Real questions.
Straight answers.

AEO is the practice of making your business easy for AI systems — ChatGPT, Gemini, Claude, Copilot and the answer engines built into search — to understand, trust and recommend. Where SEO optimised pages to rank in a list of results, AEO optimises your knowledge so AI cites you as the answer itself.

Practically, it means publishing real answers to real customer questions, structuring your business information with schema markup, and building a knowledge catalog AI systems can consume.

SEO competes for a position on a results page that a human will scan and click. AEO competes to become the source an AI trusts when composing a single answer. SEO rewarded keywords, backlinks and technical tuning. AEO rewards complete, specific, structured, trustworthy information.

The two overlap — good AEO content usually helps your SEO too — but the mindset shift is real: you're no longer writing for a ranking. You're writing to train AI to trust you.

Brand visibility score is the percentage of monitored AI responses in which your brand is mentioned across a defined set of tracked questions or prompts.

Share of voice is your brand’s proportion of all tracked brand mentions, compared with the mentions received by your competitors across the same set of AI queries.

Competitor visibility is a measure of how frequently each monitored competitor appears within AI-generated responses. It helps show which brands are gaining or losing exposure across your tracked prompts over time.

Sentiment analysis is an assessment of the tone AI systems use when referring to your brand. Mentions are classified as positive, neutral or negative and may be represented using a score from -100 to +100.

AI search is a broad term for finding information through AI-generated answers rather than conventional search-result pages. Related terms include answer engine optimisation (AEO), generative engine optimisation (GEO), AI visibility optimisation and LLM optimisation.

No — but its role is changing fast. Your website remains the human-facing shopfront: it builds trust, converts visitors and gives people somewhere to land. What's changing is that more and more customers get answers from AI before they ever reach a website — and increasingly, AI agents will act without visiting one at all.

The website stays. But it needs a knowledge engine behind it: a structured, evolving source of truth that feeds AI systems accurate information about your business.

A knowledge catalog is a structured collection of everything your business knows and wants AI systems to understand: your services, service areas, pricing guidance, FAQs, policies, processes, team credentials, reviews, case studies, booking rules and your best answers from real customer conversations.

In the old internet, your website was the hub. In the new internet, the knowledge catalog becomes the foundation — the deeper source of truth behind your digital presence.

A search where the person gets what they need without clicking through to any website. Around 58% of Google searches now end this way, and AI-generated answers are accelerating the trend.

That's why being cited inside the answer now matters more than being ranked somewhere below it.

Yes — at least as guidance. People ask AI what things cost, constantly. If you don't publish pricing information, AI gets its answer from somewhere else: a competitor, an overseas article, an out-of-date directory. That answer may be wrong, and it certainly won't explain your value.

You don't need to publish your full price book. Explain the typical range, what pushes cost up or down, what's included, and why cheaper quotes vary. That builds trust with customers and gives AI accurate material to work with.

Agentic booking is when a customer asks their AI assistant to book something on their behalf — "book a heat pump service with a company that has good reviews and clear pricing" — and the AI completes the booking directly, no website visit required.

Restaurant booking already works this way on some platforms, and home-service CRMs are building native AI integrations now. Businesses whose booking systems AI can talk to will capture this demand; businesses that only have a phone number will miss it.

Two ways. First, it answers your visitors' questions 24/7, capturing leads and reducing repetitive enquiries. Second — and this is the strategic part — it records real customer questions at scale.

Those transcripts show you what customers ask before contacting you, which services confuse them, which pricing questions keep coming up, and which objections block enquiries. That's not just support data. That's your content strategy, told to you by your own market.

JSON-LD is structured data embedded in your website that describes your business in a machine-readable format: your services, locations, prices, opening hours, reviews, credentials and FAQs. Humans never see it; AI systems and crawlers rely on it.

Good schema removes ambiguity. Instead of hoping AI infers what you do from your paragraphs, you state it precisely — which makes accurate citation and recommendation far more likely. This very page carries FAQ schema, for exactly that reason.

A vector database stores your business knowledge as embeddings — mathematical representations of meaning — so AI can search it by concept rather than exact keyword. Ask it "why is my room still cold?" and it finds your heat pump troubleshooting content, even though the words don't match. Vector databases are commonly used for semantic search, recommendation systems, image similarity, and RAG-powered AI assistants. Instead of looking only for exact words, it can find items that are similar in meaning or context by comparing their position in a high-dimensional vector space. It's the engine behind accurate, uniquely trained chatbots today, and the infrastructure your business's own AI agent will run on tomorrow.

Sooner than most businesses think. AI answers already divert more than half of searches. Agentic booking integrations are being built into major service platforms now. Agent-to-agent protocols — the plumbing that lets a customer's AI talk to a business's AI — already exist.

