Field Notes16 August 20268 min read
What NZ's vineyards, orchards, and farms are actually doing with AI
The AI wins in NZ primary industries are happening in the office, not the paddock.
The AI wins in NZ primary industries are happening in the office, not the paddock.
The AI coverage in primary industries is almost entirely about robots. Autonomous sprayers navigating between vine rows. Fruit-picking arms moving through orchards faster than human pickers. Computer vision grading lines sorting produce by size and blemish count in milliseconds. The R&D is real and, in some cases, the hardware is already operating at scale overseas.
For the majority of NZ’s primary producers, though, the robots are not the story. A Marlborough family vineyard with 40 hectares under vine is not deploying autonomous sprayers. A Bay of Plenty kiwifruit orchardist supplying Zespri is not running an AI-powered harvest arm. A Waikato sheep and beef operation is not using satellite-linked soil sensors to optimise pasture rotation. Those tools exist, but they are either beyond the economic reach of small-to-medium NZ operations or still maturing toward the reliability a commercial operation requires.
What those same businesses are using AI for is something less dramatic and significantly more accessible: the administrative and marketing work that surrounds the actual production. The inbox. The compliance paperwork. The customer newsletters that keep a cellar door mailing list engaged from harvest to harvest. The social content that stays visible to potential visitors across a long off-season.
That is the practical AI story in NZ primary industries right now, and it is largely invisible because it does not involve any interesting hardware.
Primary industry compliance in New Zealand carries a document load that many people outside the sector underestimate.
Wine producers preparing for export markets manage phytosanitary certification, country-specific labelling requirements, vintage declarations, and chemistry records across every wine in their range. A small producer with twelve wines exporting to six markets has a documentation matrix that can easily consume several weeks of administrative time per year. The underlying information, analysis results, blending records, vintage data, is usually held by the winemaker. Getting it into the right format for each market is a writing and formatting task, not a technical one.
Language models handle this kind of task well. The winemaker dictates or inputs the technical data. The model structures it into the required format for each document type. A person who understands wine and compliance reviews it for accuracy. The writing time drops by 60 to 70 percent; the accuracy stays the same because the review step is not removed.
Horticultural businesses face similar compliance work, particularly those exporting to Japan, the EU, and the US, where phytosanitary requirements, pesticide residue limits, and audit documentation are detailed and market-specific. Bay of Plenty kiwifruit suppliers within the Zespri system, and Hawke’s Bay apple growers exporting to Asia, both carry significant documentation overhead. Seasonal variation adds complexity: the records from each season have to be organised, retrievable, and formatted correctly before the next export cycle begins.
AI doesn’t remove the compliance burden. What it removes is the writing time that surrounds it, which is the part least aligned with the skills that got most primary producers into their business in the first place.
The part of the NZ primary industry sector where AI has the clearest compound effect is the direct-to-consumer layer: the cellar doors, the farm-gate stores, the specialty food brands selling through their own website rather than entirely through distributors.
NZ wine direct-to-consumer has been growing consistently. The Marlborough region alone receives hundreds of thousands of visitors annually, and cellar door visitors convert to wine club members and recurring online customers at rates that matter commercially for small producers. The challenge: keeping those customers engaged between visits across an off-season that spans most of the year.
The content requirement for that engagement is more significant than most small producers budget for. A monthly newsletter to a wine club list of 800 people, consistent social posting across the cellar door’s quieter months, and enough wine-specific content on the website that a language model asked “what’s a good Marlborough cellar door visit for a Pinot Noir enthusiast” can form a confident recommendation.
That last point is where AI search optimisation becomes directly relevant for primary industry businesses with a direct-to-consumer offer. The producers with detailed, specific, publicly available content about their vineyard, their winemaking approach, and the specific wines they make are far better positioned in AI search recommendations than those whose website has a contact page, a trade sheet, and a brief history section. A consumer who asks ChatGPT to recommend a Hawke’s Bay red wine producer worth visiting is going to get a recommendation drawn from what’s publicly readable. The producer with a website that explains their approach to Syrah in the Gimblett Gravels, why that soil type produces the structure it does, and when to visit for the best experience is the one a model can say something specific and confident about.
Building that content is where a website design project and an AIO engagement overlap for primary industry businesses. The content changes that improve AI search recommendations are the same ones that improve the quality of a new visitor’s first impression of the brand. They are not separate projects.
