Which salons does ChatGPT recommend? A Sydney beauty index

Ask ChatGPT which salons it recommends in Sydney and you get a short list of names. Ask Google or Perplexity the same question and you get another list. Being named by one AI tool doesn’t mean you’re named by the others.

We asked four AI tools 50 beauty questions about Sydney and wrote down every business they named. Below you’ll find what we saw and a few checks you can run on your own salon.

We collected the answers on 20 and 21 October 2026 and call the result our Sydney beauty AI search index. It records what appeared in the answers we saved and passes no verdict on any business in them.

What happens when someone asks AI for a salon?

Here’s an example from our saved answers. We put this question to each tool: “Who can recommend a hair salon in Sydney’s Inner West for a basic trim without a wash and blow-dry?” ChatGPT named 3 salons. Google AI Mode named 5 and so did Google AI Overviews. Perplexity named 7.

Picture the person who asked. They probably read one answer and ring one of the names in it. We can’t see what people do next, so that part is our picture and not something we measured. It does show why a salon owner would want to know which tools name their salon and which don’t.

So we built a record you can check line by line: the question we asked, the names each tool gave and the links beside them. The four tools were:

• ChatGPT, OpenAI’s chat assistant.
• Google AI Mode, the chat-style way to search inside Google.
• Google AI Overviews, the AI summary Google sometimes puts above its normal results.
• Perplexity, a search tool built around AI answers.

This is our second Sydney AI search study. Read the first, on tradies, as our Sydney trades AI search study.

What did we find?

• ChatGPT gave 177 distinct recorded business names across its 49 full answers. No recorded business name appeared in more than 5 answers in any one tool.
• Few questions got the same name from all four tools. All four gave a full answer to 49 questions. Only 13 of those had a business name that every tool shared.
• Every full answer named at least one business (199 of 199), so every answer gave a reader businesses to look at.
• A name and a link are different things. Links to fresha.com, which our study counted as a booking or directory site, appeared in 36 of 50 Google AI Overviews answers and 22 of 49 ChatGPT answers.
• Google AI Mode shared a name with Google AI Overviews on 39 of 50 questions and with Perplexity on 39 of 50. ChatGPT and Google AI Mode shared one on only 27 of 49.

What should you do about it?

These checks come from what we saw in the answers. Try them with the services and suburbs you want bookings for.

1. Write questions that name your service and suburb, then save every answer. Ask each AI tool once and note the date. Our guide on how to check your own AI search visibility shows a simple way to keep the record.
2. Open the links beside your salon’s name. If a link goes to a booking page, a directory or your own website, check that the service, suburb and booking details on it are right. Our local SEO service covers the pages and profiles around your service area.
3. Read what the answer says about the service. If it describes a service you offer in words your page doesn’t use or links a booking page with old details, that points to the page or profile to review first.
4. If the answers keep skipping your salon, talk to us. Take the saved questions and answers to our AI SEO agency and we’ll work out what to fix first. Our SEO for beauty salons page covers the work we do for salons.

Which salons does ChatGPT recommend in Sydney?

ChatGPT gave 177 distinct recorded business names across its 49 full answers. By full answer we mean one the tool finished writing, which the data files call a completed answer. A business counted when the answer offered it as a provider of the beauty service in the question. We didn’t establish how many separate businesses sit behind those names, because a branch or a different spelling counts as its own name. The exact rules are in the fine print below.

Across all four tools we recorded 1,097 name appearances, each one a business named once in one answer. The next table gives each tool’s most-named business and the number of answers that named it. That number is a count and not a score for the business.

The Google AI Overviews row shows two spellings we recorded, which our matching counts as one name.

The second table totals the names each tool gave. “After matching” treats spellings that differ only in capitals or punctuation as one name. “As recorded” counts every spelling exactly as the tool wrote it.

The full business appearances file gives the question, the service, the area and the position of every name we saw. Open the saved answers before reading too much into one appearance.

Do the AI tools name the same salons?

Not often across all four tools. They all gave a full answer to 49 of our questions and only 13 of those had a business name that every tool shared. Pairs of tools overlapped more, as the table below shows.

For each pair of tools the next table counts the questions where both gave a full answer and at least one business appeared in both.

Google AI Mode shared a name with Google AI Overviews on 39 of 50 questions and with Perplexity on 39 of 50. ChatGPT and Google AI Mode were furthest apart at 27 of 49. That’s why it’s worth checking more than one tool before you decide how visible your salon is.

