Guide · AI search optimisation
Eight steps, in order, with what I measured on my own site. Written for owners, not marketers. No tool to buy at the end.
Google shows an AI Overview at the top of a large share of New Zealand searches. ChatGPT and Microsoft Copilot answer "who should I use for" questions directly, naming two or three businesses. Perplexity does the same with citations. In each case the person reads the answer and either contacts one of the named businesses or never clicks at all.
I have the numbers for my own site. Between May and July 2026, impressions in Google Search Console for the queries I rank for fell by roughly half while average position improved, because the Overview answered the question and the click never happened. The enquiries that did come through were better qualified, because the prospect had already read a summary and wanted the person behind it. That pattern is now normal for NZ service businesses, and it means being the source the answer is built from matters more than the click.
The good news is that the work is mostly unglamorous and cheap, and most of your competitors have not done it.
Before changing anything, ask the machines the way a customer would. Open ChatGPT, Google (with AI Overviews showing), Copilot and Perplexity and ask, in plain words: "who does [your service] in [your town]", "how much does [your service] cost in New Zealand", "best [your service] near [suburb]". Ten questions is enough.
Record three things for each: whether you appear, what it says about you (it is often wrong), and which websites it cites. The citations tell you what the model trusts. If it is a directory, a competitor's FAQ page, or a Reddit thread, that is the bar you need to clear. Repeat this monthly; it takes twenty minutes and is the only measurement that matters.
Models quote the first clear answer they find. So do people on a phone. Every page should open with two or three plain sentences that answer the question the page exists for: what you do, for whom, where, roughly what it costs, and what happens next. Then the detail.
This was the single change that did the most on my site. The service pages that were rewritten to answer first climbed in position and the quality of enquiries improved. Every page on this site now has an "In short" block at the top for exactly this reason.
If two of your pages both try to answer "AI automation Auckland", Google splits the ranking between them and a model cannot tell which one is the answer. I made this mistake: my homepage and a service page were chasing the same phrases, and both suffered. Giving each phrase one home, and making the homepage about the business rather than the services, fixed it within weeks.
The test: for each question a customer might ask, can you name exactly one page on your site that answers it? If the answer is "two" or "sort of the homepage", that is the next thing to fix.
Models build a picture of a business from every place it is mentioned: your site, Google Business Profile, LinkedIn, directories, reviews, news. If your name, location, phone number and description differ between them, the picture blurs and you get left out of answers.
ChatGPT's search and Microsoft Copilot rely on Bing's index. A surprising number of New Zealand small-business sites are simply not in it, because nobody ever submitted them and Bing crawls NZ lightly. Sign up for Bing Webmaster Tools, import your site from Search Console, submit your sitemap, and set up IndexNow so every page change is pushed within minutes.
This is the cheapest step on the list and the one most often skipped. It took me an afternoon.
Many AI crawlers do not run scripts and give up on slow pages. If your content is rendered by a page builder or a front-end framework after load, a model may see an empty page. Check by viewing the page source: if the words are not in the HTML, they are not being read.
Aim for under two seconds on a phone, text in the HTML, images with real alt text and stated sizes, and no important content behind tabs, accordions or pop-ups that need a click.
Schema.org markup helps a machine confirm what the page says: that this is a LocalBusiness in Auckland, that this page describes a Service, that these are the FAQ answers, that this article was written by this person on this date. Add Organization or LocalBusiness site-wide, Service on service pages, Article on guides, Person for the owner, and FAQPage only where the questions and answers are visible on the page.
Two warnings from my own testing. Schema that says something the visible page does not is worse than none. And FAQ schema on its own did nothing: Google stopped showing FAQ rich results for most sites in 2023 and models read the visible answer. The same goes for llms.txt, the text file that is supposed to guide AI crawlers. I added one and measured no change over six weeks. It is harmless, so I leave it there, but do not pay anyone for it.
A model is more likely to name a business that more than one source vouches for. That means Google reviews with real text, case studies with numbers and a named client or sector, a LinkedIn page that is actually updated, and mentions in local directories, industry bodies and news. None of this is new; it is the same trust signals search engines have used for years, weighted more heavily.
Ask every happy client for a Google review and reply to each one. Write up your best three jobs as case studies with the before and after numbers. Post the case studies on Google Business Profile and LinkedIn with a link back. That is a month of light work and it compounds.
| Change | Effort | Measured effect |
|---|---|---|
| Answer-first rewrite of service pages | A day per page | Positions up on the rewritten pages, better-qualified enquiries |
| Separating homepage from service keywords | Half a day | Stopped the two pages competing; landing pages took the queries |
| Bing Webmaster Tools plus IndexNow | An afternoon | Pages indexed by Bing within days of each change |
| Consistent schema (Organization, Service, Person, Article) | A day | No ranking change on its own; correct business details in AI answers |
| llms.txt | Ten minutes | Nothing measurable in six weeks |
| FAQ schema without visible changes | An hour | Nothing measurable |
| Google reviews and named case studies | Ongoing | The corroboration AI answers lean on; slowest to build and hardest for a competitor to copy |
It is mostly the same work done more honestly. Clear answers, one topic per page, fast pages, accurate structured data and real reviews help both. The differences are that AI systems read the whole page rather than the title, weigh corroboration from other sources heavily, and in ChatGPT's and Copilot's case draw on Bing's index, which most NZ sites have never been submitted to.
Ask it the way a customer would: "who does X in [your town]", "best X near [suburb]", "how much does X cost in New Zealand". Do the same in Google with AI Overviews on, in Copilot and in Perplexity. Note whether you appear, what it says about you, and which site it cites. Repeat monthly.
Bing typically indexes changes within days once IndexNow is set up. Google takes two to six weeks to re-evaluate a page. ChatGPT's answers about your business can change as soon as its search tool finds the new page, but its baked-in knowledge updates only with new model versions.
I build and retrofit websites to be found by AI search, from NZ$1,000 fixed price. Book a free call and I will check where your business shows up today.
Book a free call