Local SEO
Local SEO Audits
Website Development
Google Local Service Ads
CRO
Key takeaways
- Make the H1 the exact question, word for word, and answer it in the first sentence in plain language with a real number or range. That sentence is what gets quoted.
- One topic per URL. Five questions crammed into one page costs the model too much work to extract from, so it goes elsewhere.
- A small three column table often outperforms a 2,000 word guide, because answer engines lift tables and reformat them straight into responses.
- New Zealand is an open field: most content answering local service questions was written for American search, so a genuinely local answer has far less competition for the citation.
To get a blog post quoted by ChatGPT, Google's AI Overviews, Perplexity or Claude, structure the page so the answer is the cheapest thing on it to find: make the H1 the exact question people type, answer it in the first sentence in plain language, follow with a short bulleted list of the things that change the answer, add a small comparison table, and give every question its own URL. Answer engines do not rank pages so much as retrieve passages, and they retrieve the passage that takes the least work to extract.
The video below is the full four step system we use to get local businesses recommended by AI, across keyword selection, a citation audit, getting mentioned in the places the models read, and the website itself. This article is the website part, written out properly, with the New Zealand angle that makes it worth doing here in particular.
Why New Zealand businesses have an unusually open run at this
Almost all the content answering local service questions was written for the United States. It talks about water heaters, DBAs, county permits and prices in US dollars. When someone in Tauranga asks an AI what a reroof costs, whether a plumber needs to be licensed, or what the consent rules are, the model has thin local material to work with and often answers with something imported and slightly wrong.
That is a gap, and it closes. A business publishing genuinely local answers, with New Zealand phrasing, New Zealand regulations and New Zealand dollars, becomes the obvious source to cite while the gap is open. The same page in a US market would be competing with thousands of near identical articles.
Pick the questions first: two buckets
Most businesses want to rank for everything and end up fielding calls from price shoppers. Aim for 10 to 15 high quality terms instead of a hundred random ones, split into two buckets.
- Conversion. Someone ready to spend: what does a reroof cost in Auckland, how much to replace a hot water cylinder, emergency electrician Christchurch. Whoever is retrieved here gets the call.
- Nurture. Someone a month or two out: how long does a Colorsteel roof last, mains pressure versus low pressure cylinders, what does a drainlayer actually do. These do not ring the phone this week, but they build the presence that gets you retrieved later.
Build the list from Google's People Also Ask box, which gives you the phrasing people actually use, then check volumes in Semrush, Ahrefs or Keyword Planner. The usual rule is to drop anything under 30 searches a month. In New Zealand, loosen that: a term with 20 searches a month is worth a page when the job it leads to is worth $15,000.
Find out what AI is quoting right now
Before writing anything, audit what is currently being cited. Open ChatGPT in a temporary or incognito chat so your history is not influencing the answer, then paste your questions in one at a time, not batched. Scroll past the answer to the source list underneath: those URLs are the reason the model answered the way it did.
Log every one in a spreadsheet with four columns: the phrase you searched, the exact URL, the platform it came from, and roughly where it appeared. Log the exact page, not the domain, because the model is pulling one specific passage, sometimes a single comment in a forty comment thread, and where that passage sits on the page tells you a lot.
Then ask the follow up. Something like: why did you pick that business specifically, and what signals led you to recommend them over the alternatives. It will usually just tell you. It will say the business has 312 reviews at 4.8, that ten reviews mention same day service, that the description specifies a particular install. Each of those becomes a line on your to do list.
The structure of a post AI will quote
Once you know the question, the page itself is mechanical. Six rules, in this order.
- The H1 is the exact question, word for word, question mark and all. Do not paraphrase it into something that reads better. The model matches on literal phrasing in headers.
- Answer in the first sentence, directly under the headline, in plain language, with a specific number or range. Do not bury it six paragraphs down behind an introduction about how important roofs are.
- A short bulleted list of the variables that change the answer, immediately underneath. This tells the model your answer is conditional and gives it the conditions.
