Someone wants a dentist in Auckland, or a company that builds bilingual websites, or a rough price for a service. Three years ago they searched Google. A good share of them now ask ChatGPT or Claude instead.
The difference: Google hands over ten links to choose between, while an AI gives an answer that might name two or three companies. Not being named is indistinguishable from not existing — nobody scrolls further, because there is no further.
GEO — generative engine optimisation — is the work of becoming one of the names.
How it relates to SEO
The foundation is shared. Being crawlable, well structured and accurate are prerequisites for both, so “doing GEO instead of SEO” is not a coherent plan.
Above that foundation the logic differs:
SEO competes for position. You need to be better than whoever is eleventh to reach the first page.
GEO competes for the right to be quoted. When an assistant assembles an answer it reaches for material it can safely state: definite facts, traceable sources, internal consistency. It will not quote vague self-praise, because there is no way to restate that as a sentence carrying information.
Which produces a specific consequence: vaguely written sites do badly at GEO even when their SEO is competent.
What AI prefers to quote
Imagine each passage being judged on whether an assistant would repeat it to a user. The criteria become obvious.
Specific numbers. “Affordable pricing” cannot be restated. “Single-page sites from NZ$1,499 including first-year hosting” restates cleanly and remains useful afterwards. That is part of why our prices are on the site at all.
Clear boundaries. “We build all kinds of websites” carries nothing. “We build company sites and booking systems, not e-commerce platforms” is useful — an assistant can decide whether to recommend you. Saying what you do not do makes you easier to recommend than claiming everything.
Verifiable claims. Registered company name, registration number, founding date, address, real client work. These correspond to outside records. We put our legal entity and NZBN in the site’s structured data because both can be checked on a government register, and one checkable fact outperforms ten adjectives.
Direct answers. If the question is “how much does a website cost in New Zealand”, a page titled exactly that which gives a range in its first two sentences gets quoted ahead of an article that reaches the price in paragraph five.
The technical side
Content has to be in the HTML, not painted in by script. Most AI crawlers do not run JavaScript; they read the page as it was before anything executed. If your key content is inserted by script, what they read is empty. We had this: a counter animating from 0 to 500, with 0 in the HTML, so an assistant read “0 websites delivered”.
Accurate structured data. State in machine-readable form what kind of company this is, where it is, what it offers, the price range, which languages. It changes nothing visually and everything about whether a machine understands you correctly.
Consistency across every surface. The name on your website, your Google Business Profile, your social accounts and your invoices should be the same name. When they differ, an assistant cannot confirm these are one company, and will not risk naming you.
We dealt with this ourselves recently: one company had been running two domains under two brand names, which split its own identity in half. Unifying it mattered more for GEO than any technical change.
llms.txt. A plain text file at the root of the site stating concisely what the business does, the main pages, the services and the prices. Not yet a standard, and most crawlers do not look for it, but it costs almost nothing. Ours is public — open it, there is nothing secret about the format.
Finding out whether you are mentioned
No tooling required. Test it yourself, and test it repeatedly.
Ask the way a potential customer would, not using your company name: “which companies in Auckland build bilingual websites”, “roughly what does a small business website cost in New Zealand”. Ask ChatGPT, Claude, Gemini and Perplexity separately — their training data and retrieval differ, and so do the answers.
Note three things: whether you appear, whether what it says about you is accurate, and if you do not appear, who does.
The last is the most useful. Look at the sites that were named and you will usually find they are clearer on one of the points above — more explicit pricing, or better-defined boundaries, or more that can be checked.
Where this stands today
Competition looks roughly like SEO in 2010: most companies have not started, and early positions are cheap to take.
The uncertainty deserves stating too. How these systems retrieve and cite is changing quickly, and what works now may be weighted differently in six months. So our advice is to do the parts that cannot be wasted whatever changes: write clearly, make claims checkable, keep information consistent, get the structured data right. All of it is SEO foundation as well, so there is no wrong bet to make.
Model-specific tricks can wait until they stop moving.
Our SEO and GEO service includes visibility testing across models, which is the self-check above done systematically. You are entirely able to run it yourself; the method is written out above.