The marketing work in NZ wine is where AI tools have made the most visible difference in the businesses we have worked alongside.
Tasting note drafting is the starting point for many producers. A winemaker who can articulate what makes a wine interesting verbally, but finds sitting down to write a tasting note a slow and effortful process, can dictate observations about a wine and have a structured note ready to review in minutes. The key word is review: tasting notes generated without a winemaker’s input at the editing stage tend to flatten the character of a wine into generic descriptors. With review and editing, they become a useful starting draft rather than a published output.
Media releases to trade publications, export buyer briefing documents, and importer communications follow the same pattern. The winemaker or owner has the substance. They often lack the time or inclination to structure it into a format that reads clearly under deadline. A language model drafts from notes; a person who knows the wine reviews and signs off. The process that previously took three hours takes 45 minutes.
Social media marketing for primary producers works best when the source material comes from the operation itself and the AI layer handles structuring and scheduling. A vineyard posting through harvest season has genuinely interesting content to share: vine-by-vine harvest decisions, early tastings from the tank, the specific conditions that made a particular block ripen unusually late or early. The time to write it and maintain a consistent posting schedule across a harvest is the constraint. AI drafting from voice notes or quick written summaries removes that constraint without removing the voice of the person who actually grew the fruit.
Seasonal labour is a structural feature of NZ horticulture, and the documentation overhead that surrounds it is significant. Worker onboarding materials, health and safety briefings, food safety requirements, and field protocols that change from season to season are documentation tasks that land on the shoulders of operations managers who are simultaneously trying to run a harvest.
An AI-assisted approach to seasonal documentation reduces this burden. Updated safety procedures that need to be communicated clearly in plain language. Onboarding materials that answer the questions seasonal workers most commonly ask in their first week. Field-to-packhouse handover documents that capture what a shift supervisor knows and makes it transferable. Where AI drafting helps most is in reducing the time between “this needs to be written” and “this is ready to distribute,” which in a harvest operation often means the difference between workers having clear information before the busy period begins and not having it until week two.
The review requirement applies here as well. Operational and safety documentation generated without expert review is not a shortcut, it is a liability. The AI layer handles the writing; the operations manager or health and safety advisor handles the sign-off. The division of labour is the same as it is in wine documentation, and the time saving is comparable.
The most common limiting factor in primary industry AI adoption is not cost or interest. It is the format of existing records.
An operation running on paper records, whiteboard schedules, and spreadsheets that were never designed to be queried cannot easily benefit from AI analysis. The data is there, but in a form that requires significant conversion work before it can be used. Spray records in paper logbooks, yield data tracked in custom spreadsheets that differ by season, staff hours recorded in a point-of-sale system that doesn’t export cleanly: these situations are common in small-to-medium primary industry operations, and they mean the most analytically powerful AI applications remain out of reach until the data is structured.
The businesses building digital records now, whether through industry platforms, cloud-based farm management software, or even well-designed spreadsheets maintained consistently, are creating the infrastructure that will allow AI analysis in two to three years. The operations still entirely paper-based when that capability matures will be starting from scratch at the moment their competitors are already on their second iteration.
The practical first step for any NZ primary producer thinking about AI is simpler than preparing for a technology future: audit the two or three administrative tasks that consume the most time each week that aren’t directly about growing, making, or harvesting the product. Compliance documentation, seasonal staff onboarding, customer newsletter writing, social content, trade communications. Those tasks are almost certainly where the first AI tools belong, and the investment to get started is measured in hours rather than capital expenditure.
For direct-to-consumer primary producers, the compounding observation worth sitting with is this: the window to establish AI search visibility in NZ wine, specialty food, and horticultural categories is still open. Most cellar door and farm-gate businesses have not yet invested in the content depth that AI search tools draw on when forming recommendations. The producers who build that content now, for an audience of AI models as much as for human visitors, will be difficult to displace in those recommendations once the habit of asking AI tools for travel and food recommendations solidifies among NZ and Australian consumers. That shift is already underway, and the businesses that notice it now have a genuine first-mover advantage in their regions.
The gains in NZ manufacturing aren't on the factory floor. They're in the admin layer that surrounds it.
Three areas where NZ construction firms are finding real returns from AI, and one where most are still leaving time on the table.
Local search, email, and social: where AI is changing the economics of NZ physical retail.