The realistic window is one to three years. The advantage goes to businesses that build their knowledge foundation before the wave arrives, because trust with AI systems compounds over time.

Two metrics matter most. Mention rate: how often AI names your business when answering relevant questions. Share of voice: how much of the answer draws on your information versus your competitors'.

We measure both by testing realistic customer prompts — not keywords — across ChatGPT, Gemini, Claude and AI search, then tracking how the numbers move as your knowledge catalog grows.

You can — and it will probably hurt you. Generic "10 tips for winter maintenance" content is what Google calls commodity content, and they've said plainly that it damages rather than helps your visibility. It could belong to any business, so it earns trust for none.

The problem isn't using AI. It's using AI without real input. Feed it fifty real questions from your chatbot transcripts and your actual business knowledge, and AI becomes a genuinely powerful content tool. The quality of the output depends entirely on the quality of the input.

A 30-minute video or phone session where we look at how AI currently describes and recommends your business, identify your biggest gaps, and give you at least three actions you can take immediately — whether or not you ever spend a dollar with us.

No pitch deck. No jargon. No obligation. Book yours here.

Andy Mundell — 20 years and more than 500 New Zealand websites built through On.co.nz and On.Works, hundreds of uniquely trained chatbots and knowledge bases deployed through Cleva.Bot, and AI strategy, execution and training delivered inside Kiwi businesses across every sector.

The short version: unlike consultants who talk the talk, we actually do the work. The longer version is here.

An event listener is a piece of code, commonly used in JavaScript, that waits for something to happen and then runs a response. The event might be a button click, a key press, a form submission, a page scroll, or a screen resize. Event listeners are a core part of making web pages interactive.

A bot is software that performs automated tasks, usually over a network or inside an application. Some bots handle useful work such as answering customer questions, processing forms, monitoring systems, or indexing content. Others can be harmful, such as spam bots or scraping bots, so businesses often use security tools to manage unwanted bot activity.

Artificial General Intelligence (AGI) refers to the idea of an AI system with broad, human-like ability to learn, reason, and solve unfamiliar problems across many domains. Unlike today’s specialised AI tools, AGI would not be limited to a narrow task or training purpose. It remains a research goal rather than a proven, widely available technology.

Natural Language Processing (NLP) is the area of AI that helps computers work with human language. NLP is used to understand text, analyse sentiment, detect intent, summarise content, translate languages, extract key information, and support conversational tools such as chatbots and voice assistants.

Reinforcement learning (RL) is a machine learning method where software learns by trying actions and receiving feedback. Helpful actions are rewarded, while poor choices are discouraged, allowing the system to improve its decision-making over time. It is useful when an AI must learn the best strategy in situations where the right answer is not known in advance.

Retrieval-Augmented Generation, or RAG, is a technique where an AI model retrieves relevant information from trusted sources before creating its response. Instead of relying only on what the model learned during training, RAG connects the model to current or business-specific knowledge such as website pages, policies, product data, FAQs, or support documents. This improves accuracy, relevance, and trust.

Transformers are a neural network architecture designed to understand relationships within sequences, such as words in a sentence, code, audio, or biological data. Their attention mechanism helps the model decide which parts of the input matter most, which is why transformers are central to many modern language models, translation tools, speech systems, and generative AI applications.

Cognitive search is AI-enhanced search that looks beyond exact keyword matches. It can interpret intent, understand language, and search across documents, databases, FAQs, manuals, and other content sources to return more useful answers. For businesses, it helps users find the right information even when they do not know the exact wording used in the source material.

AI agents are software systems that can take actions towards a goal, rather than only producing a single answer. They may interpret instructions, gather information, use tools, call APIs, ask follow-up questions, and decide the next step based on context. A customer support agent, for example, might check a knowledge base, answer the customer, or escalate the issue to a person.

A database is an organised digital store of information. It can hold text, numbers, files, images, records, transactions, and many other data types. Databases are usually managed through database management systems, which allow people or applications to add, update, search, retrieve, and protect information efficiently.

Structured Query Language (SQL) is a programming language used to work with relational databases. Developers and analysts use SQL to create tables, add records, update data, delete data, search for information, join related tables, and manage how database information is stored and retrieved.

An audio-to-text converter turns spoken audio into written text. Also known as speech-to-text or transcription software, it can create meeting notes, subtitles, searchable transcripts, call summaries, captions, and content that can be analysed or reused in other systems.

Embeddings are numerical representations that help AI systems compare meaning, similarity, and relationships. Text, images, audio, products, or documents can be converted into vectors so a computer can identify items that are conceptually close, even when they do not share the same keywords. Embeddings are a key part of semantic search and RAG systems.

Logistic regression is a statistical machine learning method used to estimate the probability of a limited set of outcomes, often yes/no or true/false. For example, it can help predict whether a customer will buy, a lead will convert, or a user will click a button based on patterns in previous data.

