If you've heard "GEO" lately and wondered whether it's a real discipline or a new acronym for selling SEO twice, the honest answer is: both, depending on who's selling it.
Generative Engine Optimization is the practice of making a business visible inside AI-generated answers — Google's AI Overviews, ChatGPT search, Perplexity, Copilot. The term came out of research in 2023 and has since been adopted by agencies, conferences and a fair number of people who will invoice you for it. The mechanics underneath are much less exotic than the marketing.
What actually changed
- People increasingly ask questions rather than type keywords, and get a synthesized answer instead of a list of links.
- Those answers cite sources. They are not hallucinated from nothing; they are assembled from pages that were crawled, indexed and judged relevant.
- The click often doesn't happen. But the brand name does appear — and the ones that appear get the enquiry.
- For a local business, that matters most on the queries that used to bring in good leads: "who builds websites in Auckland", "WordPress developer vs Squarespace", "how much does a website cost in NZ".
How one of these answers gets built
Roughly three steps, and each step tells you where the opportunity is:
- Retrieval. The system finds candidate pages — usually from its own search index, increasingly from a crawl of the open web. If a page isn't indexed or crawlable, it cannot be a candidate regardless of how good it is.
- Passage selection. It doesn't rank whole pages so much as pull passages that answer the question. A page with a clear, self-contained answer wins over a page where the answer is buried in paragraph nine.
- Synthesis and attribution. The model writes the answer and cites the sources it leaned on. Being cited is the prize. Being summarised without attribution is the risk everyone worries about, and it's mostly outside your control.
The uncomfortable part: none of this is a separate index. The pages that get cited are usually pages that rank, or would rank, for the same question. GEO inherits most of SEO's prerequisites — crawlability, speed, structure, authority — and then adds a few of its own.
The things that actually move the needle
- Be crawlable and fast. This is boring and it's most of the battle. Blocked robots.txt, JS-only content, and 6-second loads remove you from consideration before the interesting work starts.
- Lead with the answer. One sentence that states the thing, then the detail. "A small business website in New Zealand typically costs NZ$4,000–12,000 depending on scope" is quotable. Three paragraphs of preamble are not.
- Make content extractable. Short paragraphs, descriptive
h2s, lists and tables. Models re-use structured passages; a wall of text gets skipped. - Add structured data.
LocalBusinesson the homepage (name, area served, phone, price range),Articleon posts,FAQPagewhere you genuinely answer questions,ProductorServicewhere relevant. It doesn't guarantee a citation, but it removes ambiguity about who you are. - Be consistent as an entity. Same business name, address, phone and service list on your site, your Google Business Profile, your socials and the directories. AI answers pull entity facts from many places; contradictions get you dropped or mixed up with someone else.
- Publish the pages people actually ask about. Cost ranges, comparisons, "X vs Y", "how long does it take", local questions. These are the queries that reliably trigger an AI answer, and they're the ones most small business sites never write.
- Keep facts fresh. Dates matter. A page that says "updated October 2026" with current NZ pricing is safer to cite than a 2019 page with no date.
- Earn mentions elsewhere. Reviews, local news, supplier pages, directories, podcasts. Citations follow corroboration.
What doesn't work (or isn't settled yet)
llms.txt. A proposed file for telling AI systems what to read. It's a reasonable idea, adoption is mixed, and no major AI company has agreed to treat it as a standard. Add it if you like; don't pay for it as a strategy.- Keyword stuffing, "AI-optimised" filler, and pages written for a crawler instead of a person. Answer engines are trained on human answers; the same content that reads badly to you reads badly to them.
- Mass-produced AI pages. Google's spam policies explicitly target scaled content abuse. A CMS plus a language model plus no editorial control is a liability, not a content strategy.
- Structured data on its own. Schema describes a page; it doesn't make it worth citing.
What the crawlers look like, practically
If you're deciding what to allow in robots.txt, the names you'll see include GPTBot and OAI-SearchBot (OpenAI training and search), ClaudeBot and Claude-User (Anthropic), PerplexityBot and Perplexity-User, Google-Extended (Gemini training), and Bing's crawler for Copilot.
One decision worth making deliberately: blocking training crawlers is a legitimate choice, but blocking search and user-fetch crawlers removes you from answer results. They are separate user agents. Most businesses that block everything are blocking the citation they wanted.
How to measure it
There is no clean dashboard yet. What I look at:
- Referral traffic from
chatgpt.com,perplexity.ai,copilot.microsoft.com— small, but real, and growing. - Branded search volume (people who saw your name in an answer and Googled you after).
- Direct traffic with no referrer — AI referrals often arrive this way.
- The simplest one: ask new enquiries how they found you. "ChatGPT told me" is now a real answer.
A realistic expectation
GEO is not a replacement for SEO, and it isn't a channel you can switch on. For a small New Zealand business in 2026 the practical order is: be indexed, be fast, be consistent as an entity, publish concrete answers to the questions people actually ask, keep it current. That's the same list as good SEO — the difference is that the passages now need to be quotable, not just present.
If you want to know which pages on your site are most likely to be cited, and which are invisible, send me the URL. That's usually a short conversation.