Research · November 2026
How New Zealand business websites score for search and AI.
We ran our website audit over the homepages of 289 New Zealand businesses, drawn at random from six sectors. Most have the search basics in place. Far fewer give an AI assistant what it looks for.
- Sites audited
- 289
- Sectors
- 6
- Checks per site
- Up to 89
- Audited on
- 13 November 2026
78
The median search score, out of 100. Half the sites score between 71 and 85.
Median of 289 sites
61
The median AI readiness score, out of 100. Half the sites score between 46 and 70.
Median of 289 sites
96%
of sites score lower for AI readiness than they do for search.
278 of 289 sites
Search is mostly handled. AI readiness is not.
Every site got two scores out of 100 from the same audit anyone can run free on this site. The search score covers what search engines have asked for over many years: titles, headings, links, images and crawl settings. The AI readiness score covers what helps an AI assistant read, trust and quote a page: structured data, clear answers, named authors, dates and sources.
126 of 289 sites (44%) score 80 or more for search. Only 6 do for AI readiness, and 100 (35%) score under 50.
| Score band | Sites |
|---|---|
| 0 to 9 | 0 |
| 10 to 19 | 0 |
| 20 to 29 | 0 |
| 30 to 39 | 2 |
| 40 to 49 | 1 |
| 50 to 59 | 9 |
| 60 to 69 | 47 |
| 70 to 79 | 104 |
| 80 to 89 | 100 |
| 90 to 100 | 26 |
| Score band | Sites |
|---|---|
| 0 to 9 | 0 |
| 10 to 19 | 0 |
| 20 to 29 | 1 |
| 30 to 39 | 21 |
| 40 to 49 | 78 |
| 50 to 59 | 35 |
| 60 to 69 | 75 |
| 70 to 79 | 73 |
| 80 to 89 | 6 |
| 90 to 100 | 0 |
Number of sites in each ten-point score band. Both charts share one scale.
The basics are mostly there.
Nearly every homepage has a title, loads over HTTPS and is set up for phones. Most say what language they are written in and where to get in touch.
- Has a page title99%of 289
- Loads over HTTPS98%of 289
- Set up for phone screensA viewport tag in the page head96%of 289
- Says what language it is written in89%of 289
- Links to a contact page or section88%of 289
- Has a robots.txt file83%of 289
- Has an XML sitemap77%of 289
Shutting AI crawlers out is rare. 14 of 289 sites block at least one of OpenAI’s GPTBot, Anthropic’s ClaudeBot or Google-Extended in robots.txt. The gap in AI readiness is not about access.
About one in four is missing a basic.
Of 289 homepages, 74 have no H1 heading, 69 have no meta description and 67 do not name a canonical address. Each is a small fix.
The finer points are missed more often. Fewer than half keep their title within the length search results show in full, send the full set of tags that make a link preview, or keep their headings in order.
- Every image has an alt attributeSites with at least one image79%of 283
- Names a canonical address77%of 289
- Has a meta description76%of 289
- Has an H1 heading74%of 289
- Headings run in order, no level skipped49%of 289
- Title is 30 to 60 charactersSites with a title43%of 285
- All Open Graph tags for link previews43%of 289
- Width and height set on every imageSites with at least one image24%of 283
What an AI assistant looks for is mostly missing.
127 of 289 homepages carry no structured data at all: the markup that tells a machine in plain terms what the business is, where it is and how to reach it. Only 98 describe the business itself with most of the key details filled in.
Signs of who wrote a page, and when, are rarer still. 26 homepages show an author, 24 show a published or updated date and 18 set out questions and answers a machine can lift.
- Links to an About page or section73%of 289
- Has structured dataJSON-LD in the page56%of 289
- Describes the business in structured dataOrganization or LocalBusiness with at least 3 of 4 key details34%of 289
- Links its other profiles in structured dataA sameAs list24%of 289
- Has an llms.txt file16%of 289
- Shows who wrote the page9%of 289
- Shows a published or updated date8%of 289
- Sets out questions and answersThree or more headings written as questions6%of 289
llms.txt is a newer convention: a plain file that points AI tools at a site’s key pages. 47 sites have one. 21 of those are among the 22 Wix sites in the sample, which suggests the platform adds it, not the owner. Away from Wix it is 26 of 267.
Plain English is the exception.
We scored the sentences on each homepage with the Flesch Reading Ease formula. A higher number is easier to read, and 60 or more is the range usually called plain English.
54
The median reading ease score. Half the sites score between 46 and 61.
Median of 245 sites
27%
of homepages reach 60, the plain English mark.
