By Theresa Brady, founder of Creative Content. Dunedin, September 2026.

That is one of several patterns we found, and it matters because ranking on Google and being recommended by an AI are related jobs but not the same job. Google ranks pages. An AI answer names businesses. A page can sit on the first page of Google for exactly the thing you do and still never be named, and the gap between those two outcomes is what this article is about.

Everything here comes from our own tests or from what the platforms have published. Every claim carries a date. The results move month to month, and we re-test them every six weeks.

What the labels mean

You will see this work sold under several names. They describe the same job: making sure that when an AI assistant, or the AI answer at the top of a Google result, is asked a buying question, your business is one it names, and that what it says about you is right.

AI SEO is the broadest label: search optimisation done with AI answers in mind as well as Google's ranked results.

AEO, answer engine optimisation, is the older term. It came from the featured-snippet era and means writing so that a machine can lift a direct answer from your page.

GEO, generative engine optimisation, came from a 2023 academic paper and means shaping content so that systems which generate an answer, rather than list results, draw on it and cite it.

LLM SEO and LLMO mean the same as GEO with a different acronym. AIO is the same again, and we do not use it because it already means several other things.

Do they make any difference? Not to what you do. Google's published guidance says appearing in AI Overviews and AI Mode needs no special markup, no new file on your server and no separate discipline. The pages that get used are the ones that are clear, specific, crawlable and trusted, which has been the definition of good SEO for years. What has changed is the outcome being measured. We have written a longer guide to where each term came from and which ones are sales language.

We use three terms ourselves. AI search optimisation is the work. AI recommendation visibility is the result. The AI visibility audit is how we measure it.

How ChatGPT looks for a business

ChatGPT does not search the way Google does. It rewrites your question into one or more background searches, decides where to look before it looks, reads a handful of sources and builds an answer from them. Across August 2026 we kept seeing four distinct patterns, and which one fired depended on how the question was phrased.

Ask a general question ("what should I look for in an arborist") and it ran a web search, read a few pages and framed the answer nationally. A business appeared only if its page was one of the pages read.

Ask who to call and the answer changed shape. A map with pins, business cards with star ratings and review counts, and far fewer web pages in the Sources panel. Website copy mattered much less here. Listing data did the work, and the star ratings on every card we checked matched the business's Google reviews exactly, to the review.

Name a business ("is this company any good") and the answer was built from what others had written about it: LinkedIn, news coverage, industry registers. The company's own pages came second or not at all.

Ask a technical or regulated question ("who is qualified to assess a protected tree") and the sources were association member lists and registries. Businesses on those lists appeared. Businesses describing the same capability in their own words on their own site did not.

A business can be invisible on one pattern and prominent on another, and most owners only ever test one. The full findings, with screenshots and the fan-out research behind them, are in How ChatGPT actually picks a business.

Google's AI Overviews are a different machine. They draw on Google's own index and ranking, so a business that ranks well on Google starts in a better position with Google's AI than with ChatGPT. Better position, not a place in the answer. We cover Google's mechanism separately.

What decides whether you get named

Being cited and being named are different. A page of yours can be quoted as a source in an answer about your industry without your business being recommended in it. Being named is what brings the phone call, and when we looked at who got named and who did not, it came down to five things. We use these five names in every audit we run.

1. Something to quote

In every category we tested, the recommended businesses were described in numbers, standards and named methods lifted straight from their own pages. "Level 5 arborist, 200 removals a year, works to the ANSI A300 pruning standard" gets repeated. "Auckland's most trusted tree care specialists" gets ignored, or the business gets filed under the wrong category because the machine had to guess. Most New Zealand websites are written warmly and vaguely, for people, and leave the machine nothing to quote. That is the finding behind the copywriting rule that is hiding you from AI.

2. Proof in more than one place

Reviews and mentions, and not only on Google. Google reviews first: the rating and review count on every ChatGPT business card we checked matched the business's Google profile exactly, and we do not know how the data gets there. On the pattern where someone asks about you by name, the answer was built from what other people had written about you: review platforms your industry uses, news coverage, a supplier's case study, a council register. A business with reviews in one place and nothing anywhere else is easy to rank and hard to recommend.

3. On the lists the machine reads

Industry registers, association member lists, established directories, LinkedIn. On two of the four patterns these were the whole answer. If your industry has a register and you are not on it, you do not exist for that question. LinkedIn deserves its own line. For a consultancy or a professional firm it was the first source ChatGPT reached for, ahead of the company's own website, and it quoted what the business had written about itself there. Your LinkedIn page is not a place to be listed. It is a place where what you say about yourself gets read back.

4. A profile that agrees with your website

Category, description, service area and hours, the same on your Google Business Profile as on your site. In our tests, agreement between listings and site was rewarded on every pattern that used listings, and the businesses ChatGPT confused with another company were the ones with a common name and listings that disagreed with each other.

5. A door the crawler can open

Text inside an image does not exist to a machine. Independent testing keeps finding that AI crawlers do not run scripts the way Google's does, and OpenAI does not document what its search crawler renders, so assume content that only appears after a script runs is not read. And a robots.txt file that blocks OAI-SearchBot switches ChatGPT search off for your site entirely. We read the robots.txt files of 100 New Zealand websites on 30 August 2026. Some had blocked the search crawler without meaning to. Others believed they had opted out of AI and had not.