Which websites did the AI link to?

A tool can name your salon and link somewhere else, so we tracked names and links separately. A link might go to the salon’s own website, a booking page, a directory or a social profile. A directory is a site that lists many businesses in one place.

The next table counts the full answers that contained at least one link of each type, shown as answers with that link over full answers. One answer can carry several kinds of link, so the columns shouldn’t be added together.

Three things stand out. Google AI Mode linked to a Google business or map record in 47 of 50 answers while ChatGPT and Perplexity linked to none (0 of 49 and 0 of 50). Booking platforms appeared in 40 of 50 Google AI Overviews answers and 27 of 49 ChatGPT answers. Perplexity linked a directory in 26 of 50 answers while Google AI Mode did so in 4 of 50.

The websites below turned up most often in the links. Each cell counts the full answers that linked to that site at least once.

Fresha was the most-linked website in ChatGPT, Google AI Overviews and Perplexity. In Google AI Mode only google.com was linked more often. That’s a good reason to check the booking and directory links beside a mention as well as your own website.

Seeing a site in the links doesn’t mean we advise every salon to join it. OpenAI says ChatGPT can show cited sources and other relevant links in its Sources view,¹ so we kept ChatGPT and Perplexity source panels apart from the links shown in the answers. Our guide to how ChatGPT finds and cites sources explains more.

We also checked whether each named salon’s own website appeared in the same answer. That table sits in the fine print next to the method.

Which services and areas did we check?

We asked about 10 services grouped into 6 sectors, with 5 questions for each service. The services are haircuts, curly hair, colour and restyling, barbers, nails, brows, lashes, facials, waxing and makeup. Of the 50 questions, 13 named only Sydney. The rest named a suburb or area such as Newtown, Parramatta or Bondi.

In the next two tables the first number is name appearances and the second is full answers, so 55 / 14 for ChatGPT and hair salons means 55 names in 14 answers. ChatGPT has 14 hair salon answers rather than 15 because of its one failed check, which the fine print explains.

The same counts for each of the 10 services show where the sectors came from.

Curly hair questions brought back 21 names across 5 ChatGPT answers, 37 across 5 Google AI Mode answers, 29 across 5 Google AI Overviews answers and 41 across 5 Perplexity answers. In ChatGPT, Google AI Mode and Perplexity that was the most name appearances of any service. These counts show what our questions returned and not how many people want each service.

The calculated results file has the exact counts for every area.

How we ran the study

Before collecting anything we wrote 50 questions and locked the list. Every question names Sydney and none names a business. We drew them from 31 public Reddit discussions about salons and beauty services and reworded each one ourselves, so none is copied from a search history. The question set and source register keeps the wording, service, area and source for every question. We followed the method of our Sydney trades AI search study with beauty questions in place of trade questions. Where we changed how we match links, the fine print says so.

We asked each tool each question once, which makes 200 checks. Every question started a fresh conversation or search, with no follow-up questions and no regenerated answers. We never put a business name into a question.

“Blocked” would mean a tool stopped us with a login request or usage limit. “No Overview” would mean Google’s results page showed no AI Overview. “Failed” means our recording went wrong. None of these counts as a business missing from an answer.

We used our existing signed-in Chrome accounts. ChatGPT showed Pro and Perplexity showed Search on a free account. Google’s ordinary results showed a Sydney location taken from our internet address. We didn’t change any location permissions or look inside any account’s memory.

We saved the full text of every answer with the time we captured it, the business names in the order they appeared and the source links we could see. We counted an answer only after the tool finished writing it. Where a group of citations hid its destinations we had nothing to record, so those links stay unresolved.

The data bundle holds the records and the calculation script. Running python3 analyse.py rebuilds every table from the saved inputs.

The fine print

What we counted

• What counts as a business name. The answer had to present the business as a provider of the beauty service asked about. That includes a salon hosting independent practitioners and a provider the answer said might not suit the request. It leaves out adverts, landmarks, product brands and businesses that only appeared as source labels. A card with too little detail about the service stays in and is flagged as uncertain, as with the “Hana Cosmetics 日韩彩妆美瞳” card in one Perplexity answer.
• Names are not endorsements. We didn’t check any quality, price, review, service, availability or health claim a tool made. Where an answer joined a person and a salon into one label we kept one name. Marketplace options stay as the tool showed them.
• What the pair table compares. Each pair count includes only questions both tools answered in full. That is 49 questions when ChatGPT is in the pair and 50 otherwise, because of ChatGPT’s one failed check. The four-tool count uses 49. An answer that named no business would stay in these counts and none did.
• How the tables count a business. A business counts once per answer. Sector, service and area tables count appearances per full answer, so a cell often rests on a single question.