- A small comparison table, three columns is plenty. Answer engines lift tables and reformat them straight into responses, sometimes verbatim. A good table can outperform a 2,000 word article.
- H2s that are the related questions, each answered in its first line. Every one is another query your page can be retrieved for.
- One topic per URL. Five services or five questions crammed into one page is too much work to extract from, so the model picks a page that is only about the thing being asked.
Here is the same thing as a worked example, in the format it describes.
| Element | What it does | Worked example |
|---|---|---|
| The H1 | Matches the literal phrasing of the question being asked | How much does it cost to reroof a house in Auckland? |
| First sentence | The passage most likely to be quoted back verbatim | Published New Zealand price guides put a full Auckland reroof at roughly $15,000 to $30,000 in 2026. |
| Variables list | Tells the model which conditions change the answer | Roof size, pitch, access, removal and disposal, asbestos, material, spouting |
| The table | Gets lifted and reformatted straight into the answer | Material, installed rate per square metre, expected lifespan |
| The H2s | Each one is another question the page can be retrieved for | Does a reroof need a building consent in New Zealand? |
One thing you can stop worrying about: the meta description. It gets rewritten and the models barely look at it.
Fix the posts you already have
Most sites already have a backlog of content that is actively working against them. The classic is the ultimate guide: 2,500 words that bury the actual answer in paragraph eight, behind history, context and three paragraphs of preamble. Those rank terribly in AI because extracting from them is expensive.
Two fixes, both quick. Rewrite the opening so the answer is in the first sentence. And split pages covering five topics into five pages covering one each, with the old page kept as a hub that links to them. That is the same argument, from a different direction, as why a single services page is costing your business jobs.
Your blog is only one of the places AI looks
Answer engines retrieve rather than rank, which means they pull every mention of your business they can find. If you exist on your own website and nowhere else, there is not enough to support recommending you. The audit in step two tells you which platforms matter for your trade and your city, but for most local service businesses it comes back as some mix of the same four.
- Directories and citations. The listings that verify you exist. In New Zealand that means Yellow, Neighbourly, NoCowboys, Builderscrack, NZS.com and the trade registers, not the American set.
- Forums. Reddit carries a heavy share of citations. Find the threads already being cited, then reply properly: 200 to 400 words, lead with the actual answer, mention your city, mention where you work naturally rather than as a pitch.
- Press releases. One release, properly distributed, puts your business name on 50 to 200 indexed news sites. Head it with the question, not your business name, answer it in 400 to 600 words, and mention the business at the end as the source.
- YouTube. The models pull from it heavily for local service questions and virtually no trade business is there. Three to five minutes, titled with the exact question, recorded in the van. The transcript gets indexed and that is the point.
More detail on the non website half of this sits in how to get your business recommended by ChatGPT and AI search.
What to expect
The specific signals shift, sometimes monthly. The four steps do not, because they amount to being present everywhere the models look and being easy to quote when they get there.
On the plumbing client we built this for, ChatGPT was citing the business by week nine and AI Overviews were surfacing it for broad terms like plumber, hot water heater and emergency plumber, with calls past 300 a month. Across twelve local service businesses running the full system in 2026, it has produced $1.4 million in attributed revenue with no paid ads and no bought leads.
The checklist
- One question per URL, and the question is the H1 word for word.
- The answer is the first sentence, with a number or a range in it.
- A variables list under the answer, then a three column table.
- H2s that are real related questions, each answered in its opening line.
- New Zealand phrasing, New Zealand regulation, New Zealand dollars.
- Old guides rewritten so the answer is at the top, and multi topic pages split.
If you would rather have the whole system run for you, the keyword sort, the citation audit, the content and the profile work, that is our local SEO service. The trade specific versions are on the plumbing SEO and roofing SEO pages.
Read next: will AI search replace Google for local businesses and how to get recommended by ChatGPT and AI search.