Python is a general-purpose programming language known for being readable, flexible, and widely supported. It is used for web development, automation, data analysis, scripting, AI, machine learning, software development, and system integration. Its large ecosystem of libraries makes it popular with both beginners and experienced developers.

A Large Language Model (LLM) is a very large AI model trained to understand and generate language. Tools such as ChatGPT, Claude, and Gemini are examples of products powered by LLM technology. These models typically use transformer architecture and are trained on huge volumes of text and code so they can answer questions, draft content, summarise information, translate, and reason through tasks.

An Application Programming Interface (API) is a defined way for one software system to communicate with another. APIs describe what requests can be made, what information must be sent, and what response will come back. They allow websites, apps, databases, payment systems, AI tools, and third-party services to work together.

Transfer learning is a machine learning technique where an existing model is reused as the starting point for a new but related task. Because the model has already learned useful patterns, it often needs less data, less training time, and less computing power than building a new model from the beginning.

Artificial Intelligence (AI) is the field of creating computer systems that can perform tasks normally associated with human intelligence. These tasks may include recognising images, understanding language, making predictions, generating content, detecting patterns, recommending actions, and solving problems from data.

Prompt engineering is the practice of writing clear instructions, context, examples, and constraints so an AI model produces a better result. A good prompt can shape the model’s tone, format, role, reasoning style, source usage, and output structure. It is especially important when using generative AI for business tasks.

A neural network is a machine learning model made from layers of connected mathematical units, loosely inspired by the way brains process information. By adjusting the connections between these units during training, neural networks can learn patterns in images, text, audio, numbers, and other data.

Text-to-speech converts written text into spoken audio. It allows software to read information aloud, which is useful for accessibility, hands-free use, voice assistants, navigation, education, customer service, and any situation where listening is easier than reading from a screen.

Autoregressive models make predictions by looking at what came before in a sequence. In language, this often means predicting the next word or token based on previous words. The same idea can also be used in time-series forecasting, audio, code generation, and other tasks where earlier values help predict what comes next.

A framework is a reusable structure for building software or engineering solutions. In programming, it usually includes prebuilt code, tools, conventions, and patterns that help developers create applications faster and more consistently. Frameworks can also guide how a project is organised and how different components should work together.

A RESTful API is a web API that follows REST principles for exchanging information between systems. It commonly uses standard HTTP methods such as GET, POST, PUT, and DELETE, along with URLs that represent resources. RESTful APIs are widely used because they are simple, scalable, and easy for different applications to consume.

Hyperparameter tuning is the process of testing different training settings to improve a machine learning model. Hyperparameters are choices made before or during training, such as learning rate, model depth, batch size, or regularisation. Tuning compares different combinations to find the settings that produce the best performance.

A SQL database is a relational database that stores information in tables made up of rows and columns. SQL is used to create, query, update, and manage the data. SQL databases are useful when information needs to be structured, consistent, searchable, and related across multiple tables.

SIP trunking is a way to make and receive phone calls over an internet connection instead of traditional analogue or ISDN phone lines. It uses Session Initiation Protocol to carry voice, and can also support video and messaging services. Businesses use SIP trunks to modernise phone systems and reduce reliance on legacy lines.

Machine learning is a branch of AI where computer systems learn patterns from data instead of being programmed with every rule. After training on examples, a machine learning model can make predictions, classify information, detect anomalies, recommend actions, or improve decisions when new data is supplied.

Extensible Markup Language (XML) is a structured text format for storing and sharing data. It uses custom tags to describe information in a way that both humans and machines can read. XML is often used when systems need a strict, self-describing format for exchanging data.

A relational database organises data into tables and defines relationships between them. Each table stores information about a type of thing, such as customers, products, or orders. Keys are used to connect related records, which makes it possible to query information across multiple tables without duplicating everything.

JSON, short for JavaScript Object Notation, is a lightweight text format for exchanging structured data. It represents information using objects, arrays, names, and values, making it easy for people to read and for software to parse. JSON is commonly used in APIs, configuration files, and web applications.

A Content Delivery Network (CDN) is a distributed network of servers that stores copies of website assets closer to users. By serving images, scripts, stylesheets, videos, and other files from nearby locations, a CDN can reduce load times, lower bandwidth pressure, and improve website reliability.

An API key is a unique string used to identify and control access to an API. It helps the API provider recognise which application is making a request, apply usage limits, monitor activity, and block unauthorised access. API keys should be protected because they can grant access to services or data.

Middleware is software that sits between different systems and helps them communicate. It can translate data, route messages, manage authentication, connect applications, and coordinate services that were not originally built to work together. Middleware is often the glue in complex business software environments.

Data augmentation expands a training dataset by creating modified or additional examples from existing data. This might include cropping images, rewording text, adding noise to audio, or generating synthetic examples. It helps machine learning models become more robust when real-world training data is limited or unbalanced.