Names and spelling

To compare tools we ignored capital letters and punctuation and dropped the legal suffix “Pty Ltd”. We didn’t treat “&” as “and”, remove suburb names, merge branches or match names that only look alike. The comparison also drops accented and non-Latin letters while the download keeps every exact spelling.

So in the trim question above BOB, BOB (Newtown) and Mei at Bob on King St stay separate between answers. A shared name is a literal match after those steps. No match doesn’t prove two names belong to different businesses, which means our shared-name counts can understate how many businesses were shared.

Matching names to websites

We counted an own-website match only when the saved evidence tied a visible link to the named business. That meant a reviewed association or a link label containing the business’s full name. We never guessed from how similar a web address looked. Google saved no label on most of its links, so its figures are very low and say little about whether websites were linked.

These counts are minimums. A link we couldn’t match is unresolved and doesn’t show a website was absent. Don’t rank the tools by these figures. Don’t compare them with the trades study either, which used a fuller manual check. The citation review records the evidence for each match.

Google’s Website buttons showed no destination either. We counted Google’s expanded source cards. A match also doesn’t show who legally owns a domain or that a link supports every sentence beside it.

What the numbers can’t tell you

This is a dated sample. We collected the answers on 20 and 21 October 2026 from 50 questions adapted from 31 public discussions. They cover the services and places we chose, not every Sydney salon or how often people search. Several questions share a discussion, so they aren’t independent. Our sectors and areas have different sizes, so they can’t show market share or that one area does better than another.

The account and the moment matter. We didn’t reset our accounts between questions. Starting a new chat removes the earlier messages but doesn’t prove that account memory or other personalisation played no part. Some ChatGPT captures showed a Memory indicator. A tool may also have changed during the two days of collection. A tool’s menu label doesn’t tell us which model version answered. Another account or date may give different answers.

One check failed. ChatGPT’s check for Q011 (“Which hairdressers in Newtown, Sydney are good for a dramatic change from long blonde hair?”) never showed a sent message or an answer. After we submitted it the screen showed an empty new chat. We kept the record and didn’t retry it. It sits outside every answer and agreement count. A failed check isn’t an empty answer. No other check was missing or blocked. Google showed an AI Overview for all 50 questions.

Some locations conflict. Two Google AI Overviews answers showed a card for Alysha Maree Makeup with a Bundall QLD address beside Sydney text. Perplexity listed The Skin Bar Rozelle in a southern suburbs answer and linked a Victorian Fairfield page in a Sydney makeup answer. Another Google card listed THE EYEBROW ATELIER as a hotel. We kept all of these and flagged them. We haven’t checked every location and make no claim about those businesses.

The study doesn’t test a fix. We didn’t change any salon website, listing or review to see what happened. The checks earlier on this page are a practical suggestion drawn from the records. They aren’t evidence of what gets a salon named or of extra bookings. A salon missing from our sample isn’t shown to be invisible in general.

Using the data

Researchers and journalists can download the files below. The public download leaves out account identifiers, private conversation links and original browser files. It reproduces our arithmetic from the saved text and coding. It can’t recreate a live answer.

Raw data for researchers (JSON)

• Question set with the 50 questions and their sources
• Saved answers
• Reviewed answers with our coding decisions
• Business appearances in the original CSV format
• Calculated results
• Calculation script
• Data bundle with every raw file above

The question set’s SHA-256 fingerprint is 41f3a3c4cf30908cacff7bddfe008fa80d40a05d2cc833759d1e3a189eddbe67, so you can confirm the file is unchanged. If you quote a figure, give the collection date, the tool and how many answers it rests on. Keep the question attached to any single example. A suggested citation is: Kavera, Sydney beauty AI search index, captured 20 and 21 October 2026; 50 questions, 200 recorded checks and 199 completed answers.

Want help checking your own salon?

Bring a question your clients ask and the answer you saw. We’ll use that example to plan your next SEO step.

Our SEO audit is free. Implementation is separate unless retained. See the SEO packages for terms. Monthly plans start at Suburb SEO, $890/month + GST. No SEO package guarantees rankings, leads or revenue.

Talk about your SEO