarely gets there, which is the case we make about rebuilding. It is also exactly the work our Growth stage does.

One more test, and it’s free. Ask whoever built your website to open the page source with you and explain what’s in it: the business details, the headings, where the text lives. Anyone who actually builds websites can do that in a minute. Anyone who can’t has just told you what you paid for.

FAQs

My website looks good. Can it still be invisible to AI?

Yes. Both sites in our audits looked fine to a person. The failures were in the code, the structure and the words, none of which show up on a screen.

Does page speed affect AI search?

Indirectly but really. Google uses speed in its own rankings, and AI answers are built from search results, so a slow site starts behind. A site that takes 8 to 9 seconds on a phone is also losing the humans, which matters more.

Can I fix this without a new website?

Sometimes. Wrong business details, missing headings and a fake review can be corrected on the existing site. Thin content and a slow, template-locked build usually cannot, and that is when a rebuild is the cheaper path.

SOURCES

Would rather show us than read about it?

Book an Operations Review and we will follow one real example through your business.

30 MINUTES. NO PAYMENT TAKEN.