65 of 245 sites
Another 44 homepages had fewer than 100 words of sentences in the HTML they send, too little to score, so they are left out of the figures above. In all, 73 of 289 have under 300 words of any kind. The audit reads a page the way a crawler that does not run scripts reads it, so text a site only builds in the browser does not count here.
Fewer than half send any one security header.
These are instructions a server sends with every page, telling the browser what to refuse. They are set once, in hosting.
- Strict-Transport-SecurityAlways use HTTPS45%of 289
- X-Content-Type-OptionsDo not guess file types43%of 289
- X-Frame-OptionsDo not let other sites frame this one33%of 289
- Content-Security-PolicyOnly run scripts from named places19%of 289
On 12 sites the homepage still loads something over plain HTTP, and 6 are not served over HTTPS at all.
Accommodation trails on both scores.
Median scores for each sector, highest search score first. From the top sector to the bottom the gap is 8 points for search and 16 for AI readiness.
Retail47 sites
Search: 81AI readiness: 62Professional services54 sites
Search: 80.5AI readiness: 65.5Health46 sites
Search: 79AI readiness: 66.5Hospitality46 sites
Search: 76.5AI readiness: 57Trades46 sites
Search: 75AI readiness: 56Accommodation50 sites
Search: 73AI readiness: 50.5
Each sector has 46 to 54 audited sites, so a gap of a few points between two sectors is within the noise.
What a site is built on shows.
We could tell what 195 of the 289 sites were built on from their markup. WordPress is the most common (89 sites), then Shopify (25).
Sites where no platform could be identified, a group that includes hand-built and older sites, score lowest on both. Only 27 of those 94 carry any structured data, against 67 of 89 on WordPress.
Shopify25 sites
Search: 89AI readiness: 68Wix22 sites
Search: 85AI readiness: 66Squarespace19 sites
Search: 83AI readiness: 64WordPress89 sites
Search: 77AI readiness: 69Not identified94 sites
Search: 72AI readiness: 47
Platforms with fewer than 15 sites in the sample are not shown. These groups are small, so read the differences as a guide.
How this was measured.
- Where the sites came from
- OpenStreetMap, the open map of the world, lists businesses with their website address. We took every New Zealand listing tagged as hospitality, accommodation, health, retail, a trade or a professional office that had a .nz or .kiwi address of its own. Social pages, directories, booking sites, government and education addresses were left out. So was any address listed against more than three places, so that chains would not crowd out small businesses. That left 7,852 sites.
- The sample
- From that list we drew 60 sites at random from each of the 6 sectors, 360 in all, with a fixed seed so the same draw can be made again.
- What happened to the 360
- Of these, 289 were audited. Another 32 addresses no longer exist: the domain does not resolve. A further 21 could not be loaded: they refused or blocked the request, timed out or returned an error. Then 15 now send visitors to a different domain, so were not counted as the listed business, and 3 ask crawlers to stay out in robots.txt, so we did.
- What was measured
- The homepage only, once, on 13 November 2026, as the HTML the site sends, without running its scripts. Alongside it: robots.txt, the XML sitemap, llms.txt, the response headers and a sample of the links on the page. That is up to 89 checks a site. A check that does not apply to a site is left out of that check’s count, which is why some rows show a number under 289.
- What was not measured
- Page speed. Google’s PageSpeed Insights was not run for this study, so the performance part of the search score rests on the security header checks alone. Inner pages, rankings, traffic and anything a page only shows after its scripts run were not looked at either.
- How precise the figures are
- A share measured on all 289 sites should sit within about 6 points of the figure for the whole list, 19 times in 20. Shares for one sector or one platform are much less precise. Each sector gave the same number of sites although the sectors differ in size, so we also worked every figure out with the sectors weighted back to their real sizes: the median search score moves from 78 to 79, the median AI readiness score stays at 61, and no share shown as a bar on this page moves by more than 4 points.
- How the figures were checked
- We read 8 plain facts a second time (title, meta description, H1, canonical address, viewport, language, structured data and image alt text) from the saved pages, with a separate program that shares no code with the audit. The two agreed on all 2,306 comparisons. Running the audit on these sites also showed five checks giving wrong answers. We fixed them in the live tool first, and the figures here come from the corrected checks.
- What the scores are
- The two scores are our audit’s own weighting of its checks. They are not a Google measure, and a high score does not promise a ranking or a mention in an AI answer.
- No site is named
- Not on this page and not in the data behind it, which holds counts only. Map data © OpenStreetMap contributors, available under the Open Database Licence.
- How to cite this study
- Tally Digital, “How New Zealand business websites score for search and AI (2026)”, published 13 November 2026, https://www.tallydigital.co.nz/research/nz-business-websites-2026.
See how your own site scores
The audit behind this study is free. It runs the same checks on your site and shows what to fix first.