“Best X in Melbourne”: reverse-engineering what ChatGPT actually cites

When someone asks ChatGPT for “the best dentist in Melbourne” or “the best Vietnamese in Footscray”, the answer is built from pages it can find and quote: roundup articles, directories, review sites, forum threads and a handful of business websites. You can see exactly which ones by opening the citation links. That list of sources is your to-do list. This post walks through a method you can run yourself in about an hour, with no tools beyond a browser and a spreadsheet.

Why the citations matter more than the answer

Most business owners who try this look at the names in the answer, feel good or bad, and close the tab. The names are the least useful part. They change from one run to the next, they change with wording, and they change when the person asking is logged in with a different history.

The citations are steadier. When ChatGPT searches the web for a local query, it shows little source links next to its claims. Those links tell you which pages it trusted enough to pull from. If the same three or four sites keep turning up across different wordings of your question, those are the pages deciding who gets recommended in your category. Getting onto them is usually faster than trying to make your own homepage the source.

The hour-long method

Open a fresh chat (or a logged-out browser window) with web search on. Then:

• Write eight to ten versions of the question your customers would type. Mix “best”, “top rated”, “recommend”, “affordable”, and a few suburbs, e.g. “best physio in Richmond”, “good physio near Box Hill for back pain”.
• Run each one. For every answer, click every citation and paste the URL into a spreadsheet with three columns: query, source URL, source type.
• Label the type: roundup article, directory, review platform, forum, news, competitor site, your site.
• Sort by domain and count how often each one appears.

Run the same set again a week later. Anything that shows up in both rounds is a real source for your category, not noise.

What you will usually find in the source list

We are not going to hand you percentages, because the mix depends heavily on the trade and the suburb, and anyone quoting a neat figure for “all Melbourne businesses” is guessing. What you can expect to see is a few recurring kinds of page:

Roundup articles. “12 best brunch spots in Melbourne” on a local media or lifestyle site. These are gold for AI answers because they are already written as a ranked list with a sentence or two about each business.

Directories and review platforms. Industry directories, booking platforms, and sites that show star ratings. Your listing there becomes the text the model paraphrases, so an empty or outdated profile hurts you twice.

Forum threads. Reddit threads where locals argue about the best option. You cannot buy your way into those, and you shouldn’t try, but it tells you what real customers say about you.

Business websites. Usually a minority, and usually the ones with clear, specific pages: services named in plain words, suburbs served, prices or price ranges, FAQs. If a competitor’s own site keeps getting cited and yours doesn’t, open their page and compare it with yours line by line.

Turning the spreadsheet into a plan

Go down the list of recurring sources and ask one question for each: can we realistically appear here, and how?

For roundup articles, check when they were last updated and who wrote them. Many local lists get refreshed each year, and editors do take pitches from businesses with something concrete to say (a new location, an award, a specialist service). For directories, claim or fix your profile so the category, suburb, hours and description match your website word for word. Inconsistent details give a model two versions of you to choose from.

For your own site, the job is making the page that answers the query easy to quote. That means one page per core service, a first paragraph that states what you do and where, and structured data so machines don’t have to guess. We covered the types worth adding in schema markup for small businesses. The same logic drives Google’s side of this, which we wrote up in how AI Overviews choose Melbourne businesses.

Reviews: the part you can’t shortcut

“Best” questions pull in reputation signals. When the answer describes a business as well reviewed, it is paraphrasing ratings and review text it found somewhere. Ask real customers for honest reviews on the platforms that showed up in your spreadsheet, reply to the ones you get, and leave it there. Fake or bought reviews break platform rules, and Australian consumer law treats misleading reviews seriously. They are also the kind of thing that ends up quoted back at you.

Common mistakes we see

Testing once, logged in, and treating the result as the truth. Personalisation and wording change the answer; the repeated sources are what count.

Chasing the answer instead of the sources. Rewriting your homepage ten times won’t help if the model is getting its list from a directory where you have no profile.

Ignoring suburbs. “Best X in Melbourne” and “best X in Glen Waverley” often pull different sources. If most of your customers come from two or three suburbs, test those by name.

If you would rather have this done for you, our AEO agency service in Melbourne starts with exactly this audit, and our ChatGPT SEO work picks up from there. Or get in touch and we’ll run the queries for your category with you.

Frequently asked questions