Who the Machines Name

We ran eight buying-intent prompts across three AI answer engines from a Sydney connection on 1 September 2026. Two patterns held. Local questions were answered almost entirely from Google Business Profiles rather than from websites, and business directories were cited as sources in their own right.

Plenty has been written about optimising for AI search. Very little of it says what an AI engine actually does when someone in Sydney asks it to recommend a local business. So we measured ours, wrote down the method, and are publishing the parts that are useful to anyone else running a small Australian business.

What exactly did we test?

Eight prompts, phrased the way a buyer would type them rather than the way a marketer would: three about AI receptionists, three about finding a marketing or web agency in a named Sydney area, one about who builds AI phone agents in Sydney, and one control prompt that was simply our own business name.

Each prompt went to three answer surfaces that work without an account, all run logged out from an Australian connection with Australian region and language settings. Every answer was screenshotted in full and recorded in a machine-readable file, 26 records in total.

The two engines that require a signed-in account were deliberately not attempted. That is a real gap in the run and we would rather say so than imply we covered everything.

Which questions even get an AI answer?

Not all of them, and the pattern is useful. An AI Overview appeared on five of the eight prompts. The three that did not get one were the local ones, the questions of the form "marketing agency in this Sydney area". Those returned a map and a pack of local business listings instead.

Google's AI Mode and Perplexity both produced an answer for all eight, including the local ones. So whether a question gets an AI answer depends heavily on which surface the person is using, and local intent is treated differently from research intent almost everywhere.

The practical read for a local business: if your customers are asking a "who should I hire near me" question, the thing you are competing to appear in is often not a written AI answer at all. It is a listing. And the listing is fed by something other than your website.

Where do AI answers get local recommendations from?

This was the finding that changed our own plan. On the local prompts in Google's AI Mode, we expanded every source the answer cited. Every single one was a Google Business Profile.

Not a website. Not a blog post. Not a directory page. Each cited source showed a business name, a star rating, a review count, a suburb, a phone number, and buttons to call, get directions or visit the site. There were zero web domains cited on those answers.

If that holds, it means on-page work cannot get a business into that answer. No amount of writing, structured data or keyword placement puts you in a list that is being assembled from Business Profile records. The only entry point is the profile itself: complete, correctly categorised, with the same name and phone number everywhere, and with reviews on it.

So does the website still matter?

Yes, and the same run is the evidence. The non-local prompts, the ones about AI receptionists and about what a service costs, cited web pages heavily. Those are research questions, and research questions get answered from written content.

Two of the sources cited on one of those prompts were business directory listings, not the businesses' own sites. That is worth sitting with. A directory entry is not only a link for search engines any more; it is a document an AI engine will read and quote from when it decides who to name. For a small business that has never bothered with directories, that is a cheap gap to close.

The split is roughly this. Local intent is answered from profiles. Research intent is answered from pages and directories. Most businesses need both, and most businesses have only tried at one of them.

What happens when someone searches your business name?

Ours was cited on two of the three surfaces, so the mechanism works. But every source in those answers was either our own website or one of our own social profiles. There was not one independent page about us anywhere in the results.

That is the honest state of a young business online, and it is a specific kind of weakness rather than a vague one. An answer engine assembling a recommendation is looking for a business that other places mention. If everything it can find about you was written by you, you are a website rather than an entity, and you tend to lose to businesses that are talked about elsewhere.

There is a second-order problem worth checking on your own name. Ours collides with several unrelated businesses using a similar trading name overseas, and one answer engine asked which one was meant. If your name is not distinctive, expect to be confused with somebody else and to have to work harder to be identified.

What should a Sydney business fix first?

In this order, because it follows the evidence rather than the fashion.

First, the Google Business Profile: complete every field, choose the categories precisely, list the service areas honestly, and keep the business name identical to the one used everywhere else. Second, reviews, because the cited profiles carried ratings and review counts as part of the answer. Third, directory listings with an identical name, address and phone, because directories are cited sources. Fourth, and only fourth, on-page content written to answer real questions, which is what wins the research prompts.

Notice what is not on that list: no AI-specific file, no special markup, no rephrasing your site to sound like a chatbot. Google's own guidance says none of those are required, and nothing in this run contradicted it.

How can you run this on your own business?

You need about half an hour and no tools. Write down the eight questions a customer would actually ask before hiring you, plus your own business name as a control. Run each through the AI answer on a normal search page, through the AI mode, and through one independent answer engine, logged out, so you get the public answer rather than a personalised one. Screenshot every answer.

Then record three things per answer: did an AI answer appear at all, which sources did it cite, and which businesses did it name. Check the control first. If your own name does not return you, stop and fix the method before you believe any other result.

Repeat it monthly and keep the files. A single run tells you where you stand; a series tells you whether anything you did worked.

What this run does not prove

It is one set of prompts, from one city, on one day, on three surfaces, logged out. Answer engines vary by person, by location and by the hour, and two major assistants were not tested because they require an account.

We are publishing it anyway, with the method attached, because a measured result with stated limits beats another article confidently explaining how AI search works. Run it on yourself and you will know more about your own visibility than any general advice can tell you.

Sources: K&G AI answer visibility baseline, 1 September 2026, 8 prompts across 3 logged-out answer surfaces from an Australian connection, 26 recorded results with full-page screenshots. Google Search Central, AI optimisation guidance.

Related reading: what ranking one suburb page taught us about local SEO and how to rank higher on Google Maps.

Want to know what the answer engines say about your business? Ask us to run the same eight-prompt check on you and we will send you the screenshots either way.