In the audits we have run so far, Something to quote is where New Zealand businesses lose the most ground, and it is the one nobody expects.

In a nutshell

If the basics of local and website SEO are done well, you are most of the way there. Two things are new. The first is copywriting. Being named depends on whether your site contains sentences that say definitively what you do, for whom and where, in a form that still makes sense when a machine lifts one or two of them out of context. Most business copy was never written to survive that. The second is connection. It matters more than it used to that you are connected to sources: reviews on more than one platform, mentions in places the machine treats as reliable, a register if your industry has one, and a LinkedIn page that describes you the way you would want to be read back. The machine builds its picture of you from what everyone else says, and from what you say where it is listening.

Does your Google Business Profile matter to ChatGPT

Yes. The map-style answers are built from listing data, the ratings on the cards matched Google's in every case we checked, and the category and description on the profile were what the AI repeated back when it described a business. If your customers search locally, most of the work of being recommended by AI is the same work as being found in the map pack, which we cover on our local SEO page. If your customers search nationally, or your business takes some explaining, the profile still matters and your website and listings matter more.

Can you pay for it

Not inside the answer. No advertising product places a business in a ChatGPT recommendation or a Google AI Overview, and OpenAI states that ads do not change what ChatGPT says.

You can pay to appear beside the answer. ChatGPT ads have been live in New Zealand since April 2026, shown to logged-in adults on the Free and Go tiers, which is where most ChatGPT users sit. One format so far: a labelled sponsored card under the answer with a short title, a line of text, an image and a link. Targeting is by country plus a plain-language description of the conversations you want to appear in. Paid subscribers never see them. We are running an early campaign for a client and so far the results say more about what people are asking, and on which device, than about sales. Worth watching. Not a substitute for being named.

How long it takes

Nobody has measured this properly, including us, so treat any confident number with suspicion. What has been observed: in one published experiment, a three-page website for a deodorant that did not exist was recommended first by ChatGPT within thirty days, for the two narrow questions its pages answered, and never for the general ones. That is the fast end, and it worked because the pages were specific. Being named on the general questions is a prominence problem, built from reviews, listings and coverage over months, and there is no shortcut through it. The sources ChatGPT shows today are not the sources it will show in six months, which is why the date at the bottom of this page keeps changing.

How to test your own business

Most owners who tell us "we don't show up in AI" have run one question, in their own logged-in account, using their own business name. That test tells you almost nothing. Here is the one that does.

Open ChatGPT logged out, in a private window, so you are on the free default model that most people use. Then ask it a question the way a real buyer would, with a situation in it. Not "plumber Tauranga" but "my hot water cylinder is leaking in Tauranga, who should I call and what will it cost?" Read who it names and who it does not.

Now ask about your business by name, with your category and your town beside it: "is [your business], the electricians in Nelson, any good?" Never test the name on its own. If the machine has you confused with another company, this is the question that shows it.

Under each answer, open the Sources panel and screenshot it. That panel is the list of pages that shaped the answer. Every page on it is yours, a competitor's, or somewhere you could be, and each of those needs a different fix.

Finally, run the first question four ways: with the place, without it, naming a competitor, and as a technical question. You will see the pattern change, and you will see which patterns you are absent from. That list is the job.

The full twenty-minute version, with a second engine to check against and a look at your own robots.txt, is at the end of How ChatGPT actually picks a business.

How to see who is already citing you

You may already be quoted by an AI and not know it. Bing Webmaster Tools is free and its AI Performance report shows which of your pages Microsoft Copilot cites, how often and for which questions. It does not cover ChatGPT, but it is the only free first-party record of an AI quoting your pages. We set it up for a client this winter and found Copilot had been quoting their homepage a dozen times a day. Nobody knew. Setting it up takes ten minutes, and the steps are in How ChatGPT actually picks a business.

What we will not tell you

Nobody is an expert in this yet, including us. The platforms publish very little, the results move month to month, and most of what is written about the subject repeats one small study as if it were a rule. We read what the platforms publish, run our own tests, date everything and correct ourselves when the ground shifts. That method is the only authority anyone has in this field at the moment, and it rules some things out.

We will not sell you an llms.txt file. Google has said in writing that it ignores it.

We will not seed recommendations of your business on Reddit. ChatGPT reads Reddit constantly and cites it rarely, and Reddit is now removing around 25,000 promotional posts and comments a day.

We will not give you a single AI visibility score and call it a measurement. The platforms do not publish the data that would make one number honest, and the same question asked twice gives different answers. We report samples as samples: which prompts, how many runs, which engine, which model, what date, and the result as a range.

We will not promise you a position.

We will not call this a new discipline with its own acronym. It is search, done for a machine that reads more literally than a person does.

Go deeper

The experiments behind this page, and how to run them yourself.

Accurate as of 8 September 2026. This changes weekly: new reports, new experiments, new platform deals. We re-test the claims on this page every six weeks and update the date when we do.