A chatbot is software that lets users have a conversation with a system through text or voice. Older chatbots often followed fixed scripts, while modern AI chatbots can use natural language processing, business data, and generative AI to answer more complex questions and support customers at scale.

Generative AI is AI that can create new content, such as text, images, code, audio, video, ideas, summaries, and conversations. It learns patterns from training data and uses them to produce original outputs in response to prompts. Businesses use generative AI for content creation, support, research, automation, product design, and knowledge work.

GraphQL is a query language and runtime for APIs that lets applications request exactly the data they need. Instead of calling multiple endpoints or receiving unnecessary fields, a client can describe the shape of the response it wants. This can make API integrations more flexible and efficient.

JavaScript (JS) is a programming language used to add behaviour and interactivity to websites and web applications. It powers features such as menus, sliders, popups, maps, form validation, live updates, and dynamic page changes. Alongside HTML and CSS, it is one of the core technologies of the web.

Foundation models are large AI models trained on broad datasets so they can be adapted to many different tasks. Instead of building a model from scratch for every use case, developers can use a foundation model as a base and then prompt, fine-tune, connect, or specialise it for a particular application.

MySQL is an open-source relational database management system that uses SQL. It is commonly used to power websites, applications, content management systems, ecommerce platforms, and business software. MySQL is popular because it is mature, widely supported, and well suited to many web-based use cases.

Low code is an approach to software development that uses visual tools, reusable components, templates, and configuration instead of relying entirely on hand-written code. It helps teams create internal tools, workflows, forms, dashboards, and business applications faster, while still allowing custom code where needed.

A workflow is the ordered set of steps, decisions, tasks, and handovers required to complete a process. Workflows can be manual, automated, or a mix of both. They are used to make business processes clearer, more repeatable, and easier to improve.

A repo, short for repository, is a central place where developers store and manage source code and related project files. Repositories usually include version control, which allows teams to track changes, collaborate safely, review updates, and restore earlier versions when needed.

Edge computing moves computing power and data processing closer to the devices, sensors, systems, or users that need it. Instead of sending everything to a central data centre or cloud service, some work happens locally or nearby. This can reduce latency, save bandwidth, and support faster real-time decisions.

Synthetic data is artificially generated data designed to resemble real-world data without directly copying individual real records. It can be created through simulations, algorithms, or generative AI. Synthetic data is used for testing, research, privacy-aware development, model training, and scenarios where real data is limited, sensitive, or difficult to obtain.

GPT stands for Generative Pre-trained Transformer. It refers to a type of AI model built on transformer architecture and trained to generate language and other content from prompts. GPT-style models are used in chatbots, summarisation, drafting, coding support, search experiences, and many generative AI tools.

PostgreSQL is an advanced open-source relational database management system. It supports SQL, structured data, JSON-style data, transactions, indexing, extensions, and complex queries. It is widely used for web applications, analytics, geospatial systems, enterprise software, and data-driven products that need reliability and flexibility.

A Response Library is a curated collection of approved answers that a business can reuse across chatbots, websites, support teams, and AI search experiences. It can be built from FAQs, chatbot conversations, customer service transcripts, product knowledge, and expert-reviewed responses. Its purpose is to make answers consistent, accurate, reusable, and easier for AI systems to cite.

Markdown is a lightweight way to format plain text using simple characters. It can create headings, links, bold text, lists, quotes, code blocks, and tables without complex formatting tools. Markdown is popular for documentation, AI prompts, README files, notes, and content that needs to move cleanly between systems.

Deep learning is a type of machine learning that uses neural networks with many layers to learn complex patterns from large amounts of data. It is especially useful for tasks involving images, speech, language, recommendations, and other problems where the relationships are too complex for simple rules.

A knowledge base is a structured collection of information that helps people or systems answer questions and solve problems. It may include FAQs, help articles, policies, product details, troubleshooting guides, training notes, and internal documentation. For AI systems, a knowledge base can provide trusted source material for accurate responses. AI knowledge bases are vectorised for easy consumption by LLMs when processing RAG queries.

Semantic markup is HTML or structured content that describes the meaning and purpose of information, not just how it looks. For example, using proper headings, lists, tables, navigation, article tags, and schema helps browsers, search engines, accessibility tools, and AI systems understand the page more accurately.

Vibe coding is an informal term for building software by describing what you want to an AI coding assistant and iterating on the generated result. Instead of manually writing every line first, the developer guides the direction, tests the output, fixes issues, and refines the application through prompts, feedback, and review.

CSV data is tabular information stored in a plain-text format called Comma-Separated Values. Each line usually represents a row, and commas separate the values in each column. CSV files are widely used for importing, exporting, sharing, and analysing data because they are simple and supported by spreadsheets, databases, CRMs, and many software tools.